Compare commits

..

16 Commits

Author SHA1 Message Date
James Long f1030db291 chore: typecheck only affected packages in pre-push hook 2026-07-03 13:40:14 +00:00
Aarav Sareen a4fed69a82 feat(app): dropdown search fix (#34961)
Co-authored-by: Brendan Allan <git@brendonovich.dev>
2026-07-03 10:31:37 +00:00
opencode-agent[bot] 41a3cfcdd9 chore: update nix node_modules hashes 2026-07-03 09:37:53 +00:00
opencode-agent[bot] 72845e09fc chore: generate 2026-07-03 09:25:41 +00:00
Aiden Cline abaab29cb3 feat(opencode): add code-mode MCP adapter (#35085) 2026-07-03 09:24:27 +00:00
opencode-agent[bot] 30936a9bca chore: generate 2026-07-03 08:11:01 +00:00
Aarav Sareen a9144eccf8 feat(app): improvements to model search (#34954) 2026-07-03 08:09:51 +00:00
Aiden Cline 458ec7b372 refactor(opencode): expose MCP tools in native shape from the service (#35103) 2026-07-03 03:05:36 -05:00
usrnk1 d46c02ba73 feat(desktop): papercut fixes (#34939) 2026-07-03 08:04:59 +00:00
opencode-agent[bot] caedf36844 chore: generate 2026-07-03 07:54:40 +00:00
Aarav Sareen 4c6e2a92ce feat(app): align subagent UI with v2 (#34931)
Co-authored-by: Luke Parker <10430890+Hona@users.noreply.github.com>
2026-07-03 15:53:31 +08:00
usrnk1 bf58fae51f feat(desktop): add recently closed projects to home (#34926) 2026-07-03 15:52:40 +08:00
Aiden Cline eb3476660f chore: remove root node-gyp pin after Node 24 setup (#35092)
deploy / deploy (push) Has been cancelled
nix-eval / nix-eval (push) Has been cancelled
nix-hashes / compute-hash (blacksmith-4vcpu-ubuntu-2404, x86_64-linux) (push) Has been cancelled
nix-hashes / compute-hash (blacksmith-4vcpu-ubuntu-2404-arm, aarch64-linux) (push) Has been cancelled
nix-hashes / compute-hash (macos-15-intel, x86_64-darwin) (push) Has been cancelled
nix-hashes / compute-hash (macos-latest, aarch64-darwin) (push) Has been cancelled
nix-hashes / update-hashes (push) Has been cancelled
publish / version (push) Has been cancelled
publish / build-cli (push) Has been cancelled
publish / sign-cli-windows (push) Has been cancelled
publish / build-electron (map[bun_install_flags:--os=darwin --cpu=arm64 host:macos-26 platform_flag:--mac --arm64 target:aarch64-apple-darwin]) (push) Has been cancelled
publish / build-electron (map[bun_install_flags:--os=darwin --cpu=x64 host:macos-26-intel platform_flag:--mac --x64 target:x86_64-apple-darwin]) (push) Has been cancelled
publish / build-electron (map[host:blacksmith-4vcpu-ubuntu-2404 platform_flag:--linux target:x86_64-unknown-linux-gnu]) (push) Has been cancelled
publish / build-electron (map[host:blacksmith-4vcpu-ubuntu-2404-arm platform_flag:--linux --arm64 target:aarch64-unknown-linux-gnu]) (push) Has been cancelled
publish / build-electron (map[host:blacksmith-4vcpu-windows-2025 platform_flag:--win target:x86_64-pc-windows-msvc]) (push) Has been cancelled
publish / build-electron (map[host:windows-2025 platform_flag:--win --arm64 target:aarch64-pc-windows-msvc]) (push) Has been cancelled
publish / publish (push) Has been cancelled
storybook / storybook build (push) Has been cancelled
generate / generate (push) Has been cancelled
typecheck / typecheck (push) Has been cancelled
2026-07-03 00:45:17 -05:00
opencode-agent[bot] 7d48a22a63 chore: update nix node_modules hashes 2026-07-03 05:33:25 +00:00
opencode-agent[bot] 3f0fc22b6e chore: generate 2026-07-03 05:20:28 +00:00
Aiden Cline 2409c7a3d5 feat(codemode): add confined execution package (#35079) 2026-07-03 05:19:11 +00:00
56 changed files with 3141 additions and 2268 deletions
+3 -1
View File
@@ -17,4 +17,6 @@ if (process.versions.bun !== expectedBunVersion) {
console.warn(`Warning: Bun version ${process.versions.bun} differs from expected ${expectedBunVersion}`);
}
'
bun typecheck
# Only typecheck packages affected since the merge base with origin/dev,
# plus their dependents. Override the base with TURBO_SCM_BASE if needed.
TURBO_SCM_BASE="${TURBO_SCM_BASE:-origin/dev}" bun turbo typecheck --affected
+15 -26
View File
@@ -19,7 +19,6 @@
"@typescript/native-preview": "catalog:",
"glob": "13.0.5",
"husky": "9.1.7",
"node-gyp": "12.4.0",
"oxlint": "1.60.0",
"oxlint-tsgolint": "0.21.0",
"prettier": "3.6.2",
@@ -3600,7 +3599,7 @@
"entities": ["entities@7.0.1", "", {}, "sha512-TWrgLOFUQTH994YUyl1yT4uyavY5nNB5muff+RtWaqNVCAK408b5ZnnbNAUEWLTCpum9w6arT70i1XdQ4UeOPA=="],
"env-paths": ["env-paths@2.2.1", "", {}, "sha512-+h1lkLKhZMTYjog1VEpJNG7NZJWcuc2DDk/qsqSTRRCOXiLjeQ1d1/udrUGhqMxUgAlwKNZ0cf2uqan5GLuS2A=="],
"env-paths": ["env-paths@3.0.0", "", {}, "sha512-dtJUTepzMW3Lm/NPxRf3wP4642UWhjL2sQxc+ym2YMj1m/H2zDNQOlezafzkHwn6sMstjHTwG6iQQsctDW/b1A=="],
"err-code": ["err-code@2.0.3", "", {}, "sha512-2bmlRpNKBxT/CRmPOlyISQpNj+qSeYvcym/uT0Jx2bMOlKLtSy1ZmLuVxSEKKyor/N5yhvp/ZiG1oE3DEYMSFA=="],
@@ -4522,7 +4521,7 @@
"node-fetch-native": ["node-fetch-native@1.6.7", "", {}, "sha512-g9yhqoedzIUm0nTnTqAQvueMPVOuIY16bqgAJJC8XOOubYFNwz6IER9qs0Gq2Xd0+CecCKFjtdDTMA4u4xG06Q=="],
"node-gyp": ["node-gyp@12.4.0", "", { "dependencies": { "env-paths": "^2.2.0", "exponential-backoff": "^3.1.1", "graceful-fs": "^4.2.6", "nopt": "^9.0.0", "proc-log": "^6.0.0", "semver": "^7.3.5", "tar": "^7.5.4", "tinyglobby": "^0.2.12", "undici": "^6.25.0", "which": "^6.0.0" }, "bin": { "node-gyp": "bin/node-gyp.js" } }, "sha512-OMcPNvqTCFUnNaBlmdgq+lfNqY7gTiSmNRDjY3uAXRyudeKZEZxu3CLtjMQrx4zZxCX2b/mpNqTtwuCJgXhHkw=="],
"node-gyp": ["node-gyp@12.3.0", "", { "dependencies": { "env-paths": "^2.2.0", "exponential-backoff": "^3.1.1", "graceful-fs": "^4.2.6", "nopt": "^9.0.0", "proc-log": "^6.0.0", "semver": "^7.3.5", "tar": "^7.5.4", "tinyglobby": "^0.2.12", "undici": "^6.25.0", "which": "^6.0.0" }, "bin": { "node-gyp": "bin/node-gyp.js" } }, "sha512-QNcUWM+HgJplcPzBvFBZ9VXacyGZ4+VTOb80PwWR+TlVzoHbRKULNEzpRsnaoxG3Wzr7Qh7BYxGDU3CbKib2Yg=="],
"node-gyp-build": ["node-gyp-build@4.8.4", "", { "bin": { "node-gyp-build": "bin.js", "node-gyp-build-optional": "optional.js", "node-gyp-build-test": "build-test.js" } }, "sha512-LA4ZjwlnUblHVgq0oBF3Jl/6h/Nvs5fzBLwdEF4nuxnFdsfajde4WfxtJr3CaiH+F6ewcIB/q4jQ4UzPyid+CQ=="],
@@ -5372,7 +5371,7 @@
"uncrypto": ["uncrypto@0.1.3", "", {}, "sha512-Ql87qFHB3s/De2ClA9e0gsnS6zXG27SkTiSJwjCc9MebbfapQfuPzumMIUMi38ezPZVNFcHI9sUIepeQfw8J8Q=="],
"undici": ["undici@6.26.0", "", {}, "sha512-4yqz8a3n5HmGTlsbADNtr/dJlhkh/55Rq798G6ibiULcXbDtaLpTl1pvdqcbFfeoj3iSi52lePFM7h9H21cw/A=="],
"undici": ["undici@8.3.0", "", {}, "sha512-TkUDgb6tl7KOGZ+7e8E3d2FYgUQgF6z5YypqjWmixVQSQERFcVrVg0ySADm2LVLRh5ljAaHTCR5Fmz3Q34rB7Q=="],
"undici-types": ["undici-types@7.16.0", "", {}, "sha512-Zz+aZWSj8LE6zoxD+xrjh4VfkIG8Ya6LvYkZqtUQGJPZjYl53ypCaUwWqo7eI0x66KBGeRo+mlBEkMSeSZ38Nw=="],
@@ -5574,7 +5573,7 @@
"y18n": ["y18n@5.0.8", "", {}, "sha512-0pfFzegeDWJHJIAmTLRP2DwHjdF5s7jo9tuztdQxAhINCdvS+3nGINqPd00AphqJR/0LhANUS6/+7SCb98YOfA=="],
"yallist": ["yallist@5.0.0", "", {}, "sha512-YgvUTfwqyc7UXVMrB+SImsVYSmTS8X/tSrtdNZMImM+n7+QTriRXyXim0mBrTXNeqzVF0KWGgHPeiyViFFrNDw=="],
"yallist": ["yallist@4.0.0", "", {}, "sha512-3wdGidZyq5PB084XLES5TpOSRA3wjXAlIWMhum2kRcv/41Sn2emQ0dycQW4uZXLejwKvg6EsvbdlVL+FYEct7A=="],
"yaml": ["yaml@2.9.0", "", { "bin": { "yaml": "bin.mjs" } }, "sha512-2AvhNX3mb8zd6Zy7INTtSpl1F15HW6Wnqj0srWlkKLcpYl/gMIMJiyuGq2KeI2YFxUPjdlB+3Lc10seMLtL4cA=="],
@@ -5622,6 +5621,8 @@
"@actions/github/undici": ["undici@5.29.0", "", { "dependencies": { "@fastify/busboy": "^2.0.0" } }, "sha512-raqeBD6NQK4SkWhQzeYKd1KmIG6dllBOTt55Rmkt4HtI9mwdWtJljnrXjAFUBLTSN67HWrOIZ3EPF4kjUw80Bg=="],
"@actions/http-client/undici": ["undici@6.26.0", "", {}, "sha512-4yqz8a3n5HmGTlsbADNtr/dJlhkh/55Rq798G6ibiULcXbDtaLpTl1pvdqcbFfeoj3iSi52lePFM7h9H21cw/A=="],
"@ai-sdk/alibaba/@ai-sdk/openai-compatible": ["@ai-sdk/openai-compatible@2.0.41", "", { "dependencies": { "@ai-sdk/provider": "3.0.8", "@ai-sdk/provider-utils": "4.0.23" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-kNAGINk71AlOXx10Dq/PXw4t/9XjdK8uxfpVElRwtSFMdeSiLVt58p9TPx4/FJD+hxZuVhvxYj9r42osxWq79g=="],
"@ai-sdk/amazon-bedrock/@ai-sdk/anthropic": ["@ai-sdk/anthropic@3.0.81", "", { "dependencies": { "@ai-sdk/provider": "3.0.10", "@ai-sdk/provider-utils": "4.0.27" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-B1JDd9Ugq9R5AgIaW3674lhGCMMYJcPUxnrZh8fzbGojgg4QvHFRv6eZahGQAUsmGHbcf74G9bdSBDLWQGY2GA=="],
@@ -5874,8 +5875,6 @@
"@dot/log/chalk": ["chalk@4.1.2", "", { "dependencies": { "ansi-styles": "^4.1.0", "supports-color": "^7.1.0" } }, "sha512-oKnbhFyRIXpUuez8iBMmyEa4nbj4IOQyuhc/wy9kY7/WVPcwIO9VA668Pu8RkO7+0G76SLROeyw9CpQ061i4mA=="],
"@effect/platform-node/undici": ["undici@8.3.0", "", {}, "sha512-TkUDgb6tl7KOGZ+7e8E3d2FYgUQgF6z5YypqjWmixVQSQERFcVrVg0ySADm2LVLRh5ljAaHTCR5Fmz3Q34rB7Q=="],
"@electron/asar/commander": ["commander@5.1.0", "", {}, "sha512-P0CysNDQ7rtVw4QIQtm+MRxV66vKFSvlsQvGYXZWR3qFU0jlMKHZZZgw8e+8DSah4UDKMqnknRDQz+xuQXQ/Zg=="],
"@electron/asar/glob": ["glob@7.2.3", "", { "dependencies": { "fs.realpath": "^1.0.0", "inflight": "^1.0.4", "inherits": "2", "minimatch": "^3.1.1", "once": "^1.3.0", "path-is-absolute": "^1.0.0" } }, "sha512-nFR0zLpU2YCaRxwoCJvL6UvCH2JFyFVIvwTLsIf21AuHlMskA1hhTdk+LlYJtOlYt9v6dvszD2BGRqBL+iQK9Q=="],
@@ -5886,16 +5885,12 @@
"@electron/fuses/fs-extra": ["fs-extra@9.1.0", "", { "dependencies": { "at-least-node": "^1.0.0", "graceful-fs": "^4.2.0", "jsonfile": "^6.0.1", "universalify": "^2.0.0" } }, "sha512-hcg3ZmepS30/7BSFqRvoo3DOMQu7IjqxO5nCDt+zM9XWjb33Wg7ziNT+Qvqbuc3+gWpzO02JubVyk2G4Zvo1OQ=="],
"@electron/get/env-paths": ["env-paths@3.0.0", "", {}, "sha512-dtJUTepzMW3Lm/NPxRf3wP4642UWhjL2sQxc+ym2YMj1m/H2zDNQOlezafzkHwn6sMstjHTwG6iQQsctDW/b1A=="],
"@electron/get/undici": ["undici@7.26.0", "", {}, "sha512-3O9Tf67pGhgOv9jM35AbhkXAKi13f3oy3aE4CSgr+TckGeY+/iu97ZXN+J7DpHPzLbVApFd1IFhcnBjREYXYcg=="],
"@electron/notarize/fs-extra": ["fs-extra@9.1.0", "", { "dependencies": { "at-least-node": "^1.0.0", "graceful-fs": "^4.2.0", "jsonfile": "^6.0.1", "universalify": "^2.0.0" } }, "sha512-hcg3ZmepS30/7BSFqRvoo3DOMQu7IjqxO5nCDt+zM9XWjb33Wg7ziNT+Qvqbuc3+gWpzO02JubVyk2G4Zvo1OQ=="],
"@electron/osx-sign/isbinaryfile": ["isbinaryfile@4.0.10", "", {}, "sha512-iHrqe5shvBUcFbmZq9zOQHBoeOhZJu6RQGrDpBgenUm/Am+F3JM2MgQj+rK3Z601fzrL5gLZWtAPH2OBaSVcyw=="],
"@electron/rebuild/node-gyp": ["node-gyp@12.3.0", "", { "dependencies": { "env-paths": "^2.2.0", "exponential-backoff": "^3.1.1", "graceful-fs": "^4.2.6", "nopt": "^9.0.0", "proc-log": "^6.0.0", "semver": "^7.3.5", "tar": "^7.5.4", "tinyglobby": "^0.2.12", "undici": "^6.25.0", "which": "^6.0.0" }, "bin": { "node-gyp": "bin/node-gyp.js" } }, "sha512-QNcUWM+HgJplcPzBvFBZ9VXacyGZ4+VTOb80PwWR+TlVzoHbRKULNEzpRsnaoxG3Wzr7Qh7BYxGDU3CbKib2Yg=="],
"@electron/universal/fs-extra": ["fs-extra@11.3.5", "", { "dependencies": { "graceful-fs": "^4.2.0", "jsonfile": "^6.0.1", "universalify": "^2.0.0" } }, "sha512-eKpRKAovdpZtR1WopLHxlBWvAgPny3c4gX1G5Jhwmmw4XJj0ifSD5qB5TOo8hmA0wlRKDAOAhEE1yVPgs6Fgcg=="],
"@electron/universal/minimatch": ["minimatch@9.0.9", "", { "dependencies": { "brace-expansion": "^2.0.2" } }, "sha512-OBwBN9AL4dqmETlpS2zasx+vTeWclWzkblfZk7KTA5j3jeOONz/tRCnZomUyvNg83wL5Zv9Ss6HMJXAgL8R2Yg=="],
@@ -5946,8 +5941,6 @@
"@npmcli/query/postcss-selector-parser": ["postcss-selector-parser@7.1.1", "", { "dependencies": { "cssesc": "^3.0.0", "util-deprecate": "^1.0.2" } }, "sha512-orRsuYpJVw8LdAwqqLykBj9ecS5/cRHlI5+nvTo8LcCKmzDmqVORXtOIYEEQuL9D4BxtA1lm5isAqzQZCoQ6Eg=="],
"@npmcli/run-script/node-gyp": ["node-gyp@12.3.0", "", { "dependencies": { "env-paths": "^2.2.0", "exponential-backoff": "^3.1.1", "graceful-fs": "^4.2.6", "nopt": "^9.0.0", "proc-log": "^6.0.0", "semver": "^7.3.5", "tar": "^7.5.4", "tinyglobby": "^0.2.12", "undici": "^6.25.0", "which": "^6.0.0" }, "bin": { "node-gyp": "bin/node-gyp.js" } }, "sha512-QNcUWM+HgJplcPzBvFBZ9VXacyGZ4+VTOb80PwWR+TlVzoHbRKULNEzpRsnaoxG3Wzr7Qh7BYxGDU3CbKib2Yg=="],
"@octokit/auth-app/@octokit/request": ["@octokit/request@10.0.10", "", { "dependencies": { "@octokit/endpoint": "^11.0.3", "@octokit/request-error": "^7.0.2", "@octokit/types": "^16.0.0", "content-type": "^2.0.0", "json-with-bigint": "^3.5.3", "universal-user-agent": "^7.0.2" } }, "sha512-KxNC2pTqqhszMNrf12ZRd4PonRgyJdsM4F/jySiddQK+DsRcfBtUvqn8t7UsyZhnRJHvX46OohDt5N3VqIWC2w=="],
"@octokit/auth-app/@octokit/request-error": ["@octokit/request-error@7.1.0", "", { "dependencies": { "@octokit/types": "^16.0.0" } }, "sha512-KMQIfq5sOPpkQYajXHwnhjCC0slzCNScLHs9JafXc4RAJI+9f+jNDlBNaIMTvazOPLgb4BnlhGJOTbnN0wIjPw=="],
@@ -6222,8 +6215,6 @@
"conf/dot-prop": ["dot-prop@10.1.0", "", { "dependencies": { "type-fest": "^5.0.0" } }, "sha512-MVUtAugQMOff5RnBy2d9N31iG0lNwg1qAoAOn7pOK5wf94WIaE3My2p3uwTQuvS2AcqchkcR3bHByjaM0mmi7Q=="],
"conf/env-paths": ["env-paths@3.0.0", "", {}, "sha512-dtJUTepzMW3Lm/NPxRf3wP4642UWhjL2sQxc+ym2YMj1m/H2zDNQOlezafzkHwn6sMstjHTwG6iQQsctDW/b1A=="],
"config-chain/ini": ["ini@1.3.8", "", {}, "sha512-JV/yugV2uzW5iMRSiZAyDtQd+nxtUnjeLt0acNdw98kKLrvuRVyB80tsREOE7yvGVgalhZ6RNXCmEHkUKBKxew=="],
"crc/buffer": ["buffer@5.7.1", "", { "dependencies": { "base64-js": "^1.3.1", "ieee754": "^1.1.13" } }, "sha512-EHcyIPBQ4BSGlvjB16k5KgAJ27CIsHY/2JBmCRReo48y9rQ3MaUzWX3KVlBa4U7MyX02HdVj0K7C3WaB3ju7FQ=="],
@@ -6362,7 +6353,9 @@
"nitro/undici": ["undici@7.26.0", "", {}, "sha512-3O9Tf67pGhgOv9jM35AbhkXAKi13f3oy3aE4CSgr+TckGeY+/iu97ZXN+J7DpHPzLbVApFd1IFhcnBjREYXYcg=="],
"node-gyp/semver": ["semver@7.8.5", "", { "bin": { "semver": "bin/semver.js" } }, "sha512-Y7/KDsb8LjooZpwaqGyulO6DQlksgCncchHGk+sZIY4SBvUocMBEFH5Ur1fI4dV+Jvl0w6cjvucaIi40puRioA=="],
"node-gyp/env-paths": ["env-paths@2.2.1", "", {}, "sha512-+h1lkLKhZMTYjog1VEpJNG7NZJWcuc2DDk/qsqSTRRCOXiLjeQ1d1/udrUGhqMxUgAlwKNZ0cf2uqan5GLuS2A=="],
"node-gyp/undici": ["undici@6.26.0", "", {}, "sha512-4yqz8a3n5HmGTlsbADNtr/dJlhkh/55Rq798G6ibiULcXbDtaLpTl1pvdqcbFfeoj3iSi52lePFM7h9H21cw/A=="],
"node-gyp-build-optional-packages/detect-libc": ["detect-libc@2.1.2", "", {}, "sha512-Btj2BOOO83o3WyH59e8MgXsxEQVcarkUOpEYrubB0urwnN10yQ364rsiByU11nZlqWYZm05i/of7io4mzihBtQ=="],
@@ -6466,6 +6459,8 @@
"sucrase/commander": ["commander@4.1.1", "", {}, "sha512-NOKm8xhkzAjzFx8B2v5OAHT+u5pRQc2UCa2Vq9jYL/31o2wi9mxBA7LIFs3sV5VSC49z6pEhfbMULvShKj26WA=="],
"tar/yallist": ["yallist@5.0.0", "", {}, "sha512-YgvUTfwqyc7UXVMrB+SImsVYSmTS8X/tSrtdNZMImM+n7+QTriRXyXim0mBrTXNeqzVF0KWGgHPeiyViFFrNDw=="],
"terser/acorn": ["acorn@8.16.0", "", { "bin": { "acorn": "bin/acorn" } }, "sha512-UVJyE9MttOsBQIDKw1skb9nAwQuR5wuGD3+82K6JgJlm/Y+KI92oNsMNGZCYdDsVtRHSak0pcV5Dno5+4jh9sw=="],
"terser/commander": ["commander@2.20.3", "", {}, "sha512-GpVkmM8vF2vQUkj2LvZmD35JxeJOLCwJ9cUkugyk2nuhbv3+mJvpLYYt+0+USMxE+oj+ey/lJEnhZw75x/OMcQ=="],
@@ -6952,6 +6947,8 @@
"ansi-align/string-width/strip-ansi": ["strip-ansi@6.0.1", "", { "dependencies": { "ansi-regex": "^5.0.1" } }, "sha512-Y38VPSHcqkFrCpFnQ9vuSXmquuv5oXOKpGeT6aGrr3o3Gc9AlVa6JBfUSOCnbxGGZF+/0ooI7KrPuUSztUdU5A=="],
"app-builder-lib/@electron/get/env-paths": ["env-paths@2.2.1", "", {}, "sha512-+h1lkLKhZMTYjog1VEpJNG7NZJWcuc2DDk/qsqSTRRCOXiLjeQ1d1/udrUGhqMxUgAlwKNZ0cf2uqan5GLuS2A=="],
"app-builder-lib/@electron/get/fs-extra": ["fs-extra@8.1.0", "", { "dependencies": { "graceful-fs": "^4.2.0", "jsonfile": "^4.0.0", "universalify": "^0.1.0" } }, "sha512-yhlQgA6mnOJUKOsRUFsgJdQCvkKhcz8tlZG5HBQfReYZy46OwLcY+Zia0mtdHsOo9y/hP+CxMN0TU9QxoOtG4g=="],
"app-builder-lib/@electron/get/semver": ["semver@6.3.1", "", { "bin": { "semver": "bin/semver.js" } }, "sha512-BR7VvDCVHO+q2xBEWskxS6DJE1qRnb7DxzUrogb71CWoSficBxYsiAGd+Kl0mmq/MprG9yArRkyrQxTO6XjMzA=="],
@@ -7082,10 +7079,6 @@
"lazystream/readable-stream/string_decoder": ["string_decoder@1.1.1", "", { "dependencies": { "safe-buffer": "~5.1.0" } }, "sha512-n/ShnvDi6FHbbVfviro+WojiFzv+s8MPMHBczVePfUpDJLwoLT0ht1l4YwBCbi8pJAveEEdnkHyPyTP/mzRfwg=="],
"minipass-flush/minipass/yallist": ["yallist@4.0.0", "", {}, "sha512-3wdGidZyq5PB084XLES5TpOSRA3wjXAlIWMhum2kRcv/41Sn2emQ0dycQW4uZXLejwKvg6EsvbdlVL+FYEct7A=="],
"minipass-pipeline/minipass/yallist": ["yallist@4.0.0", "", {}, "sha512-3wdGidZyq5PB084XLES5TpOSRA3wjXAlIWMhum2kRcv/41Sn2emQ0dycQW4uZXLejwKvg6EsvbdlVL+FYEct7A=="],
"motion/framer-motion/motion-dom": ["motion-dom@12.40.0", "", { "dependencies": { "motion-utils": "^12.39.0" } }, "sha512-HxU3ZaBwNPVQUBQf1xxgq+7JrPNZvjLVxgbpEZL7RrWJnsxOf0/OM+yrHG9ogLQ31Do/r57Oz2gQWPK+6q62mg=="],
"motion/framer-motion/motion-utils": ["motion-utils@12.39.0", "", {}, "sha512-8nadJAJjTtqRkmRF36FoJTrywK9nnFmnPwnSMyxaOCU7GDjN9RTMJIxx9De8ErM+vpPhMccr/6fo5WciyQLnMQ=="],
@@ -7098,8 +7091,6 @@
"opencode/@ai-sdk/cerebras/@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@4.0.33", "", { "dependencies": { "@ai-sdk/provider": "3.0.12", "@standard-schema/spec": "^1.1.0", "eventsource-parser": "^3.0.8" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-nJ0bAfegMAIJtrzMJtbzer1cS3nb7c7DsyU1S4nrPm7ZU0Mn6SBBZv5IGZZGTbpWTJwqKTSPeZJTXalbAxt1BA=="],
"openid-client/lru-cache/yallist": ["yallist@4.0.0", "", {}, "sha512-3wdGidZyq5PB084XLES5TpOSRA3wjXAlIWMhum2kRcv/41Sn2emQ0dycQW4uZXLejwKvg6EsvbdlVL+FYEct7A=="],
"p-locate/p-limit/yocto-queue": ["yocto-queue@0.1.0", "", {}, "sha512-rVksvsnNCdJ/ohGc6xgPwyN8eheCxsiLM8mxuE/t/mOVqJewPuO1miLpTHQiRgTKCLexL4MeAFVagts7HmNZ2Q=="],
"pkg-dir/find-up/locate-path": ["locate-path@5.0.0", "", { "dependencies": { "p-locate": "^4.1.0" } }, "sha512-t7hw9pI+WvuwNJXwk5zVHpyhIqzg2qTlklJOf0mVxGSbe3Fp2VieZcduNYjaLDoy6p9uGpQEGWG87WpMKlNq8g=="],
@@ -7368,8 +7359,6 @@
"app-builder-lib/@electron/get/fs-extra/universalify": ["universalify@0.1.2", "", {}, "sha512-rBJeI5CXAlmy1pV+617WB9J63U6XcazHHF2f2dbJix4XzpUF0RS3Zbj0FGIOCAva5P/d/GBOYaACQ1w+0azUkg=="],
"app-builder-lib/hosted-git-info/lru-cache/yallist": ["yallist@4.0.0", "", {}, "sha512-3wdGidZyq5PB084XLES5TpOSRA3wjXAlIWMhum2kRcv/41Sn2emQ0dycQW4uZXLejwKvg6EsvbdlVL+FYEct7A=="],
"archiver-utils/glob/jackspeak/@isaacs/cliui": ["@isaacs/cliui@8.0.2", "", { "dependencies": { "string-width": "^5.1.2", "string-width-cjs": "npm:string-width@^4.2.0", "strip-ansi": "^7.0.1", "strip-ansi-cjs": "npm:strip-ansi@^6.0.1", "wrap-ansi": "^8.1.0", "wrap-ansi-cjs": "npm:wrap-ansi@^7.0.0" } }, "sha512-O8jcjabXaleOG9DQ0+ARXWZBTfnP4WNAqzuiJK7ll44AmxGKv/J2M4TPjxjY3znBCfvBXFzucm1twdyFybFqEA=="],
"archiver-utils/glob/minimatch/brace-expansion": ["brace-expansion@2.1.1", "", { "dependencies": { "balanced-match": "^1.0.0" } }, "sha512-WR1cURNjuvBLMZBMbqM0UoE+WAfdUcEV1ccD8PVBVOI+Z3ND4+SZbN8RsfT2bMuG1qwz5RFvPukSZm5fF2D5eA=="],
@@ -7392,6 +7381,8 @@
"editorconfig/minimatch/brace-expansion/balanced-match": ["balanced-match@1.0.2", "", {}, "sha512-3oSeUO0TMV67hN1AmbXsK4yaqU7tjiHlbxRDZOpH0KW9+CeX4bRAaX0Anxt0tx2MrpRpWwQaPwIlISEJhYU5Pw=="],
"electron-builder-squirrel-windows/app-builder-lib/@electron/get/env-paths": ["env-paths@2.2.1", "", {}, "sha512-+h1lkLKhZMTYjog1VEpJNG7NZJWcuc2DDk/qsqSTRRCOXiLjeQ1d1/udrUGhqMxUgAlwKNZ0cf2uqan5GLuS2A=="],
"electron-builder-squirrel-windows/app-builder-lib/@electron/get/fs-extra": ["fs-extra@8.1.0", "", { "dependencies": { "graceful-fs": "^4.2.0", "jsonfile": "^4.0.0", "universalify": "^0.1.0" } }, "sha512-yhlQgA6mnOJUKOsRUFsgJdQCvkKhcz8tlZG5HBQfReYZy46OwLcY+Zia0mtdHsOo9y/hP+CxMN0TU9QxoOtG4g=="],
"electron-builder-squirrel-windows/app-builder-lib/@electron/get/semver": ["semver@6.3.1", "", { "bin": { "semver": "bin/semver.js" } }, "sha512-BR7VvDCVHO+q2xBEWskxS6DJE1qRnb7DxzUrogb71CWoSficBxYsiAGd+Kl0mmq/MprG9yArRkyrQxTO6XjMzA=="],
@@ -7500,8 +7491,6 @@
"electron-builder-squirrel-windows/app-builder-lib/@electron/get/fs-extra/universalify": ["universalify@0.1.2", "", {}, "sha512-rBJeI5CXAlmy1pV+617WB9J63U6XcazHHF2f2dbJix4XzpUF0RS3Zbj0FGIOCAva5P/d/GBOYaACQ1w+0azUkg=="],
"electron-builder-squirrel-windows/app-builder-lib/hosted-git-info/lru-cache/yallist": ["yallist@4.0.0", "", {}, "sha512-3wdGidZyq5PB084XLES5TpOSRA3wjXAlIWMhum2kRcv/41Sn2emQ0dycQW4uZXLejwKvg6EsvbdlVL+FYEct7A=="],
"electron-builder/yargs/cliui/strip-ansi/ansi-regex": ["ansi-regex@5.0.1", "", {}, "sha512-quJQXlTSUGL2LH9SUXo8VwsY4soanhgo6LNSm84E1LBcE8s3O0wpdiRzyR9z/ZZJMlMWv37qOOb9pdJlMUEKFQ=="],
"electron-builder/yargs/string-width/strip-ansi/ansi-regex": ["ansi-regex@5.0.1", "", {}, "sha512-quJQXlTSUGL2LH9SUXo8VwsY4soanhgo6LNSm84E1LBcE8s3O0wpdiRzyR9z/ZZJMlMWv37qOOb9pdJlMUEKFQ=="],
+4 -4
View File
@@ -1,8 +1,8 @@
{
"nodeModules": {
"x86_64-linux": "sha256-ZmAIcG3wPgssKejLe/EMJOjPuMHCyRiiItRvSqIjaME=",
"aarch64-linux": "sha256-lbAHoVI/CvgdrfKZ/9dKZ/O7DxJvuWEOrDgcCuFGPBk=",
"aarch64-darwin": "sha256-Nrs4ijvzQ3J8AYigJSlsr66feUvLZ5iiYCfBdaZNJEc=",
"x86_64-darwin": "sha256-ltCaJSO7kZWuMOU49269xvboNVX35OJDFj0gXCDf62Y="
"x86_64-linux": "sha256-ovrz0pALxRek5VEqSYkVpzb9YeiiVQWxfAHZQ3kX0/0=",
"aarch64-linux": "sha256-+sqBt11Nl1fDrL1JvFAcN8JMWahspfbBPuMeM7YdagU=",
"aarch64-darwin": "sha256-PsZfcmMpz/Xzudjv4CQZu+wPfnyijrlVYeP2x9tRffk=",
"x86_64-darwin": "sha256-4TpIc3Gh9nDaA5Y1zkCq0KraMdWu+PWtYVY3p7/CsJ0="
}
}
-1
View File
@@ -99,7 +99,6 @@
"@typescript/native-preview": "catalog:",
"glob": "13.0.5",
"husky": "9.1.7",
"node-gyp": "12.4.0",
"oxlint": "1.60.0",
"oxlint-tsgolint": "0.21.0",
"prettier": "3.6.2",
@@ -0,0 +1,19 @@
import { describe, expect, test } from "bun:test"
import { matchesModelSearch } from "./dialog-select-model-search"
describe("matchesModelSearch", () => {
test("matches model names across separators", () => {
expect(matchesModelSearch("gpt 5", ["GPT-5.5"])).toBe(true)
expect(matchesModelSearch("gpt-5", ["GPT-5.5"])).toBe(true)
expect(matchesModelSearch("gpt5", ["GPT-5.5"])).toBe(true)
})
test("matches any searchable model field", () => {
expect(matchesModelSearch("open ai", ["GPT-5.5", "gpt-5.5", "OpenAI"])).toBe(true)
expect(matchesModelSearch("gpt 5", ["GPT-5.5", "gpt-5.5", "OpenAI"])).toBe(true)
})
test("does not match unrelated searches", () => {
expect(matchesModelSearch("claude", ["GPT-5.5", "gpt-5.5", "OpenAI"])).toBe(false)
})
})
@@ -0,0 +1,18 @@
export const normalizeModelSearch = (value: string) =>
value
.toLowerCase()
.replace(/[^\p{Letter}\p{Number}]+/gu, " ")
.trim()
.replace(/\s+/g, " ")
export const compactModelSearch = (value: string) => normalizeModelSearch(value).replaceAll(" ", "")
export const matchesModelSearch = (query: string, values: string[]) => {
const search = normalizeModelSearch(query)
if (!search) return true
const compactSearch = compactModelSearch(query)
return values.some(
(value) => normalizeModelSearch(value).includes(search) || compactModelSearch(value).includes(compactSearch),
)
}
@@ -1,5 +1,15 @@
import { Popover as Kobalte } from "@kobalte/core/popover"
import { Component, ComponentProps, createMemo, For, JSX, Show, ValidComponent } from "solid-js"
import {
Component,
ComponentProps,
createEffect,
createMemo,
For,
JSX,
onCleanup,
Show,
ValidComponent,
} from "solid-js"
import { createStore } from "solid-js/store"
import { useLocal } from "@/context/local"
import { useDialog } from "@opencode-ai/ui/context/dialog"
@@ -17,6 +27,9 @@ import { MenuV2 } from "@opencode-ai/ui/v2/menu-v2"
import { ModelTooltip } from "./model-tooltip"
import { useLanguage } from "@/context/language"
import { decode64 } from "@/utils/base64"
import { handleDocumentSearchKeydown } from "@/utils/search-keydown"
import { createEventListener } from "@solid-primitives/event-listener"
import { matchesModelSearch } from "./dialog-select-model-search"
const isFree = (provider: string, cost: { input: number } | undefined) =>
provider === "opencode" && (!cost || cost.input === 0)
@@ -242,14 +255,9 @@ export function ModelSelectorPopoverV2(props: {
.filter((item) => (props.provider ? item.provider.id === props.provider : true)),
)
const models = createMemo(() => {
const search = store.search.trim().toLowerCase()
const search = store.search.trim()
const filtered = search
? allModels().filter(
(item) =>
item.name.toLowerCase().includes(search) ||
item.id.toLowerCase().includes(search) ||
item.provider.name.toLowerCase().includes(search),
)
? allModels().filter((item) => matchesModelSearch(search, [item.name, item.id, item.provider.name]))
: allModels()
return [...filtered].sort((a, b) => a.name.localeCompare(b.name))
@@ -333,19 +341,23 @@ export function ModelSelectorPopoverV2(props: {
queueMicrotask(() => activeItem()?.scrollIntoView({ block: "nearest" }))
}
const setSearch = (value: string) => {
const search = value.trim().toLowerCase()
const search = value.trim()
const first = [...allModels()]
.sort((a, b) => a.name.localeCompare(b.name))
.find(
(item) =>
!search ||
item.name.toLowerCase().includes(search) ||
item.id.toLowerCase().includes(search) ||
item.provider.name.toLowerCase().includes(search),
)
.find((item) => matchesModelSearch(search, [item.name, item.id, item.provider.name]))
setStore({ search: value, active: first ? modelKey(first) : manageKey })
}
createEffect(() => {
if (!store.open) return
createEventListener(
document,
"keydown",
(event: KeyboardEvent) => handleDocumentSearchKeydown(searchRef, event, store.search, setSearch),
true,
)
})
return (
<MenuV2 open={store.open} modal={false} placement="top-start" gutter={6} onOpenChange={setOpen}>
<MenuV2.Trigger as={props.triggerAs ?? "div"} {...props.triggerProps}>
+3 -1
View File
@@ -218,6 +218,7 @@ export const PromptInput: Component<PromptInputProps> = (props) => {
let scrollRef!: HTMLDivElement
let slashPopoverRef!: HTMLDivElement
let restoreEndOnFocus = true
let savedCursor: number | null = null
const mirror = { input: false }
const inset = 56
@@ -590,7 +591,7 @@ export const PromptInput: Component<PromptInputProps> = (props) => {
const restoreFocus = () => {
requestAnimationFrame(() => {
const cursor = prompt.cursor() ?? promptLength(prompt.current())
const cursor = savedCursor ?? prompt.cursor() ?? promptLength(prompt.current())
editorRef.focus()
setCursorPosition(editorRef, cursor)
queueScroll()
@@ -627,6 +628,7 @@ export const PromptInput: Component<PromptInputProps> = (props) => {
const isImeComposing = (event: KeyboardEvent) => event.isComposing || composing() || event.keyCode === 229
const handleBlur = () => {
savedCursor = currentCursor()
closePopover()
setComposing(false)
}
@@ -1,4 +1,4 @@
import { For, Show, splitProps, type Accessor, type ComponentProps } from "solid-js"
import { createEffect, For, onCleanup, Show, splitProps, type Accessor, type ComponentProps } from "solid-js"
import { createStore } from "solid-js/store"
import { DropdownMenu } from "@opencode-ai/ui/dropdown-menu"
import { Icon } from "@opencode-ai/ui/icon"
@@ -8,6 +8,7 @@ import { getProjectAvatarVariant } from "@/context/layout"
import { useLanguage } from "@/context/language"
import { displayName, getProjectAvatarSource } from "@/pages/layout/helpers"
import { pathKey } from "@/utils/path-key"
import { handleDocumentSearchKeydown } from "@/utils/search-keydown"
export type PromptProject = {
name?: string
@@ -101,6 +102,16 @@ export function createPromptProjectController(input: {
setStore({ open: false, search: "", active: "" })
input.controls().add(language.t("command.project.open"), server)
}
const setSearch = (value: string) => {
const search = value.trim().toLowerCase()
const first = input
.controls()
.available.find((project) => !search || displayName(project).toLowerCase().includes(search))
setStore({
search: value,
active: first ? projectKey(first) : actionKey(servers().length > 1 ? undefined : servers()[0]?.key),
})
}
return {
selected,
@@ -127,16 +138,7 @@ export function createPromptProjectController(input: {
}
setStore({ open: false, search: "", active: "" })
},
setSearch(value: string) {
const search = value.trim().toLowerCase()
const first = input
.controls()
.available.find((project) => !search || displayName(project).toLowerCase().includes(search))
setStore({
search: value,
active: first ? projectKey(first) : actionKey(servers().length > 1 ? undefined : servers()[0]?.key),
})
},
setSearch,
clearSearch() {
setStore({ search: "", active: initialActive() })
setTimeout(() => searchRef?.focus())
@@ -170,6 +172,9 @@ export function createPromptProjectController(input: {
focusSearch() {
setTimeout(() => requestAnimationFrame(() => searchRef?.focus()))
},
handleSearchKeydown(event: KeyboardEvent) {
return handleDocumentSearchKeydown(searchRef, event, store.search, setSearch)
},
}
}
@@ -243,6 +248,13 @@ export function PromptProjectSelector(props: {
return project ? props.controller.projectKey(project) : undefined
}
createEffect(() => {
if (!props.controller.open()) return
const handler = (event: KeyboardEvent) => props.controller.handleSearchKeydown(event)
document.addEventListener("keydown", handler, true)
onCleanup(() => document.removeEventListener("keydown", handler, true))
})
return (
<DropdownMenu
open={props.controller.open()}
@@ -399,7 +399,7 @@
overflow: hidden;
font-size: 13px;
font-weight: 440;
line-height: 1;
line-height: 16px;
text-overflow: ellipsis;
white-space: nowrap;
}
@@ -3,7 +3,7 @@ import { createSignal, Show, type JSXElement } from "solid-js"
import "./titlebar-tab-popover.css"
// Initial hover delay before the preview appears, per design.
const OPEN_DELAY = 200
const OPEN_DELAY = 400
// Mouse-out delay: begin closing immediately (a brief exit animation plays).
const CLOSE_DELAY = 0
// After a preview closes, hovering a neighbouring tab within this window skips
+11 -1
View File
@@ -1,7 +1,8 @@
import { createSimpleContext } from "@opencode-ai/ui/context"
import { createEffect, createMemo, createRoot } from "solid-js"
import { createStore } from "solid-js/store"
import { createServerProjects, ServerConnection, useServer } from "./server"
import { createServerProjects, RECENTLY_CLOSED_DISPLAY_LIMIT, ServerConnection, useServer } from "./server"
import { pathKey } from "@/utils/path-key"
import { useServerHealth } from "@/utils/server-health"
import { createServerSdkContext } from "./server-sdk"
import { createServerSyncContext } from "./server-sync"
@@ -127,6 +128,14 @@ function createServerCtx(
}
const projectsList = createMemo(() => projects.list().map(enrich))
const recentlyClosedList = createMemo(() => {
const known = new Set(sync.data.project.map((project) => pathKey(project.worktree)))
return projects
.recentlyClosed()
.filter((worktree) => known.has(pathKey(worktree)))
.slice(0, RECENTLY_CLOSED_DISPLAY_LIMIT)
.map((worktree) => enrich({ worktree, expanded: false }))
})
const isLocal =
(conn?.type === "sidecar" && conn.variant === "base") || (conn?.type === "http" && isLocalHost(conn.http.url))
@@ -139,6 +148,7 @@ function createServerCtx(
projects: {
...projects,
list: projectsList,
recentlyClosed: recentlyClosedList,
},
}
}
+11 -2
View File
@@ -5,10 +5,11 @@ import { createSimpleContext } from "@opencode-ai/ui/context"
import { makeEventListener } from "@solid-primitives/event-listener"
import { useServerSync } from "./server-sync"
import { useServerSDK } from "./server-sdk"
import { ServerConnection, useServer } from "./server"
import { RECENTLY_CLOSED_DISPLAY_LIMIT, ServerConnection, useServer } from "./server"
import { usePlatform } from "./platform"
import { Project } from "@opencode-ai/sdk/v2"
import { Persist, persisted, removePersisted } from "@/utils/persist"
import { pathKey } from "@/utils/path-key"
import { decode64 } from "@/utils/base64"
import { same } from "@/utils/same"
import { createScrollPersistence, type SessionScroll } from "./layout-scroll"
@@ -493,7 +494,7 @@ export const { use: useLayout, provider: LayoutProvider } = createSimpleContext(
const root = rootFor(project.worktree)
if (root === project.worktree) continue
server.projects.close(project.worktree)
server.projects.remove(project.worktree)
if (!seen.has(root)) {
server.projects.open(root)
@@ -613,6 +614,14 @@ export const { use: useLayout, provider: LayoutProvider } = createSimpleContext(
},
projects: {
list,
recentlyClosed: createMemo(() => {
const known = new Set(serverSync().data.project.map((project) => pathKey(project.worktree)))
return server.projects
.recentlyClosed()
.filter((worktree) => known.has(pathKey(worktree)))
.slice(0, RECENTLY_CLOSED_DISPLAY_LIMIT)
.map((worktree) => enrich({ worktree, expanded: false }))
}),
open(directory: string) {
const root = rootFor(directory)
if (server.projects.list().find((x) => x.worktree === root)) return
+83 -1
View File
@@ -99,7 +99,7 @@ describe("createServerProjects", () => {
test("keeps active and explicit server buckets in one reactive store", () => {
createRoot((dispose) => {
const [scope] = createSignal(ServerScope.local)
const [store, setStore] = createStore({ projects: {}, lastProject: {} })
const [store, setStore] = createStore({ projects: {}, lastProject: {}, recentlyClosed: {} })
const active = createServerProjects({ scope, store, setStore })
const remote = createServerProjects({ scope: () => "https://debian.example" as ServerScope, store, setStore })
@@ -115,6 +115,88 @@ describe("createServerProjects", () => {
dispose()
})
})
test("tracks recently closed projects and drops them when reopened", () => {
createRoot((dispose) => {
const [scope] = createSignal(ServerScope.local)
const [store, setStore] = createStore({ projects: {}, lastProject: {}, recentlyClosed: {} })
const projects = createServerProjects({ scope, store, setStore })
projects.open("/a")
projects.open("/b")
projects.close("/a")
expect(projects.recentlyClosed()).toEqual(["/a"])
projects.close("/b")
expect(projects.recentlyClosed()).toEqual(["/b", "/a"])
projects.open("/a")
expect(projects.recentlyClosed()).toEqual(["/b"])
expect(projects.list()).toEqual([{ worktree: "/a", expanded: true }])
dispose()
})
})
test("remove drops a project without recording it as recently closed", () => {
createRoot((dispose) => {
const [scope] = createSignal(ServerScope.local)
const [store, setStore] = createStore({ projects: {}, lastProject: {}, recentlyClosed: {} })
const projects = createServerProjects({ scope, store, setStore })
projects.open("/repo/subdir")
projects.remove("/repo/subdir")
expect(projects.list()).toEqual([])
expect(projects.recentlyClosed()).toEqual([])
dispose()
})
})
test("retains recently closed history beyond the visible display limit", () => {
createRoot((dispose) => {
const [scope] = createSignal(ServerScope.local)
const [store, setStore] = createStore({ projects: {}, lastProject: {}, recentlyClosed: {} })
const projects = createServerProjects({ scope, store, setStore })
// Closing 6 projects keeps all 6 in the store even though only 5 are displayed;
// this prevents display-filtered entries from evicting still-visible ones.
for (const dir of ["/1", "/2", "/3", "/4", "/5", "/6"]) {
projects.open(dir)
projects.close(dir)
}
expect(projects.recentlyClosed()).toEqual(["/6", "/5", "/4", "/3", "/2", "/1"])
dispose()
})
})
test("caps recently closed history at the store limit", () => {
createRoot((dispose) => {
const [scope] = createSignal(ServerScope.local)
const [store, setStore] = createStore({ projects: {}, lastProject: {}, recentlyClosed: {} })
const projects = createServerProjects({ scope, store, setStore })
for (let i = 1; i <= 20; i++) {
projects.open(`/p${i}`)
projects.close(`/p${i}`)
}
expect(projects.recentlyClosed()).toHaveLength(16)
expect(projects.recentlyClosed()[0]).toBe("/p20")
expect(projects.recentlyClosed().at(-1)).toBe("/p5")
dispose()
})
})
test("dedupes recently closed entries by normalized path", () => {
createRoot((dispose) => {
const [scope] = createSignal(ServerScope.local)
const [store, setStore] = createStore({ projects: {}, lastProject: {}, recentlyClosed: {} })
const projects = createServerProjects({ scope, store, setStore })
projects.close("/repo")
projects.close("/repo/")
expect(projects.recentlyClosed()).toEqual(["/repo/"])
dispose()
})
})
})
describe("migrateCanonicalLocalServerState", () => {
+40 -5
View File
@@ -2,12 +2,23 @@ import { createSimpleContext } from "@opencode-ai/ui/context"
import { type Accessor, batch, createMemo } from "solid-js"
import { createStore, type SetStoreFunction, type Store } from "solid-js/store"
import { Persist, persisted } from "@/utils/persist"
import { pathKey } from "@/utils/path-key"
import { ServerScope } from "@/utils/server-scope"
type StoredProject = { worktree: string; expanded: boolean }
type StoredServer = string | ServerConnection.HttpBase | ServerConnection.Http
type ServerProjectState = { projects: Record<string, StoredProject[]>; lastProject: Record<string, string> }
type ServerProjectState = {
projects: Record<string, StoredProject[]>
lastProject: Record<string, string>
recentlyClosed: Record<string, string[]>
}
const HEALTH_POLL_INTERVAL_MS = 10_000
// The store retains more history than is displayed. Consumers filter recently closed entries
// against the live project list (dropping deleted projects) and then cap the visible count via
// RECENTLY_CLOSED_DISPLAY_LIMIT. Retaining extra history ensures entries that are temporarily
// filtered out do not evict still-visible ones from the persisted store.
const RECENTLY_CLOSED_HISTORY_LIMIT = 16
export const RECENTLY_CLOSED_DISPLAY_LIMIT = 5
export function normalizeServerUrl(input: string) {
const trimmed = input.trim()
@@ -72,19 +83,42 @@ export function createServerProjects<T extends ServerProjectState>(input: {
}) {
const setStore = input.setStore as unknown as SetStoreFunction<ServerProjectState>
const current = () => input.store.projects[input.scope()] ?? []
const currentClosed = () => input.store.recentlyClosed?.[input.scope()] ?? []
const remove = (directory: string) => {
setStore(
"projects",
input.scope(),
current().filter((project) => project.worktree !== directory),
)
}
return {
list: current,
recentlyClosed: currentClosed,
remove,
open(directory: string) {
const scope = input.scope()
const key = pathKey(directory)
const closed = currentClosed()
if (closed.some((worktree) => pathKey(worktree) === key)) {
setStore(
"recentlyClosed",
scope,
closed.filter((worktree) => pathKey(worktree) !== key),
)
}
if (current().some((project) => project.worktree === directory)) return
setStore("projects", scope, [{ worktree: directory, expanded: true }, ...current()])
},
// User-initiated close: removes the project and records it in recently closed.
// Internal, non-user removals (e.g. sandbox/worktree normalization) should use remove().
close(directory: string) {
setStore(
"projects",
input.scope(),
current().filter((project) => project.worktree !== directory),
remove(directory)
const key = pathKey(directory)
const closed = [directory, ...currentClosed().filter((worktree) => pathKey(worktree) !== key)].slice(
0,
RECENTLY_CLOSED_HISTORY_LIMIT,
)
setStore("recentlyClosed", input.scope(), closed)
},
expand(directory: string) {
const index = current().findIndex((project) => project.worktree === directory)
@@ -235,6 +269,7 @@ export const { use: useServer, provider: ServerProvider } = createSimpleContext(
list: [] as StoredServer[],
projects: {} as Record<string, StoredProject[]>,
lastProject: {} as Record<string, string>,
recentlyClosed: {} as Record<string, string[]>,
}),
)
+1
View File
@@ -604,6 +604,7 @@ export const dict = {
"home.title": "Home",
"home.projects": "Projects",
"home.project.add": "Add project",
"home.recentlyClosed": "Recently closed",
"home.server.collapse": "Collapse server projects",
"home.server.expand": "Expand server projects",
"home.sessions.search.placeholder": "Search sessions",
+105 -6
View File
@@ -270,6 +270,10 @@ export function NewHome() {
})
const focusedSync = () => focusedServerCtx()?.sync ?? sync()
const projects = createMemo(() => focusedServerCtx()?.projects.list() ?? layout.projects.list())
const recentlyClosed = createMemo(
() => focusedServerCtx()?.projects.recentlyClosed() ?? layout.projects.recentlyClosed(),
)
const homedir = createMemo(() => focusedSync().data.path.home ?? "")
const selectedProject = createMemo(() => projects().find((project) => project.worktree === selection().directory))
const newSessionProject = createMemo(
() =>
@@ -518,10 +522,13 @@ export function NewHome() {
<div class="mx-auto grid h-full w-full max-w-[1080px] grid-rows-[auto_minmax(0,1fr)_auto] gap-4 px-3 lg:grid-cols-[280px_minmax(0,720px)] lg:grid-rows-1 lg:gap-8 lg:px-6">
<HomeProjectColumn
projects={projects()}
recentlyClosed={recentlyClosed()}
homedir={homedir()}
selected={selection()}
focusServer={focusServer}
selectProject={selectProject}
openNewSession={openProjectNewSession}
openRecentProject={(conn, directory) => addProjects(conn, [directory])}
chooseProject={(conn) => void chooseProject(conn)}
editProject={editProject}
closeProject={(conn, directory) => {
@@ -638,10 +645,13 @@ export function NewHome() {
function HomeProjectColumn(props: {
projects: LocalProject[]
recentlyClosed: LocalProject[]
homedir: string
selected: HomeProjectSelection
focusServer: (server: ServerConnection.Any) => void
selectProject: (server: ServerConnection.Any, directory: string) => void
openNewSession: (server: ServerConnection.Any, directory: string) => void
openRecentProject: (server: ServerConnection.Any, directory: string) => void
chooseProject: (server: ServerConnection.Any) => void
editProject: (server: ServerConnection.Any, project: LocalProject) => void
closeProject: (server: ServerConnection.Any, directory: string) => void
@@ -670,8 +680,10 @@ function HomeProjectColumn(props: {
aria-label={props.language.t("home.projects")}
>
<div class="flex h-7 min-w-0 shrink-0 items-center justify-between pl-1.5 pr-3">
<div class={HOME_SECTION_LABEL}>{props.language.t("home.projects")}</div>
<Show when={global.servers.list().length === 1}>
<div class="text-v2-text-text-muted [font-weight:530]">{props.language.t("home.projects")}</div>
<Show
when={global.servers.list().length === 1 && !(props.projects.length === 0 && props.recentlyClosed.length > 0)}
>
<TooltipV2 placement="bottom" value={props.language.t("home.project.add")}>
<IconButtonV2
data-action="home-add-project"
@@ -691,7 +703,21 @@ function HomeProjectColumn(props: {
when={global.servers.list().length > 1}
fallback={
<div class="pr-3">
<HomeProjectList {...props} server={global.servers.list()[0]!} />
<Show
when={props.projects.length > 0}
fallback={
<HomeProjectEmpty
server={global.servers.list()[0]!}
recentlyClosed={props.recentlyClosed}
homedir={props.homedir}
chooseProject={props.chooseProject}
openRecentProject={props.openRecentProject}
language={props.language}
/>
}
>
<HomeProjectList {...props} server={global.servers.list()[0]!} />
</Show>
</div>
}
>
@@ -897,6 +923,79 @@ function HomeProjectList(props: {
)
}
function HomeProjectEmpty(props: {
server: ServerConnection.Any
recentlyClosed: LocalProject[]
homedir: string
chooseProject: (server: ServerConnection.Any) => void
openRecentProject: (server: ServerConnection.Any, directory: string) => void
language: ReturnType<typeof useLanguage>
}) {
const global = useGlobal()
const unreachable = () => global.servers.health[ServerConnection.key(props.server)]?.healthy === false
return (
<div class="flex min-w-0 flex-col gap-1">
<button
type="button"
data-action="home-add-project-row"
class={`${HOME_PROJECT_NAV_ROW} disabled:opacity-60 [&>[data-slot=icon-svg]]:text-v2-icon-icon-muted`}
disabled={unreachable()}
onClick={() => props.chooseProject(props.server)}
>
<IconV2 name="folder-add-left" size="small" />
<span class={HOME_PROJECT_NAV_LABEL}>{props.language.t("home.project.add")}</span>
</button>
<Show when={props.recentlyClosed.length > 0}>
<div class="mt-3 flex h-7 min-w-0 shrink-0 items-center pl-1.5 pr-3">
<div class="text-v2-text-text-faint [font-weight:530]">{props.language.t("home.recentlyClosed")}</div>
</div>
<For each={props.recentlyClosed}>
{(project) => (
<HomeRecentlyClosedRow
project={project}
server={props.server}
homedir={props.homedir}
openRecentProject={props.openRecentProject}
language={props.language}
/>
)}
</For>
</Show>
</div>
)
}
function HomeRecentlyClosedRow(props: {
project: LocalProject
server: ServerConnection.Any
homedir: string
openRecentProject: (server: ServerConnection.Any, directory: string) => void
language: ReturnType<typeof useLanguage>
}) {
const global = useGlobal()
const unreachable = () => global.servers.health[ServerConnection.key(props.server)]?.healthy === false
const path = () => {
const home = props.homedir
const worktree = props.project.worktree
if (home && (worktree === home || worktree.startsWith(`${home}/`))) return `~${worktree.slice(home.length)}`
return worktree
}
return (
<TooltipV2 placement="right" value={path()}>
<button
type="button"
data-component="home-recently-closed-row"
class={`${HOME_PROJECT_NAV_ROW} disabled:opacity-60`}
disabled={unreachable()}
onClick={() => props.openRecentProject(props.server, props.project.worktree)}
>
<HomeProjectAvatar project={props.project} outline />
<span class={HOME_PROJECT_NAV_LABEL}>{displayName(props.project)}</span>
</button>
</TooltipV2>
)
}
function HomeProjectRow(props: {
project: LocalProject
server: ServerConnection.Any
@@ -979,13 +1078,13 @@ function HomeProjectRow(props: {
)
}
function HomeProjectAvatar(props: { project: LocalProject }) {
function HomeProjectAvatar(props: { project: LocalProject; outline?: boolean }) {
const name = createMemo(() => displayName(props.project))
return (
<ProjectAvatar
fallback={name()}
src={getProjectAvatarSource(props.project.id, props.project.icon)}
variant={getProjectAvatarVariant(props.project.icon?.color)}
src={props.outline ? undefined : getProjectAvatarSource(props.project.id, props.project.icon)}
variant={props.outline ? "outline" : getProjectAvatarVariant(props.project.icon?.color)}
/>
)
}
@@ -1393,14 +1393,14 @@ export function MessageTimeline(props: {
<button
type="button"
data-slot="session-title-parent"
class="min-w-0 max-w-[40%] truncate px-2 text-[13px] font-[530] leading-4 tracking-[-0.04px] text-v2-text-text-faint transition-colors hover:text-v2-text-text-muted"
class="min-w-0 max-w-[40%] truncate pl-2 text-[13px] font-[530] leading-4 tracking-[-0.04px] text-v2-text-text-faint transition-colors hover:text-v2-text-text-muted"
onClick={navigateParent}
>
{parentTitle()}
</button>
<span
data-slot="session-title-separator"
class="-translate-y-[0.5px] px-1 text-[11px] font-medium text-v2-text-text-faint"
class="-translate-y-[0.5px] pl-2 pr-1 text-[11px] font-medium text-v2-text-text-faint"
aria-hidden="true"
>
/
@@ -3,6 +3,7 @@ import { useLanguage } from "@/context/language"
import { useLocal } from "@/context/local"
import { useSettings } from "@/context/settings"
import { useDialog } from "@opencode-ai/ui/context/dialog"
import { getCursorPosition, setCursorPosition } from "@/components/prompt-input/editor-dom"
import { useSessionLayout } from "./session-layout"
import { createSessionOwnership } from "./session-ownership"
@@ -26,9 +27,23 @@ export const useComposerCommands = () => {
const chooseModel = async () => {
const owner = sessionOwnership.capture()
const editor = document.querySelector<HTMLElement>('[data-component="prompt-input"]')
const selection = window.getSelection()
const cursor =
editor && selection?.rangeCount && editor.contains(selection.anchorNode) ? getCursorPosition(editor) : null
const restoreComposer = () => {
// Kobalte restores focus during its teardown effect; defer past it so the
// composer keeps focus and the caret returns to where the user left it.
requestAnimationFrame(() => {
const editor = document.querySelector<HTMLElement>('[data-component="prompt-input"]')
if (!editor) return
editor.focus()
if (cursor !== null) setCursorPosition(editor, cursor)
})
}
const { DialogSelectModel } = await import("@/components/dialog-select-model")
owner.run(() => {
void dialog.show(() => <DialogSelectModel model={local.model} />)
void dialog.show(() => <DialogSelectModel model={local.model} />, restoreComposer)
})
}
+116
View File
@@ -0,0 +1,116 @@
const editableSelector = "input, textarea, select, [contenteditable=''], [contenteditable='true']"
export function handleDocumentSearchKeydown(
input: HTMLInputElement | undefined,
event: KeyboardEvent,
inputValue: string,
setInputValue: (value: string) => void,
) {
if (!input) return false
if (event.defaultPrevented || event.isComposing) return false
if (event.target === input) return false
if (event.target instanceof Element && event.target.closest(editableSelector)) return false
const action = searchKeyAction(event)
if (!action) return false
event.preventDefault()
event.stopPropagation()
input.focus()
const start = input.selectionStart ?? inputValue.length
const end = input.selectionEnd ?? inputValue.length
if (action.type === "selectAll") {
input.setSelectionRange(0, inputValue.length)
return true
}
if (action.type === "move") {
moveSelection(input, inputValue, action.delta, event.shiftKey)
return true
}
if (action.type === "home") {
setBoundarySelection(input, start, 0, event.shiftKey)
return true
}
if (action.type === "end") {
setBoundarySelection(input, start, inputValue.length, event.shiftKey)
return true
}
if (action.type === "deleteBackward") {
if (start !== end)
return updateValue(input, inputValue.slice(0, start) + inputValue.slice(end), start, setInputValue)
if (start === 0) return true
return updateValue(input, inputValue.slice(0, start - 1) + inputValue.slice(end), start - 1, setInputValue)
}
if (action.type === "deleteForward") {
if (start !== end)
return updateValue(input, inputValue.slice(0, start) + inputValue.slice(end), start, setInputValue)
if (end === inputValue.length) return true
return updateValue(input, inputValue.slice(0, start) + inputValue.slice(end + 1), start, setInputValue)
}
return updateValue(
input,
inputValue.slice(0, start) + action.value + inputValue.slice(end),
start + action.value.length,
setInputValue,
)
}
function searchKeyAction(event: KeyboardEvent) {
if ((event.ctrlKey || event.metaKey) && !event.altKey && event.key.toLowerCase() === "a") {
return { type: "selectAll" } as const
}
if (event.ctrlKey || event.metaKey || event.altKey) return undefined
if (event.key.length === 1) return { type: "insert", value: event.key } as const
if (event.key === "Backspace") return { type: "deleteBackward" } as const
if (event.key === "Delete") return { type: "deleteForward" } as const
if (event.key === "ArrowLeft") return { type: "move", delta: -1 } as const
if (event.key === "ArrowRight") return { type: "move", delta: 1 } as const
if (event.key === "Home") return { type: "home" } as const
if (event.key === "End") return { type: "end" } as const
return undefined
}
function moveSelection(input: HTMLInputElement, inputValue: string, delta: -1 | 1, extend: boolean) {
const start = input.selectionStart ?? inputValue.length
const end = input.selectionEnd ?? inputValue.length
if (!extend && start !== end) {
const caret = delta < 0 ? start : end
input.setSelectionRange(caret, caret)
return
}
if (!extend) {
const caret = Math.max(0, Math.min(inputValue.length, start + delta))
input.setSelectionRange(caret, caret)
return
}
const backward = input.selectionDirection === "backward"
const anchor = backward ? end : start
const focus = backward ? start : end
const next = Math.max(0, Math.min(inputValue.length, focus + delta))
input.setSelectionRange(Math.min(anchor, next), Math.max(anchor, next), next < anchor ? "backward" : "forward")
}
function setBoundarySelection(input: HTMLInputElement, anchor: number, focus: number, extend: boolean) {
if (!extend) {
input.setSelectionRange(focus, focus)
return
}
input.setSelectionRange(Math.min(anchor, focus), Math.max(anchor, focus), focus < anchor ? "backward" : "forward")
}
function updateValue(input: HTMLInputElement, value: string, caret: number, setInputValue: (value: string) => void) {
input.value = value
setInputValue(value)
input.setSelectionRange(caret, caret)
return true
}
+36 -35
View File
@@ -8,11 +8,11 @@ The package is currently private to this workspace. Its API is designed around t
```ts
// One execution
yield* CodeMode.execute({ tools, code })
yield * CodeMode.execute({ tools, code })
// A reusable runtime
const runtime = CodeMode.make({ tools, limits })
yield* runtime.execute(code)
yield * runtime.execute(code)
// One agent-facing code tool
const codeTool = runtime.agentTool()
@@ -55,7 +55,9 @@ const runtime = CodeMode.make({
},
})
const result = yield* runtime.execute(`
const result =
yield *
runtime.execute(`
const order = await tools.orders.lookup({ id: "order_42" })
return { id: order.id, needsAttention: order.status !== "complete" }
`)
@@ -72,8 +74,8 @@ Successful result values are JSON-safe data. A program that returns `undefined`,
```ts
const tool = Tool.make({
description,
input, // Effect Schema (validating) or JSON Schema (render-only)
output, // optional; same choice
input, // Effect Schema (validating) or JSON Schema (render-only)
output, // optional; same choice
run,
})
```
@@ -89,13 +91,15 @@ The description and schemas are part of the model-visible tool contract. Keep de
Use `CodeMode.execute` for a single execution:
```ts
const result = yield* CodeMode.execute({
tools: { orders: { lookup: lookupOrder } },
code: `return await tools.orders.lookup({ id: "order_42" })`,
limits: { maxToolCalls: 10 },
onToolCallStart: (call) => Effect.logDebug("CodeMode tool started", call),
onToolCallEnd: (call) => Effect.logDebug("CodeMode tool settled", call),
})
const result =
yield *
CodeMode.execute({
tools: { orders: { lookup: lookupOrder } },
code: `return await tools.orders.lookup({ id: "order_42" })`,
limits: { maxToolCalls: 10 },
onToolCallStart: (call) => Effect.logDebug("CodeMode tool started", call),
onToolCallEnd: (call) => Effect.logDebug("CodeMode tool settled", call),
})
```
The Effect environment is inferred from the supplied tools. CodeMode does not erase service requirements introduced by tool implementations.
@@ -110,10 +114,10 @@ const runtime = CodeMode.make({
limits: { timeoutMs: 30_000 },
})
runtime.catalog() // structured tool descriptions
runtime.instructions() // model-facing syntax and tool guide
runtime.catalog() // structured tool descriptions
runtime.instructions() // model-facing syntax and tool guide
runtime.execute(source) // ExecuteResult
runtime.agentTool() // { name, description, input, output, execute }
runtime.agentTool() // { name, description, input, output, execute }
```
`catalog`, `instructions`, and `agentTool` are projections of the same configured tool tree. `agentTool().description` is exactly `instructions()`.
@@ -222,10 +226,10 @@ CodeMode is an orchestration language, not a general JavaScript runtime.
The limits are exactly three knobs:
| Limit | Default | Bounds |
| --- | ---: | --- |
| `timeoutMs` | none - no timeout | Wall-clock execution time. |
| `maxToolCalls` | none - unlimited | Tool calls admitted during the execution. |
| Limit | Default | Bounds |
| ---------------- | -------------------: | -------------------------------------------------------------------- |
| `timeoutMs` | none - no timeout | Wall-clock execution time. |
| `maxToolCalls` | none - unlimited | Tool calls admitted during the execution. |
| `maxOutputBytes` | none - no truncation | Model-facing output: the serialized result value plus captured logs. |
No limit has a default, on purpose: execution budgets are host policy, not library policy - a host that wants a bound sets one; a host that can interrupt the execution fiber (as OpenCode does on user cancel) may set no timeout, and a host with its own tool-output truncation (as OpenCode has) may leave `maxOutputBytes` unset. A host with neither should set `maxOutputBytes`, or oversized results silently flood model context.
@@ -254,28 +258,25 @@ Two interpreter internals are fixed constants rather than knobs: at most 8 tool
Failures are data:
| Kind | Meaning |
| --- | --- |
| `ParseError` | Source is empty or cannot be parsed. |
| `UnsupportedSyntax` | Parsed JavaScript is outside the supported subset. |
| `UnknownTool` | A program referenced a tool the host did not provide. |
| `InvalidToolInput` | Tool input failed schema decoding or safe-data copying. |
| `InvalidToolOutput` | Tool output failed schema decoding or safe-data copying. |
| `InvalidDataValue` | Program data violated the plain-data contract (depth, circularity, blocked properties, non-data values). |
| `ToolCallLimitExceeded` | Calls exceeded `maxToolCalls`. |
| `TimeoutExceeded` | Execution exceeded `timeoutMs`. |
| `ToolFailure` | A tool refused or failed. |
| `ExecutionFailure` | The program threw or another execution error occurred. |
| Kind | Meaning |
| ----------------------- | -------------------------------------------------------------------------------------------------------- |
| `ParseError` | Source is empty or cannot be parsed. |
| `UnsupportedSyntax` | Parsed JavaScript is outside the supported subset. |
| `UnknownTool` | A program referenced a tool the host did not provide. |
| `InvalidToolInput` | Tool input failed schema decoding or safe-data copying. |
| `InvalidToolOutput` | Tool output failed schema decoding or safe-data copying. |
| `InvalidDataValue` | Program data violated the plain-data contract (depth, circularity, blocked properties, non-data values). |
| `ToolCallLimitExceeded` | Calls exceeded `maxToolCalls`. |
| `TimeoutExceeded` | Execution exceeded `timeoutMs`. |
| `ToolFailure` | A tool refused or failed. |
| `ExecutionFailure` | The program threw or another execution error occurred. |
Unknown host failures, defects, invalid outputs, and copying failures are sanitized. To return a safe operational refusal, fail with `toolError`:
```ts
import { toolError } from "@opencode-ai/codemode"
run: ({ id }) =>
authorized(id)
? loadOrder(id)
: Effect.fail(toolError("Order is unavailable"))
run: ({ id }) => (authorized(id) ? loadOrder(id) : Effect.fail(toolError("Order is unavailable")))
```
Only the supplied message is model-visible. The optional cause is never returned in `ExecuteResult`; hosts should perform any required internal logging before crossing this boundary.
+404 -374
View File
@@ -39,6 +39,7 @@ package and was **deleted** in Wave 3 (done, see below).
From issue #34787 and design discussion. Do not relitigate these casually.
### Core direction
- Generic CodeMode lives in its own package: `@opencode-ai/codemode` (repo scope convention;
the issue's `@opencode/codemode` name was normalized to the `@opencode-ai/*` convention).
- **Keep the hand-rolled interpreter.** No QuickJS/V8/sandbox-engine dependency. We own and
@@ -52,6 +53,7 @@ From issue #34787 and design discussion. Do not relitigate these casually.
products/blog posts) in code, comments, commit messages, or docs in this repo.
### MCP / tools
- The MCP adapter lives in OpenCode, not here. It converts MCP definitions into ordinary
`Tool.make(...)` definitions and hands CodeMode a plain tool tree.
- Permissions stay in the OpenCode adapter (each tool's `run` wraps the permission ask).
@@ -61,8 +63,9 @@ From issue #34787 and design discussion. Do not relitigate these casually.
`tools.<server>.<tool>` namespaces before handing them over.
### Discovery / search
- **Search only - no separate `describe`.** `tools.$codemode.search({ query?, namespace?,
limit? })` over the final tool tree, owned by this package.
limit? })` over the final tool tree, owned by this package.
- Search result item shape: `{ path, description, signature }` in an `{ items, total }`
wrapper. The `signature` string embeds the full input/output TypeScript types - in search
results it is the pretty, JSDoc-annotated multiline form (Fix 7), so per-field schema
@@ -82,6 +85,7 @@ From issue #34787 and design discussion. Do not relitigate these casually.
- Tools without an output schema render `unknown` as their return type.
### Schemas / Tool.make
- `Tool.make` carries rich metadata so search can render real signatures.
- Support **Effect Schema** (first-class, validating) and **JSON Schema** (initially
render-only - used for TypeScript rendering; the adapter may validate on its own). Leave
@@ -90,6 +94,7 @@ From issue #34787 and design discussion. Do not relitigate these casually.
normalization for plugin authors can come later.
### Attachments / output
- **No `output.text/file/image` API in v1.** (Deleted in Wave 2.)
- Tool calls return native structured payloads into the sandbox. Files/images emitted by
child tools **never enter the sandbox** - the OpenCode adapter strips and accumulates them
@@ -100,6 +105,7 @@ From issue #34787 and design discussion. Do not relitigate these casually.
image bytes into context or drop attachments.
### Runtime behavior
- Limits are EXACTLY the three public knobs: `{ timeoutMs, maxToolCalls, maxOutputBytes }` -
matching the original locked spec exactly. NO limit has a default (user direction, Fix 6
for the first two; extended to `maxOutputBytes` in the truncation-layering fix below):
@@ -129,7 +135,7 @@ From issue #34787 and design discussion. Do not relitigate these casually.
- `console.*` is captured into `logs` on the result; the host appends them to model-facing
output. Not a tool call; costs no tool budget.
- Simple tool-call **start/end hooks** for nested progress: `onToolCallStart({ index, name,
input })` and `onToolCallEnd({ index, name, input, durationMs, outcome, message? })`.
input })` and `onToolCallEnd({ index, name, input, durationMs, outcome, message? })`.
Interrupted calls fire no end event. No `CurrentToolCall` context service (removed in
Wave 2).
@@ -148,8 +154,9 @@ and `bun run typecheck`; from `packages/opencode`, `bun run typecheck` and
`test/tool/registry.test.ts`).
### Wave 0 - scaffold (done)
- `packages/codemode` created from the experiments implementation: `src/{index,codemode,tool,
tool-error,tool-runtime}.ts`, README, AGENTS.md, tests.
tool-error,tool-runtime}.ts`, README, AGENTS.md, tests.
- `package.json`: name `@opencode-ai/codemode`, deps `acorn@8.15.0`, `typescript: catalog:`,
`effect: catalog:` (both repos pin effect `4.0.0-beta.83`; opencode's effect patch only
touches `unstable/httpapi`, which this package doesn't use).
@@ -157,6 +164,7 @@ and `bun run typecheck`; from `packages/opencode`, `bun run typecheck` and
Context.Service key string renamed to `@opencode-ai/codemode/CurrentToolCall`.
### Wave 1a - forgiving JS semantics (done)
Ported from the old opencode rune work; `test/parity.test.ts` (24 tests) is the acceptance
spec. The seeded interpreter was deliberately strict; these behaviors replaced that:
@@ -173,6 +181,7 @@ spec. The seeded interpreter was deliberately strict; these behaviors replaced t
null/undefined still throws (real JS throws too).
### Wave 1b-i - stdlib value types: Date, RegExp, Map, Set (done)
`src/values.ts` holds `SandboxDate/SandboxRegExp/SandboxMap/SandboxSet` (own module so both
`codemode.ts` and `tool-runtime.ts` import without a cycle). Design:
@@ -202,19 +211,20 @@ spec. The seeded interpreter was deliberately strict; these behaviors replaced t
interpolation renders `/regex/` and ISO dates directly.
### Wave 2 - API layer (done)
The package's public contract, reshaped for the Wave 3 adapter. 101 tests / 0 fail after this
wave; both packages typecheck clean.
- **`Tool.make` schema flexibility** (`src/tool.ts`): `input`/`output` each accept an Effect
Schema (validating, decoded both directions as before) OR a raw JSON Schema document
(render-only - no validation, values pass through; rendering handles `$defs`/`definitions`
+ `$ref`). `output` is **optional** -> signature renders `Promise<unknown>` and the host
result is exposed as-is. Discrimination via `Schema.isSchema`. New helpers exported from
`tool.ts`: `inputTypeScript`/`outputTypeScript`/`decodeInput`/`decodeOutput`/
`jsonSchemaToTypeScript`; `tool-runtime.ts` consumes them (no direct `Schema.*` use there
anymore). Types `JsonSchema`/`ToolSchema` exported from the index. Note: an empty
`Schema.Struct({})` renders as `{ } | Array<unknown>` (effect's JSON Schema emission) -
cosmetic, fixed in Wave 4.
- `$ref`). `output` is **optional** -> signature renders `Promise<unknown>` and the host
result is exposed as-is. Discrimination via `Schema.isSchema`. New helpers exported from
`tool.ts`: `inputTypeScript`/`outputTypeScript`/`decodeInput`/`decodeOutput`/
`jsonSchemaToTypeScript`; `tool-runtime.ts` consumes them (no direct `Schema.*` use there
anymore). Types `JsonSchema`/`ToolSchema` exported from the index. Note: an empty
`Schema.Struct({})` renders as `{ } | Array<unknown>` (effect's JSON Schema emission) -
cosmetic, fixed in Wave 4.
- **`output.*` API deleted**: `OutputItem`(+Schema), result `output` fields, the `output`
global/namespace dispatch, `invokeOutput`/`outputItem`/helpers, interpreter output fields,
instructions line, README section, seeded tests. AGENTS.md keeps a rephrased
@@ -228,22 +238,23 @@ wave; both packages typecheck clean.
(`ToolError`/`ToolRuntimeError` message, else "Tool execution failed"). Interrupted calls
fire no end event (timeout kills the whole execution anyway).
- **Limits collapse**: public `ExecutionLimits` = `{ timeoutMs?, maxToolCalls?,
maxOutputBytes? }` (defaults 10_000 / 100 / 32_000). This wave kept the other knobs as
maxOutputBytes? }` (defaults 10_000 / 100 / 32_000). This wave kept the other knobs as
internal defaults reachable through an `@internal` `InternalExecutionLimits` type; Fix 5
later deleted that type and the internal limit system entirely.
- **`maxOutputBytes` truncation** (CodeMode-owned, never fails): applied via `boundOutput` in
a final `Effect.map` over every result path (success/timeout/normalized failure). Oversized
serialized values become truncated text + ` [result truncated: N bytes exceeds the M-byte
output limit; return a smaller value]`; logs keep leading lines within the remaining budget
+ `[logs truncated: showing K of N lines]`; result gains `truncated: true` (also added to
`ExecuteResultSchema`). UTF-8-safe truncation (no split code points). (The in-sandbox
`maxDataBytes` check that used to throw first on oversized raw values died in Fix 5 -
truncation is now the only result-size mechanism.)
output limit; return a smaller value]`; logs keep leading lines within the remaining budget
- `[logs truncated: showing K of N lines]`; result gains `truncated: true` (also added to
`ExecuteResultSchema`). UTF-8-safe truncation (no split code points). (The in-sandbox
`maxDataBytes` check that used to throw first on oversized raw values died in Fix 5 -
truncation is now the only result-size mechanism.)
- **Search polish**: default limit 12 -> **10** (`defaultSearchLimit`); exact-path lookup - a
trimmed query equal to one tool path (optionally `tools.`-prefixed) returns that tool alone
(`total: 1`), bypassing ranking. Tokenization/ranking/shape unchanged.
### Wave 3 - OpenCode MCP adapter (done)
`packages/opencode/src/session/code-mode.ts` rewritten as a thin adapter over this package;
the vendored rune interpreter is gone. Same `define(mcpTools, mcpDefs, servers)` signature, so
`tools.ts` gating (flag on + MCP tools exist -> single `execute` tool, early-return suppresses
@@ -257,7 +268,7 @@ per-MCP registration; MCP resource tools unaffected) is unchanged.
invoked) - so signature rendering, the inline-vs-search switch, and `$codemode.search`
availability all come from this package and stay consistent with execution.
- **`run` path**: per-child permission ask first (`ctx.ask({ permission: entry.key, patterns:
["*"], always: ["*"] })`, exactly the old gating; approving `execute` approves no child).
["*"], always: ["*"] })`, exactly the old gating; approving `execute` approves no child).
Denials and host failures are mapped to `toolError(message)` so they surface as safe,
catchable in-program failures (MCP `isError` text propagates as `e.message`; without this
they'd be sanitized to "Tool execution failed"). Dispatch reuses the ai-sdk wrapper from
@@ -270,10 +281,10 @@ per-MCP registration; MCP resource tools unaffected) is unchanged.
No handles, no `Result<T>` envelope, no base64 in the sandbox, no data-size tuning (the
`maxDataBytes` budget that existed at the time was deleted in Fix 5).
- **Execute result**: `{ output: formatValue(value) + trailing "Logs:" section (success AND
error - logs are plain pre-formatted lines now), attachments: accumulated }` through the
error - logs are plain pre-formatted lines now), attachments: accumulated }` through the
existing `Tool.ExecuteResult.attachments` -> `message-v2.ts` vision plumbing; attachments
ride on both success and error results. Diagnostic `suggestions` not already contained in
the message are appended to error output. Native outer truncation stays on (adapter never
the message are appended to error output. Native outer truncation stays on (adapter never
sets `metadata.truncated`); CodeMode's own `maxOutputBytes` (32 KB default at the time)
cut first - since the truncation-layering fix, native truncation is the only layer.
Limits: `{ timeoutMs: 30_000 }` at the time (matched the default MCP request timeout);
@@ -296,6 +307,7 @@ per-MCP registration; MCP resource tools unaffected) is unchanged.
describe/`renderType`/`rankTools` tests died with the old design (58+17+24 -> 34+16).
### Wave 4 - instructions/prompting + polish (done)
Instructions are now the budgeted-catalog + prompting-guidance form; verified e2e against a
real MCP config. Package still 101 tests / 0 fail; opencode adapter suites still 34 + 16; both
packages typecheck clean.
@@ -321,7 +333,7 @@ packages typecheck clean.
exported); `CodeMode.execute` (one-shot) passes it too, preserving the
`execute`==`make().execute` law. A speculative `tools.$codemode.search` call on a small
catalog now succeeds instead of `UnknownTool`, and unknown-tool suggestions always point at
search. Search is *advertised* in the instructions only when the inlined list is PARTIAL,
search. Search is _advertised_ in the instructions only when the inlined list is PARTIAL,
keeping small-catalog instructions tight.
- **Prompting content** in `instructions()`, mapping 1:1 to the section 5 transcript failures:
parse-string-results-as-JSON, return-small, console-for-intermediates, and
@@ -340,7 +352,7 @@ packages typecheck clean.
- **E2E (verified, headless)**: from the repo root with `OPENCODE_EXPERIMENTAL_CODE_MODE=1`,
the scratch `.opencode/opencode.jsonc` (context7, github, playwright, sentry, memory,
sequential-thinking; left uncommitted/as-is), and `bun packages/opencode/src/index.ts run
--dangerously-skip-permissions -m opencode/claude-sonnet-4-5 "..."`. Confirmed: a single
--dangerously-skip-permissions -m opencode/claude-sonnet-4-5 "..."`. Confirmed: a single
`execute` tool registered alongside core tools (per-MCP registration suppressed; MCP
resource tools unaffected); the live description read back as "Available tools (PARTIAL -
56 of 88 shown; find the rest with tools.$codemode.search):" with correct per-namespace
@@ -352,6 +364,7 @@ packages typecheck clean.
images, output truncation.
### Wave 5 - Promise generalization (done)
First-class promise values in the interpreter; the direct-tool-call-only `Promise.all`
restriction (and its bespoke AST checks) is gone. Package suite is 136 tests / 0 fail (35 new
in `test/promise.test.ts`); adapter suites and both typechecks unchanged/green; the opencode
@@ -495,7 +508,7 @@ adapter needed **no changes**.
`tools.*`." (the second line drops the tools clause when the tree is empty).
- **`## Workflow`**: numbered steps - find a tool via `tools.$codemode.search` -> read
the `{ path, description, signature }` matches -> call by path -> `typeof res ===
"string" ? JSON.parse(res) : res` -> return only the needed fields. When the catalog is
"string" ? JSON.parse(res) : res` -> return only the needed fields. When the catalog is
COMPLETE the search/read steps collapse into "Pick a tool from the list under
`## Available tools`" and the steps renumber (4 instead of 5).
- **`## Rules`**: call-by-exact-path; TEXT-is-JSON -> JSON.parse; return small (never raw
@@ -526,70 +539,72 @@ adapter needed **no changes**.
**Fix 4 - token-budgeted catalog (was bytes)** (user direction: signatures need a token
budget; namespaces must always be present):
- `src/token.ts` added: copy of `@opencode-ai/core/util/token` (`round(chars / 4)`), so
the package stays dependency-free; keep in sync if the core heuristic changes.
- `DiscoveryOptions.maxInlineCatalogBytes` -> `maxInlineCatalogTokens` (default 4,000
estimated tokens ~ the old 16,000 bytes at 4 chars/token - behavior parity, not a size
reduction). `discoveryPlan` charges `estimate(catalogLine(tool))` per line; cheapest-first
+ stop-on-first-miss unchanged at the time (stop-on-first-miss replaced by round-robin in
- `src/token.ts` added: copy of `@opencode-ai/core/util/token` (`round(chars / 4)`), so
the package stays dependency-free; keep in sync if the core heuristic changes.
- `DiscoveryOptions.maxInlineCatalogBytes` -> `maxInlineCatalogTokens` (default 4,000
estimated tokens ~ the old 16,000 bytes at 4 chars/token - behavior parity, not a size
reduction). `discoveryPlan` charges `estimate(catalogLine(tool))` per line; cheapest-first
- stop-on-first-miss unchanged at the time (stop-on-first-miss replaced by round-robin in
Fix 8). Namespace stub lines were and remain unbudgeted - every
namespace always appears with its tool count, even at budget 0 (asserted in package and
adapter tests).
- Ripple: chars/4 rounding erases small line-length differences, so equal-cost lines fall
to the lexicographic path tiebreak; the adapter's PARTIAL test now asserts the
lexicographic tail (`op_99`) is excluded instead of `op_149`. Fixed-prose measurements
(2026-07): preamble ~44 + Workflow ~146 + Rules ~362 + Syntax ~453 ~ 1,100 tokens fixed;
worst-case net description ~ fixed + 4,000 ~ 5,100 estimated tokens.
- Ripple: chars/4 rounding erases small line-length differences, so equal-cost lines fall
to the lexicographic path tiebreak; the adapter's PARTIAL test now asserts the
lexicographic tail (`op_99`) is excluded instead of `op_149`. Fixed-prose measurements
(2026-07): preamble ~44 + Workflow ~146 + Rules ~362 + Syntax ~453 ~ 1,100 tokens fixed;
worst-case net description ~ fixed + 4,000 ~ 5,100 estimated tokens.
**Fix 5 - internal limits removed** (user direction: only the three PUBLIC limits survive as
configurable knobs; the internal limit system dies):
- `ExecutionLimits` (`timeoutMs` 10_000 / `maxToolCalls` 100 / `maxOutputBytes` 32_000 at
the time; Fix 6 later removed the first two defaults. Same validation: safe integers,
timeoutMs >= 1, others >= 0, RangeError otherwise) is now
the ENTIRE limit surface - exactly the shape section 2's original locked spec named.
`ResolvedExecutionLimits` shrank to those three fields; the `@internal`
`InternalExecutionLimits` type is deleted.
- **Deleted outright**: `maxOperations` and the whole operation-budget machinery
(`recordWork`/`recordOperation`/`budget.operations`, plus the `workUnits`/
`cheapArrayMethods` cost helpers); `maxSourceBytes` (the pre-parse source-size check);
`maxDataBytes` (every byte-accounting path: `runtimeValueBytes`, `boundedProgramValue`,
the container-size caches (`containerSizes`/`objectCounts`), Map/Set incremental `bytes`
fields in `values.ts`, string-growth `limitString` checks, tool-argument/result byte
checks in `tool-runtime.ts`, and the final-result size check); `maxAuditBytes` (log and
audit-trail byte accounting - `toolCalls` records and the start/end hooks are unchanged);
`maxCollectionLength` (every array-length/object-field-count check - this knob was
actively harmful: an MCP tool returning 20k rows failed). The `OperationLimitExceeded`
and `AuditLimitExceeded` diagnostic kinds are gone from the `DiagnosticKind` union and
`ExecuteResultSchema` (fine - the package is unreleased).
- **Fixed constants, not knobs**: `TOOL_CALL_CONCURRENCY = 8` (codemode.ts; the fork
semaphore) and `MAX_VALUE_DEPTH = 32` (tool-runtime.ts; the `copyIn` depth check - kept
only because it produces a clearer error than a native stack-overflow RangeError; still
`InvalidDataValue`). The `DataLimits` plumbing through `tool-runtime.ts` is gone -
`copyIn(value, label)` needs no limits argument, and `ToolRuntime.make` takes just
`(tools, maxToolCalls, hooks?, searchIndex?)`.
- **Verified fact**: timeout interruption does NOT depend on the operation budget - the
Effect fiber runtime auto-yields between interpreter steps, so `timeoutMs` interrupts
even a pure `while (true) {}` loop (empirically verified: a 200ms timeout fired at
~225ms with maxOperations set to MAX_SAFE_INTEGER before the deletion). A regression
test in `codemode.test.ts` asserts exactly this (`while(true){}` + `timeoutMs: 200` ->
`TimeoutExceeded`, elapsed well under a few seconds).
- **Kept (correctness, not budgets)**: circular detection (`copyIn` walks +
`rejectCircularInsertion` on mutations), plain-objects-only, blocked properties
(`__proto__`/`constructor`/`prototype`), data-only checks, and all three public-limit
behaviors unchanged.
- Behavior deltas beyond the intended kills: in-sandbox structures deeper than 32 levels
now fail at the data boundary (`copyIn`) instead of at construction; array index
assignment allows any non-negative integer index (holes permitted, message now "must be
a non-negative integer"); interpreter-produced deep/hostile structures that overflow the
native stack during a walk still normalize to the existing "Execution exceeded the
maximum nesting depth." data diagnostic - failures remain data everywhere.
- Tests: deleted the knob-only tests (stdlib Map/Set collection-length growth x2,
enumeration operation-budget, codemode maxDataBytes/maxSourceBytes/maxOperations/
maxConcurrency-RangeError assertions, and the adapter's runaway-loop-via-operation-limit
test - superseded by the package timeout regression test); rewrote the helpers that used
`InternalExecutionLimits` as a convenience to plain `ExecutionLimits`
(promise/enumeration/stdlib run helpers). Package suite: 154 pass / 0 fail; adapter
suites: 34 + 16.
- `ExecutionLimits` (`timeoutMs` 10_000 / `maxToolCalls` 100 / `maxOutputBytes` 32_000 at
the time; Fix 6 later removed the first two defaults. Same validation: safe integers,
timeoutMs >= 1, others >= 0, RangeError otherwise) is now
the ENTIRE limit surface - exactly the shape section 2's original locked spec named.
`ResolvedExecutionLimits` shrank to those three fields; the `@internal`
`InternalExecutionLimits` type is deleted.
- **Deleted outright**: `maxOperations` and the whole operation-budget machinery
(`recordWork`/`recordOperation`/`budget.operations`, plus the `workUnits`/
`cheapArrayMethods` cost helpers); `maxSourceBytes` (the pre-parse source-size check);
`maxDataBytes` (every byte-accounting path: `runtimeValueBytes`, `boundedProgramValue`,
the container-size caches (`containerSizes`/`objectCounts`), Map/Set incremental `bytes`
fields in `values.ts`, string-growth `limitString` checks, tool-argument/result byte
checks in `tool-runtime.ts`, and the final-result size check); `maxAuditBytes` (log and
audit-trail byte accounting - `toolCalls` records and the start/end hooks are unchanged);
`maxCollectionLength` (every array-length/object-field-count check - this knob was
actively harmful: an MCP tool returning 20k rows failed). The `OperationLimitExceeded`
and `AuditLimitExceeded` diagnostic kinds are gone from the `DiagnosticKind` union and
`ExecuteResultSchema` (fine - the package is unreleased).
- **Fixed constants, not knobs**: `TOOL_CALL_CONCURRENCY = 8` (codemode.ts; the fork
semaphore) and `MAX_VALUE_DEPTH = 32` (tool-runtime.ts; the `copyIn` depth check - kept
only because it produces a clearer error than a native stack-overflow RangeError; still
`InvalidDataValue`). The `DataLimits` plumbing through `tool-runtime.ts` is gone -
`copyIn(value, label)` needs no limits argument, and `ToolRuntime.make` takes just
`(tools, maxToolCalls, hooks?, searchIndex?)`.
- **Verified fact**: timeout interruption does NOT depend on the operation budget - the
Effect fiber runtime auto-yields between interpreter steps, so `timeoutMs` interrupts
even a pure `while (true) {}` loop (empirically verified: a 200ms timeout fired at
~225ms with maxOperations set to MAX_SAFE_INTEGER before the deletion). A regression
test in `codemode.test.ts` asserts exactly this (`while(true){}` + `timeoutMs: 200` ->
`TimeoutExceeded`, elapsed well under a few seconds).
- **Kept (correctness, not budgets)**: circular detection (`copyIn` walks +
`rejectCircularInsertion` on mutations), plain-objects-only, blocked properties
(`__proto__`/`constructor`/`prototype`), data-only checks, and all three public-limit
behaviors unchanged.
- Behavior deltas beyond the intended kills: in-sandbox structures deeper than 32 levels
now fail at the data boundary (`copyIn`) instead of at construction; array index
assignment allows any non-negative integer index (holes permitted, message now "must be
a non-negative integer"); interpreter-produced deep/hostile structures that overflow the
native stack during a walk still normalize to the existing "Execution exceeded the
maximum nesting depth." data diagnostic - failures remain data everywhere.
- Tests: deleted the knob-only tests (stdlib Map/Set collection-length growth x2,
enumeration operation-budget, codemode maxDataBytes/maxSourceBytes/maxOperations/
maxConcurrency-RangeError assertions, and the adapter's runaway-loop-via-operation-limit
test - superseded by the package timeout regression test); rewrote the helpers that used
`InternalExecutionLimits` as a convenience to plain `ExecutionLimits`
(promise/enumeration/stdlib run helpers). Package suite: 154 pass / 0 fail; adapter
suites: 34 + 16.
**Fix 6 - no default timeout / tool-call cap** (user direction): `timeoutMs` and
`maxToolCalls` lost their defaults (were 10_000 / 100) - absent now means no timeout /
@@ -639,196 +654,200 @@ adapter suites: 34 + 16.
**Fix 8 - condensed instructions + round-robin catalog fairness + plural-aware search**
(user direction: the fixed instruction prose was too verbose; two discovery fixes ride
along). All in `tool-runtime.ts`; no interpreter changes.
- **Syntax section inverted**: the three dense allowlist lines (~453 estimated tokens)
are replaced by four short lines (~188) built on "models already know JavaScript; name
only what is unusual or missing": (1) standard modern JS works - functions/closures,
destructuring, template literals, loops, try/catch, spread, optional chaining, the
usual Array/String/Object/Math/JSON methods, plus Date/RegExp/Map/Set and
Promise.all/allSettled/race/resolve/reject; (2) TypeScript type annotations are
stripped before execution, decorators are not supported; (3) NOT supported (each fails
with a message naming the alternative): classes, generators, for await...of,
.then/.catch/.finally (use await with try/catch), `x instanceof Error` (caught errors
are plain `{ name, message }` objects), splice; (4) the data-boundary note (Dates ->
ISO strings; Map/Set/RegExp -> `{}`). Every claim was verified against the interpreter
before writing: probed empirically - classes/generators/for-await/.then/.catch/
.finally/`instanceof Error`/splice/decorators/BigInt/labeled statements/tagged
templates/object getters all fail with clear diagnostics; TS annotations/`as`/
interfaces/type aliases are stripped and TS **enums actually work** (transpileModule
compiles them to an IIFE the interpreter runs), hence enums deliberately unmentioned.
`supportedSyntaxMessage` (the in-diagnostic text in `codemode.ts`) is untouched.
- **Workflow/Rules deduped**: the call-by-exact-path, JSON.parse-string-results, and
return-small content now lives ONLY in the numbered Workflow steps (with their
compliance-driving justifications inline: "most tools return JSON as a string", "raw
payloads get truncated and waste context"); Rules keeps only bullets adding new
content - filter/aggregate collections in code, console.* intermediates (logs ride
back), Promise.all parallelism, Object.keys/for...in enumeration, browse-namespace
(PARTIAL only), and the media rule compressed to one line. The no-.then/.catch
guidance moved to the Syntax not-supported line. Content upgrades: the PARTIAL search
step gained query-style guidance (`- short phrases like "list issues" work best`; a
clearly-a-query-string example, not a tool name), and the exact-path guidance is now
"call it with the result's `path` as-is (never guess segments)" / COMPLETE: "use it
as-is rather than guessing segments".
- **Fixed-prose measurements** (instructions split on `"\n## "`, catalog budget 0,
bytes/3.7 - same method as Fix 4; chars/4 in parentheses):
preamble 44 -> 44 (41 -> 41), Workflow 146 -> 187 (135 -> 171), Rules 362 -> 191
(332 -> 176), Syntax 453 -> 188 (419 -> 174); fixed prose total 1,005 -> 610 (927 -> 562),
~ 40% reduction with no behavioral content dropped. Workflow grew slightly because it
absorbed the deduped parse/return-small justifications.
- **Round-robin namespace inlining** (`discoveryPlan`): the ported stop-on-first-miss
behavior (alphabetically-late namespaces starved to "none shown" while an early
namespace inlines everything) is replaced by round-robin fairness - in each round
(namespaces alphabetical), every namespace still holding un-inlined tools attempts to
place its next-cheapest line against the shared token budget; a namespace whose next
line does not fit is done while the others keep going; stop when all are done. Every
namespace gets some representation before any namespace gets everything. Kept:
`estimate` (chars/4) budget accounting, unbudgeted namespace stub lines, per-namespace
`(N tools)`/`(N tools, K shown)`/`(N tools, none shown)` labels, COMPLETE vs PARTIAL
header, alphabetical namespace order in the output, cheapest-first within each
namespace's shown set.
- **Plural/singular search fix**: `tokenize`d terms matched one-directionally (term must
be substring of indexed text), so query "issues" missed a tool whose text only says
"issue". Now each term expands to `termForms` - the term plus naive singular variants
(trailing "es" stripped when length > 3, trailing "s" when length > 2) - and each of
the four field checks passes when ANY form matches. Weights, exact-path lookup, and
namespace scoping untouched. A true plural path match still outranks a singular-only
description match (path substring 8 + searchable 2 > description 4 + searchable 2).
- **Tests**: package instruction/structure assertions updated to the new text; new
syntax-section test (leads with "Standard modern JavaScript works", names the
verified not-supported list, keeps the data-boundary note); the budget-exhaustion
test rewritten to assert the new fairness (alpha.expensive not fitting must NOT
prevent beta.cheap from showing: PARTIAL 2 of 3, `- beta (1 tool)` fully shown); new
plural/singular test (query "issues" finds a singular-only tool; ranking still
prefers the true "issues" path match). Adapter: description assertions updated; the
large-catalog PARTIAL test now asserts `zeta_only_tool` IS shown (`- zeta (1 tool)` +
its inlined line) - it was "none shown" under starvation. README updated (budgeted
catalog paragraph -> round-robin; search paragraph -> singular variants;
instructions-structure paragraph -> new section contents). Package suite: 169 pass /
0 fail; adapter suites: 34 + 16.
- **Syntax section inverted**: the three dense allowlist lines (~453 estimated tokens)
are replaced by four short lines (~188) built on "models already know JavaScript; name
only what is unusual or missing": (1) standard modern JS works - functions/closures,
destructuring, template literals, loops, try/catch, spread, optional chaining, the
usual Array/String/Object/Math/JSON methods, plus Date/RegExp/Map/Set and
Promise.all/allSettled/race/resolve/reject; (2) TypeScript type annotations are
stripped before execution, decorators are not supported; (3) NOT supported (each fails
with a message naming the alternative): classes, generators, for await...of,
.then/.catch/.finally (use await with try/catch), `x instanceof Error` (caught errors
are plain `{ name, message }` objects), splice; (4) the data-boundary note (Dates ->
ISO strings; Map/Set/RegExp -> `{}`). Every claim was verified against the interpreter
before writing: probed empirically - classes/generators/for-await/.then/.catch/
.finally/`instanceof Error`/splice/decorators/BigInt/labeled statements/tagged
templates/object getters all fail with clear diagnostics; TS annotations/`as`/
interfaces/type aliases are stripped and TS **enums actually work** (transpileModule
compiles them to an IIFE the interpreter runs), hence enums deliberately unmentioned.
`supportedSyntaxMessage` (the in-diagnostic text in `codemode.ts`) is untouched.
- **Workflow/Rules deduped**: the call-by-exact-path, JSON.parse-string-results, and
return-small content now lives ONLY in the numbered Workflow steps (with their
compliance-driving justifications inline: "most tools return JSON as a string", "raw
payloads get truncated and waste context"); Rules keeps only bullets adding new
content - filter/aggregate collections in code, console.\* intermediates (logs ride
back), Promise.all parallelism, Object.keys/for...in enumeration, browse-namespace
(PARTIAL only), and the media rule compressed to one line. The no-.then/.catch
guidance moved to the Syntax not-supported line. Content upgrades: the PARTIAL search
step gained query-style guidance (`- short phrases like "list issues" work best`; a
clearly-a-query-string example, not a tool name), and the exact-path guidance is now
"call it with the result's `path` as-is (never guess segments)" / COMPLETE: "use it
as-is rather than guessing segments".
- **Fixed-prose measurements** (instructions split on `"\n## "`, catalog budget 0,
bytes/3.7 - same method as Fix 4; chars/4 in parentheses):
preamble 44 -> 44 (41 -> 41), Workflow 146 -> 187 (135 -> 171), Rules 362 -> 191
(332 -> 176), Syntax 453 -> 188 (419 -> 174); fixed prose total 1,005 -> 610 (927 -> 562),
~ 40% reduction with no behavioral content dropped. Workflow grew slightly because it
absorbed the deduped parse/return-small justifications.
- **Round-robin namespace inlining** (`discoveryPlan`): the ported stop-on-first-miss
behavior (alphabetically-late namespaces starved to "none shown" while an early
namespace inlines everything) is replaced by round-robin fairness - in each round
(namespaces alphabetical), every namespace still holding un-inlined tools attempts to
place its next-cheapest line against the shared token budget; a namespace whose next
line does not fit is done while the others keep going; stop when all are done. Every
namespace gets some representation before any namespace gets everything. Kept:
`estimate` (chars/4) budget accounting, unbudgeted namespace stub lines, per-namespace
`(N tools)`/`(N tools, K shown)`/`(N tools, none shown)` labels, COMPLETE vs PARTIAL
header, alphabetical namespace order in the output, cheapest-first within each
namespace's shown set.
- **Plural/singular search fix**: `tokenize`d terms matched one-directionally (term must
be substring of indexed text), so query "issues" missed a tool whose text only says
"issue". Now each term expands to `termForms` - the term plus naive singular variants
(trailing "es" stripped when length > 3, trailing "s" when length > 2) - and each of
the four field checks passes when ANY form matches. Weights, exact-path lookup, and
namespace scoping untouched. A true plural path match still outranks a singular-only
description match (path substring 8 + searchable 2 > description 4 + searchable 2).
- **Tests**: package instruction/structure assertions updated to the new text; new
syntax-section test (leads with "Standard modern JavaScript works", names the
verified not-supported list, keeps the data-boundary note); the budget-exhaustion
test rewritten to assert the new fairness (alpha.expensive not fitting must NOT
prevent beta.cheap from showing: PARTIAL 2 of 3, `- beta (1 tool)` fully shown); new
plural/singular test (query "issues" finds a singular-only tool; ranking still
prefers the true "issues" path match). Adapter: description assertions updated; the
large-catalog PARTIAL test now asserts `zeta_only_tool` IS shown (`- zeta (1 tool)` +
its inlined line) - it was "none shown" under starvation. README updated (budgeted
catalog paragraph -> round-robin; search paragraph -> singular variants;
instructions-structure paragraph -> new section contents). Package suite: 169 pass /
0 fail; adapter suites: 34 + 16.
**Fix 9 - prompting trims per user review of Fix 8** (user reviewed the condensed
instructions and directed further cuts):
- Default `maxInlineCatalogTokens` 4,000 -> **2,000** (user wants ~2k tokens of signatures
auto-inlined; round-robin fairness from Fix 8 spreads it across all namespaces).
- Console rule and files/images rule DROPPED from `## Rules`. Replaced by a single
`unknown`-treatment warning: "A result typed `Promise<unknown>` has no guaranteed
shape - verify what actually came back before relying on its fields." (Deliberately
does NOT suggest console.log - user review: naming it there nudges models to log AND
return the same data; the prompt stays console-neutral, neither for nor against.)
The media-stripping MECHANISM is unchanged and still tested; only the prose about it
is gone - the `[N images attached]` marker is self-explanatory in context.
- Kept as-is per user: the JSON.parse workflow step (maps to the original motivating
transcript failure; NOT copied from prior art - see section 5 note), the browse-namespace rule
(undecided), no no-fetch/ambient-authority rule added (proposed, not approved).
- Explicitly REJECTED for now: auto-parsing JSON-looking text results at the adapter
boundary ("could get weird" - type flips, program-sees vs tool-sent divergence). Logged
as a next-iteration follow-up below.
- Default `maxInlineCatalogTokens` 4,000 -> **2,000** (user wants ~2k tokens of signatures
auto-inlined; round-robin fairness from Fix 8 spreads it across all namespaces).
- Console rule and files/images rule DROPPED from `## Rules`. Replaced by a single
`unknown`-treatment warning: "A result typed `Promise<unknown>` has no guaranteed
shape - verify what actually came back before relying on its fields." (Deliberately
does NOT suggest console.log - user review: naming it there nudges models to log AND
return the same data; the prompt stays console-neutral, neither for nor against.)
The media-stripping MECHANISM is unchanged and still tested; only the prose about it
is gone - the `[N images attached]` marker is self-explanatory in context.
- Kept as-is per user: the JSON.parse workflow step (maps to the original motivating
transcript failure; NOT copied from prior art - see section 5 note), the browse-namespace rule
(undecided), no no-fetch/ambient-authority rule added (proposed, not approved).
- Explicitly REJECTED for now: auto-parsing JSON-looking text results at the adapter
boundary ("could get weird" - type flips, program-sees vs tool-sent divergence). Logged
as a next-iteration follow-up below.
**DSL-expansion pass - interpreter-surface batch from section 4** (the deferred medium-tier JS
parity items, done as one focused pass; no public API or limit changes):
- **`instanceof` + real Error values**: the `errorConstructors` names (`Error`,
`TypeError`, `RangeError`, `SyntaxError`, `ReferenceError`, `EvalError`, `URIError`) are
bound globals (`ErrorConstructorReference`, callable with or without `new`; `typeof` ->
`"function"`). Error values stay the same plain `{ name, message }` null-prototype
objects as before - the constructor name additionally rides on a NON-ENUMERABLE symbol
key (`ErrorBrand`), which every `Object.entries`-based walk (copyIn/copyOut, spread,
JSON.stringify) is blind to, so serialization is byte-identical to the old shape and the
brand is lost on spread/boundary copies exactly like JS loses the prototype.
`caughtErrorValue` produces `{ name, message }` wrappers via `createErrorValue`, so
caught interpreter AND tool failures are `instanceof Error` and carry the `name` the
equivalent real-JS failure would have (follow-up fix, user-directed - "closest to real
JS"): `InterpreterRuntimeError` gained an `errorName` field ("Error" default) set
fluently at throw sites via `.as(name)` - `JSON.parse` failures are `"SyntaxError"` (and
now include the engine's position detail in the message; safe - derived from the
program-supplied string), invalid regex patterns/flags `"SyntaxError"`, unknown
identifiers and TDZ access `"ReferenceError"`, assignment to a constant `"TypeError"`,
a bad `normalize` form `"RangeError"`; a host Error reaching the catch path directly
keeps its own name when it is one of the standard seven. Tool failures and everything
without a specific analogue stay `"Error"` - internal class names never leak. Specific
names satisfy the specific `instanceof` (`e instanceof SyntaxError`), matching JS.
The operator is handled in `evaluateBinaryExpression`
BEFORE the data-only operand check (like `typeof`, it observes any lhs - promises and
functions included); recognized rhs: the error constructors (a specific type matches its
own brand or `Error`, never a sibling), `Date`/`RegExp`/`Map`/`Set` (sandbox classes),
`Array`, `Object` (any object/function-ish value), `Promise` (`SandboxPromise`), and
`Number`/`String`/`Boolean` (always false - no boxed values exist); anything else is a
catchable error naming the recognized constructors.
- **Array methods**: `splice` (mutating, returns the removed elements; insertions run
`rejectCircularInsertion` like push/unshift; one-arg form removes to the end, undefined
delete count removes nothing), `fill` (circular-checked value) and `copyWithin`
(host-delegated), and `keys`/`values`/`entries` returning **arrays** (the Map/Set
convention - for...of and spread work either way). The `retryableArrayMethods`
"rewrite using map/filter" hint set emptied out and was deleted with its branch; unknown
array properties still read `undefined`.
- **String methods**: `localeCompare(that)` (locale/options arguments ignored - host
default locale; the dominant use is a sort comparator), `normalize(form?)` (invalid form
-> catchable error naming the four valid forms), `trimLeft`/`trimRight` as
trimStart/trimEnd aliases.
- **Actionable regex failures**: `toHostRegex` and `constructRegExp` now show the
offending pattern (or flags) plus the engine reason (deduped "Invalid regular
expression:" prefix via `regexFailureReason`) and a shared escaping hint
(`escapeRegexHint`); flags failures list the valid flag letters; the
replaceAll/matchAll missing-`g` errors spell out the exact `/pattern/g` to write and
the single-match alternative.
- **copyIn split (the important one)**: `copyIn(value, label, preserveSandboxValues =
false)` - recursion moved to a private `copyBounded`; `boundedData` (every intra-sandbox
checkpoint: `Object.*` helpers, coercion/Array.from/join inputs, template
interpolation, expression-result checkpoints) is now `copyIn(value, label, true)`,
which passes `SandboxDate`/`SandboxRegExp`/`SandboxMap`/`SandboxSet` through **by
reference as leaves** (contents not walked - Map/Set members are validated at their
mutation sites) while keeping the depth (`MAX_VALUE_DEPTH`), circularity,
plain-objects-only, blocked-property, and data-only checks; un-awaited promises keep
the await-hinting rejection in BOTH modes (deliberate - JS-parity pass-through was
considered and skipped to preserve the nudge). The HOST boundary (final result,
tool-call arguments, `JSON.stringify`, tool-result intake) uses the default mode and
still serializes JSON forms (Date -> ISO, RegExp/Map/Set -> `{}`); host instances met on
the preserving path are defensively wrapped into sandbox equivalents. Ripple: the
`Object.*` helpers treat sandbox values as empty objects (`Object.keys(map)` -> `[]`,
assign sources contribute nothing, hasOwn -> false - JS has no own enumerable props
there), so interpreter internals (`.map`/`.time`/`.regex`) can never leak; the
template-literal sandbox carve-out collapsed into `boundedData`. Object/array spread
already preserved instances (reference copies, no checkpoint) - now tested.
- **Console formatting**: `formatConsoleArgument` is total and deep
(`formatConsoleValue`): numbers render via `String` (`NaN`/`Infinity`/`-Infinity`
literally - never the JSON `null`; finite numbers match their JSON form), nested
strings are JSON-quoted, sandbox values keep their friendly forms at ANY depth (ISO
date, `/regex/flags`, `Map(n) [...]`, `Set(n) [...]`), opaque references become
in-place `[CodeMode reference]` markers instead of collapsing the whole argument,
cycles render `[Circular]` (reachable via Map/Set members, which mutation never
checkpoints), and depth beyond `MAX_CONSOLE_DEPTH = 32` (fixed constant, not a knob)
degrades to `...` - console can no longer fail a program. `console.table` guards with
`containsOpaqueReference` (sandbox cells render, e.g. ISO dates) and its row/cell
walkers treat sandbox values as scalar cells.
- **Prose**: the instructions Syntax not-supported line dropped its `instanceof
Error`/splice mentions (nothing else reworded); README updated (checkpoint
preservation vs boundary serialization, error values/`instanceof`, new array/string
methods, regex-failure behavior); `supportedSyntaxMessage` left untouched (it lists
supported syntax, was already non-exhaustive, and stays accurate).
- **Tests**: package suite 169 -> 209 (parity: Error/instanceof + real-JS error-name
coverage, splice/fill/copyWithin/keys/values/entries, localeCompare/normalize/trim-alias
describes; stdlib: checkpoint survival incl. tool-arg boundary pinning, stdlib
`instanceof`, regex-message assertions; codemode: NaN/Infinity + nested/cyclic console
rendering, table cells, caught-tool-failure `instanceof`); adapter suites unchanged
(34 + 16, green); both packages `tsgo --noEmit` clean.
- **`instanceof` + real Error values**: the `errorConstructors` names (`Error`,
`TypeError`, `RangeError`, `SyntaxError`, `ReferenceError`, `EvalError`, `URIError`) are
bound globals (`ErrorConstructorReference`, callable with or without `new`; `typeof` ->
`"function"`). Error values stay the same plain `{ name, message }` null-prototype
objects as before - the constructor name additionally rides on a NON-ENUMERABLE symbol
key (`ErrorBrand`), which every `Object.entries`-based walk (copyIn/copyOut, spread,
JSON.stringify) is blind to, so serialization is byte-identical to the old shape and the
brand is lost on spread/boundary copies exactly like JS loses the prototype.
`caughtErrorValue` produces `{ name, message }` wrappers via `createErrorValue`, so
caught interpreter AND tool failures are `instanceof Error` and carry the `name` the
equivalent real-JS failure would have (follow-up fix, user-directed - "closest to real
JS"): `InterpreterRuntimeError` gained an `errorName` field ("Error" default) set
fluently at throw sites via `.as(name)` - `JSON.parse` failures are `"SyntaxError"` (and
now include the engine's position detail in the message; safe - derived from the
program-supplied string), invalid regex patterns/flags `"SyntaxError"`, unknown
identifiers and TDZ access `"ReferenceError"`, assignment to a constant `"TypeError"`,
a bad `normalize` form `"RangeError"`; a host Error reaching the catch path directly
keeps its own name when it is one of the standard seven. Tool failures and everything
without a specific analogue stay `"Error"` - internal class names never leak. Specific
names satisfy the specific `instanceof` (`e instanceof SyntaxError`), matching JS.
The operator is handled in `evaluateBinaryExpression`
BEFORE the data-only operand check (like `typeof`, it observes any lhs - promises and
functions included); recognized rhs: the error constructors (a specific type matches its
own brand or `Error`, never a sibling), `Date`/`RegExp`/`Map`/`Set` (sandbox classes),
`Array`, `Object` (any object/function-ish value), `Promise` (`SandboxPromise`), and
`Number`/`String`/`Boolean` (always false - no boxed values exist); anything else is a
catchable error naming the recognized constructors.
- **Array methods**: `splice` (mutating, returns the removed elements; insertions run
`rejectCircularInsertion` like push/unshift; one-arg form removes to the end, undefined
delete count removes nothing), `fill` (circular-checked value) and `copyWithin`
(host-delegated), and `keys`/`values`/`entries` returning **arrays** (the Map/Set
convention - for...of and spread work either way). The `retryableArrayMethods`
"rewrite using map/filter" hint set emptied out and was deleted with its branch; unknown
array properties still read `undefined`.
- **String methods**: `localeCompare(that)` (locale/options arguments ignored - host
default locale; the dominant use is a sort comparator), `normalize(form?)` (invalid form
-> catchable error naming the four valid forms), `trimLeft`/`trimRight` as
trimStart/trimEnd aliases.
- **Actionable regex failures**: `toHostRegex` and `constructRegExp` now show the
offending pattern (or flags) plus the engine reason (deduped "Invalid regular
expression:" prefix via `regexFailureReason`) and a shared escaping hint
(`escapeRegexHint`); flags failures list the valid flag letters; the
replaceAll/matchAll missing-`g` errors spell out the exact `/pattern/g` to write and
the single-match alternative.
- **copyIn split (the important one)**: `copyIn(value, label, preserveSandboxValues =
false)` - recursion moved to a private `copyBounded`; `boundedData` (every intra-sandbox
checkpoint: `Object.*` helpers, coercion/Array.from/join inputs, template
interpolation, expression-result checkpoints) is now `copyIn(value, label, true)`,
which passes `SandboxDate`/`SandboxRegExp`/`SandboxMap`/`SandboxSet` through **by
reference as leaves** (contents not walked - Map/Set members are validated at their
mutation sites) while keeping the depth (`MAX_VALUE_DEPTH`), circularity,
plain-objects-only, blocked-property, and data-only checks; un-awaited promises keep
the await-hinting rejection in BOTH modes (deliberate - JS-parity pass-through was
considered and skipped to preserve the nudge). The HOST boundary (final result,
tool-call arguments, `JSON.stringify`, tool-result intake) uses the default mode and
still serializes JSON forms (Date -> ISO, RegExp/Map/Set -> `{}`); host instances met on
the preserving path are defensively wrapped into sandbox equivalents. Ripple: the
`Object.*` helpers treat sandbox values as empty objects (`Object.keys(map)` -> `[]`,
assign sources contribute nothing, hasOwn -> false - JS has no own enumerable props
there), so interpreter internals (`.map`/`.time`/`.regex`) can never leak; the
template-literal sandbox carve-out collapsed into `boundedData`. Object/array spread
already preserved instances (reference copies, no checkpoint) - now tested.
- **Console formatting**: `formatConsoleArgument` is total and deep
(`formatConsoleValue`): numbers render via `String` (`NaN`/`Infinity`/`-Infinity`
literally - never the JSON `null`; finite numbers match their JSON form), nested
strings are JSON-quoted, sandbox values keep their friendly forms at ANY depth (ISO
date, `/regex/flags`, `Map(n) [...]`, `Set(n) [...]`), opaque references become
in-place `[CodeMode reference]` markers instead of collapsing the whole argument,
cycles render `[Circular]` (reachable via Map/Set members, which mutation never
checkpoints), and depth beyond `MAX_CONSOLE_DEPTH = 32` (fixed constant, not a knob)
degrades to `...` - console can no longer fail a program. `console.table` guards with
`containsOpaqueReference` (sandbox cells render, e.g. ISO dates) and its row/cell
walkers treat sandbox values as scalar cells.
- **Prose**: the instructions Syntax not-supported line dropped its `instanceof
Error`/splice mentions (nothing else reworded); README updated (checkpoint
preservation vs boundary serialization, error values/`instanceof`, new array/string
methods, regex-failure behavior); `supportedSyntaxMessage` left untouched (it lists
supported syntax, was already non-exhaustive, and stays accurate).
- **Tests**: package suite 169 -> 209 (parity: Error/instanceof + real-JS error-name
coverage, splice/fill/copyWithin/keys/values/entries, localeCompare/normalize/trim-alias
describes; stdlib: checkpoint survival incl. tool-arg boundary pinning, stdlib
`instanceof`, regex-message assertions; codemode: NaN/Infinity + nested/cyclic console
rendering, table cells, caught-tool-failure `instanceof`); adapter suites unchanged
(34 + 16, green); both packages `tsgo --noEmit` clean.
**Truncation layering - CodeMode truncation off in OpenCode** (user direction; resolves the
section 4 outer-truncation item the OPPOSITE way from "kill the outer one"):
- `maxOutputBytes` lost its 32,000 default and now behaves exactly like the other two
limits: absent = no truncation. All three limits are uniformly no-default - budgets are
host policy. `ResolvedExecutionLimits.maxOutputBytes` is `number | undefined`;
`boundOutput` only runs when the host set the limit. Explicit values validate as before
(safe integer >= 0).
- OpenCode continues to pass NO limits, which now also means no CodeMode truncation.
`execute` is a normal `Tool.define` tool, so OpenCode's native tool-output truncation
applies with no special-casing - verified by tracing `wrap()` (`tool.ts:130-144`,
50KB/2000-line thresholds in `truncate.ts`, full output dumped to a file under
`tool-output/`): the `metadata.truncated` self-truncation exemption never fires for
`execute` (its metadata never sets that key). One truncation layer, the host's - and it
is the richer one (file dump + explore/grep hint vs an inline marker).
- Hosts without their own output bounding set `maxOutputBytes` explicitly; README table
and prose updated, adapter comment rewritten. Tests: codemode +1 (absent limit -> 100KB
value + 50KB log line pass through unbounded, `truncated` undefined); the adapter test
that relied on the old default now asserts the oversized result reaches the shared
wrapper un-truncated. Suites: 210 + 50, tsgo clean both.
- `maxOutputBytes` lost its 32,000 default and now behaves exactly like the other two
limits: absent = no truncation. All three limits are uniformly no-default - budgets are
host policy. `ResolvedExecutionLimits.maxOutputBytes` is `number | undefined`;
`boundOutput` only runs when the host set the limit. Explicit values validate as before
(safe integer >= 0).
- OpenCode continues to pass NO limits, which now also means no CodeMode truncation.
`execute` is a normal `Tool.define` tool, so OpenCode's native tool-output truncation
applies with no special-casing - verified by tracing `wrap()` (`tool.ts:130-144`,
50KB/2000-line thresholds in `truncate.ts`, full output dumped to a file under
`tool-output/`): the `metadata.truncated` self-truncation exemption never fires for
`execute` (its metadata never sets that key). One truncation layer, the host's - and it
is the richer one (file dump + explore/grep hint vs an inline marker).
- Hosts without their own output bounding set `maxOutputBytes` explicitly; README table
and prose updated, adapter comment rewritten. Tests: codemode +1 (absent limit -> 100KB
value + 50KB log line pass through unbounded, `truncated` undefined); the adapter test
that relied on the old default now asserts the oversized result reaches the shared
wrapper un-truncated. Suites: 210 + 50, tsgo clean both.
**Docs polish** (post-API-review): stale `DiscoveryOptions` JSDoc fixed (claimed default
4,000 and alphabetical cheapest-first - now 2,000 and round-robin, matching Fix 8/9 reality)
@@ -837,132 +856,137 @@ regular dependency; hosts depend on it themselves because the API surface is Eff
**Registry promotion + permission-aware catalog** (the "promote to a proper tool service"
restructure; fixes the section 4 permission-advertising bug):
- **The adapter moved** `src/session/code-mode.ts` -> `src/tool/code-mode.ts` and is now a
registry-resident tool service on the TaskTool precedent: `CodeModeTool =
Tool.define(CODE_MODE_TOOL, ...)` whose init depends on `MCP.Service`, `Agent.Service`,
and `Session.Service`. It is yielded in `ToolRegistry.layer`, gated into `builtin` by
`flags.experimentalCodeMode` (like the lsp/plan experiments), and `MCP.node` joined the
registry's `node.deps` (`MCP.node` has no ToolRegistry dependency, so no cycle). The
session-level special-casing in `session/tools.ts` (ad-hoc `SessionCodeMode.define` +
append) is deleted; the early return that suppresses raw per-MCP registration when the
flag is on stays session-side, keyed on the same flag+tool-count condition.
- **Enablement** lives in `ToolRegistry.tools()` next to the WebSearchTool check: the MCP
tool count is consulted once (an Effect) before the synchronous filter, and code mode
passes the predicate iff `flags.experimentalCodeMode` && count > 0.
- **Description split on the `describeTask` precedent**: the tool's static base
description is a two-line summary; `describeCodeMode(agent)` in `registry.tools()`
appends the full CodeMode instructions (workflow/rules/syntax + grouped catalog,
`catalogInstructions` in the adapter) at the same composition point as task - so
`plugin.trigger("tool.definition")` sees the base description first.
- **Permission-aware catalog + dispatch** (the bug fix): the visibility predicate from
`llm/request.ts` `resolveTools` is hoisted to `Permission.visibleTools(tools, ruleset)`
(a record filter over `Permission.disabled` - only a hard `deny` with pattern `"*"`
hides a tool; ask-level rules stay fully visible and prompt at call time) and
`resolveTools` now uses it, so the two paths cannot drift. `describeCodeMode` filters
with the merged agent+session ruleset that `SessionTools.resolve` passes into the
registry before building the catalog/search index; `execute` rebuilds the runtime per
execution from a fresh, filtered `mcp.tools()` snapshot using the same merged ruleset
(`Agent.get(ctx.agent)` + `Session.get(ctx.sessionID)`, matching the merge
`SessionTools.context` wires into `ctx.ask`) - a denied tool is not dispatchable
even if the model guesses its name and yields the normal unknown-tool diagnostic.
Documented gap (out of scope by design): per-message `user.tools[key] === false` arrives
at request-prep after descriptions are built and has no child-call equivalent.
- **Preserved behavior**: cancellation race + pre-aborted-signal guard, `toSandboxResult`
unwrap order, attachment accumulation, `CODE_MODE_TOOL` at all title sites, no execution
limits (native truncation only), `displayInput`, per-child `ctx.ask` gating (now wired
through `Tool.Context` exactly like every registry tool).
- **Explicit non-goal**: memoizing the catalog builder keyed on (ToolsChanged generation,
permission ruleset) was considered and deliberately skipped - the per-turn rebuild is
cheap (grouping + string rendering); revisit only if profiling shows it matters.
- **Tests**: the two adapter suites moved to `test/tool/{code-mode,code-mode-integration}
.test.ts` (mocked `MCP.Service`/`Agent.Service`/`Session.Service` replacing the direct
`define(...)` construction; description assertions target `catalogInstructions`, the
registry's composition input) and gained permission coverage: deny excluded from
catalog/search, ask-level stays visible and callable, denied tool undispatchable
(unknown-tool diagnostic), `Permission.visibleTools` semantics. `test/tool/
registry.test.ts` gained four registry-level tests: registered with flag+MCP tools,
excluded without MCP tools, excluded with flag off, and deny/ask catalog filtering
through `registry.tools()`. Suites: 43 + 16 adapter tests, 16 registry tests, all green.
- **The adapter moved** `src/session/code-mode.ts` -> `src/tool/code-mode.ts` and is now a
registry-resident tool service on the TaskTool precedent: `CodeModeTool =
Tool.define(CODE_MODE_TOOL, ...)` whose init depends on `MCP.Service`, `Agent.Service`,
and `Session.Service`. It is yielded in `ToolRegistry.layer`, gated into `builtin` by
`flags.experimentalCodeMode` (like the lsp/plan experiments), and `MCP.node` joined the
registry's `node.deps` (`MCP.node` has no ToolRegistry dependency, so no cycle). The
session-level special-casing in `session/tools.ts` (ad-hoc `SessionCodeMode.define` +
append) is deleted; the early return that suppresses raw per-MCP registration when the
flag is on stays session-side, keyed on the same flag+tool-count condition.
- **Enablement** lives in `ToolRegistry.tools()` next to the WebSearchTool check: the MCP
tool count is consulted once (an Effect) before the synchronous filter, and code mode
passes the predicate iff `flags.experimentalCodeMode` && count > 0.
- **Description split on the `describeTask` precedent**: the tool's static base
description is a two-line summary; `describeCodeMode(agent)` in `registry.tools()`
appends the full CodeMode instructions (workflow/rules/syntax + grouped catalog,
`catalogInstructions` in the adapter) at the same composition point as task - so
`plugin.trigger("tool.definition")` sees the base description first.
- **Permission-aware catalog + dispatch** (the bug fix): the visibility predicate from
`llm/request.ts` `resolveTools` is hoisted to `Permission.visibleTools(tools, ruleset)`
(a record filter over `Permission.disabled` - only a hard `deny` with pattern `"*"`
hides a tool; ask-level rules stay fully visible and prompt at call time) and
`resolveTools` now uses it, so the two paths cannot drift. `describeCodeMode` filters
with the merged agent+session ruleset that `SessionTools.resolve` passes into the
registry before building the catalog/search index; `execute` rebuilds the runtime per
execution from a fresh, filtered `mcp.tools()` snapshot using the same merged ruleset
(`Agent.get(ctx.agent)` + `Session.get(ctx.sessionID)`, matching the merge
`SessionTools.context` wires into `ctx.ask`) - a denied tool is not dispatchable
even if the model guesses its name and yields the normal unknown-tool diagnostic.
Documented gap (out of scope by design): per-message `user.tools[key] === false` arrives
at request-prep after descriptions are built and has no child-call equivalent.
- **Preserved behavior**: cancellation race + pre-aborted-signal guard, `toSandboxResult`
unwrap order, attachment accumulation, `CODE_MODE_TOOL` at all title sites, no execution
limits (native truncation only), `displayInput`, per-child `ctx.ask` gating (now wired
through `Tool.Context` exactly like every registry tool).
- **Explicit non-goal**: memoizing the catalog builder keyed on (ToolsChanged generation,
permission ruleset) was considered and deliberately skipped - the per-turn rebuild is
cheap (grouping + string rendering); revisit only if profiling shows it matters.
- **Tests**: the two adapter suites moved to `test/tool/{code-mode,code-mode-integration}
.test.ts` (mocked `MCP.Service`/`Agent.Service`/`Session.Service` replacing the direct
`define(...)` construction; description assertions target `catalogInstructions`, the
registry's composition input) and gained permission coverage: deny excluded from
catalog/search, ask-level stays visible and callable, denied tool undispatchable
(unknown-tool diagnostic), `Permission.visibleTools` semantics. `test/tool/
registry.test.ts` gained four registry-level tests: registered with flag+MCP tools,
excluded without MCP tools, excluded with flag off, and deny/ask catalog filtering
through `registry.tools()`. Suites: 43 + 16 adapter tests, 16 registry tests, all green.
**Shared MCP invocation middle (`McpInvoke.invoke`)** (closes the section 4 "plugin hooks skip
child calls" gap):
- `packages/opencode/src/mcp/invoke.ts` extracts the duplicated "invoke an MCP tool"
middle into one shared `McpInvoke.invoke(input)`: plugin `tool.execute.before` hook ->
permission ask (`{ permission: key, patterns: ["*"], always: ["*"] }` via the caller's
`ctx.ask`) -> dispatch through the ai-sdk tool's execute inside the `Tool.execute`
tracing span (`tool.name`/`tool.call_id`/`session.id`/`message.id` attributes) ->
plugin `tool.execute.after` hook. It returns the RAW result the ai-sdk execute
resolved with; each caller keeps its own shaping edge - the legacy per-MCP loop in
`SessionTools.resolve` applies its existing model-facing shaping/truncation, code
mode applies `toSandboxResult`. It lives under `src/mcp/` because both callers
already depend on MCP and the function is about invoking an MCP-backed ai-sdk tool,
not about sessions or code mode.
- **After-hook payload**: fired inside `McpInvoke.invoke` with the raw MCP result -
which is exactly what the legacy loop always passed (the raw `CallToolResult`, not
the shaped `{title, output, metadata}`), so legacy behavior is preserved bit-for-bit
and the hook payload cannot drift between callers. No callback/edge-firing design
was needed.
- **Synthetic child callID**: code-mode child calls pass `${parentCallID}/${n}` as the
hook/span callID (`parentCallID` = the `execute` call's `ctx.callID`, falling back to
the entry key; `n` = per-execution counter starting at 1, shared across all child
calls in one program). callID is an opaque string - nothing parses it. The ai-sdk
`toolCallId` (`options.toolCallId`) stays each caller's existing value
(`ctx.callID ?? entry.key` for code mode).
- **Child-scoped hook failures**: `CodeModeTool` (which now also yields
`Plugin.Service`) wraps the whole child call - hooks, ask, dispatch - in
`toCatchable` (the generalization of the old `askPermission` catchCause), so a plugin
hook failure fails ONLY that child call as a catchable in-program `toolError`; other
calls in the same program keep running and interruption still propagates as
interruption. Legacy semantics unchanged: a hook failure fails the tool call.
- **Tests**: `test/tool/code-mode.test.ts` +2 (child calls fire before/after with the
MCP key and `parent/1`, `parent/2` ids, after hook carries the raw MCP result; a
failing before hook is caught in-program, gates dispatch, and leaves the outer
execute ok) - both code-mode harnesses gained a `Plugin.Service` mock (pass-through
trigger by default, overridable). New `test/session/tools.test.ts` (3 tests) pins
`SessionTools.resolve` at the real-registry seam (LayerNode.compile, fake MCP layer):
flag on + MCP tools -> `execute` present, raw MCP keys suppressed; flag off -> raw
keys present, `execute` absent; and the legacy raw-MCP execute fires before/after
hooks keyed by the ai-sdk toolCallId with the raw result payload. Suites: adapter
45 + 16, session/tool/permission all green; this package untouched (211 pass).
- `packages/opencode/src/mcp/invoke.ts` extracts the duplicated "invoke an MCP tool"
middle into one shared `McpInvoke.invoke(input)`: plugin `tool.execute.before` hook ->
permission ask (`{ permission: key, patterns: ["*"], always: ["*"] }` via the caller's
`ctx.ask`) -> dispatch through the ai-sdk tool's execute inside the `Tool.execute`
tracing span (`tool.name`/`tool.call_id`/`session.id`/`message.id` attributes) ->
plugin `tool.execute.after` hook. It returns the RAW result the ai-sdk execute
resolved with; each caller keeps its own shaping edge - the legacy per-MCP loop in
`SessionTools.resolve` applies its existing model-facing shaping/truncation, code
mode applies `toSandboxResult`. It lives under `src/mcp/` because both callers
already depend on MCP and the function is about invoking an MCP-backed ai-sdk tool,
not about sessions or code mode.
- **After-hook payload**: fired inside `McpInvoke.invoke` with the raw MCP result -
which is exactly what the legacy loop always passed (the raw `CallToolResult`, not
the shaped `{title, output, metadata}`), so legacy behavior is preserved bit-for-bit
and the hook payload cannot drift between callers. No callback/edge-firing design
was needed.
- **Synthetic child callID**: code-mode child calls pass `${parentCallID}/${n}` as the
hook/span callID (`parentCallID` = the `execute` call's `ctx.callID`, falling back to
the entry key; `n` = per-execution counter starting at 1, shared across all child
calls in one program). callID is an opaque string - nothing parses it. The ai-sdk
`toolCallId` (`options.toolCallId`) stays each caller's existing value
(`ctx.callID ?? entry.key` for code mode).
- **Child-scoped hook failures**: `CodeModeTool` (which now also yields
`Plugin.Service`) wraps the whole child call - hooks, ask, dispatch - in
`toCatchable` (the generalization of the old `askPermission` catchCause), so a plugin
hook failure fails ONLY that child call as a catchable in-program `toolError`; other
calls in the same program keep running and interruption still propagates as
interruption. Legacy semantics unchanged: a hook failure fails the tool call.
- **Tests**: `test/tool/code-mode.test.ts` +2 (child calls fire before/after with the
MCP key and `parent/1`, `parent/2` ids, after hook carries the raw MCP result; a
failing before hook is caught in-program, gates dispatch, and leaves the outer
execute ok) - both code-mode harnesses gained a `Plugin.Service` mock (pass-through
trigger by default, overridable). New `test/session/tools.test.ts` (3 tests) pins
`SessionTools.resolve` at the real-registry seam (LayerNode.compile, fake MCP layer):
flag on + MCP tools -> `execute` present, raw MCP keys suppressed; flag off -> raw
keys present, `execute` absent; and the legacy raw-MCP execute fires before/after
hooks keyed by the ai-sdk toolCallId with the raw result payload. Suites: adapter
45 + 16, session/tool/permission all green; this package untouched (211 pass).
**Signature rendering + compound-assignment parity fixes** (externally reported, both
verified real with failing tests before fixing):
- **Non-identifier property names in rendered signatures** (`src/tool.ts`): `renderSchema`
emitted raw property names, so schema properties like `foo-bar`/`@type`/`x.y`/`123`
rendered invalid TypeScript (`{ foo-bar?: string }`). Fixed with a `renderKey` helper -
bare identifiers stay bare, everything else is `JSON.stringify`-quoted - applied in the
single `field` closure both the compact and pretty renderings share. The
`identifierSegment` regex now lives in `tool.ts` (exported) and `tool-runtime.ts`'s
bracket-notation `toolExpression` imports it: one source of truth for "is this a bare
identifier" across object keys and tool paths. Tests: `signature.test.ts` +4 (compact,
pretty with JSDoc on a quoted key, JSON Schema input+output, Effect Schema struct).
- **Numeric schema unions keep their real alternatives** (`src/tool.ts`): the old
`anyOf`/`oneOf` renderer collapsed any union containing `{ type: "number" }` to just
`number`, dropping real JSON Schema alternatives (`string | number`, `number | null`,
etc.). The collapse is now restricted to Effect's number-schema artifact
(`number | "NaN" | "Infinity" | "-Infinity"`, emitted as single-value string enums),
while raw JSON Schema unions render every branch. Tests: `signature.test.ts` +3.
- **Compound assignment now matches binary-operator semantics** (`src/codemode.ts`):
`applyCompoundAssignment` did raw JS ops on interpreter wrapper objects, so `x += y`
diverged from `x = x + y` (sandbox Date `d += 1` produced `"[object Object]1"`;
`d -= 400` gave `NaN` instead of epoch arithmetic). The operator table + coercion moved
verbatim out of `evaluateBinaryExpression` into a shared `applyBinaryOperator`;
compound assignment validates against a `compoundOperators` set (`+=` ... `>>>=`) and
dispatches through it (`operator.slice(0, -1)`). Logical assignments (`&&=`/`||=`/`??=`)
keep their separate short-circuit path (`evaluateLogicalAssignment`), and both
assignment call sites still wrap results in `boundedData`. Deliberate side effect:
compound assignment now rejects opaque references, consistent with binary operators.
Tests: `parity.test.ts` +5 (Date `+=` concat parity, Date `-=`/`/=` epoch parity,
string `+=` object/array, member-target compound, 13-case operator sweep vs real JS).
Package suite: 220 pass.
- **Non-identifier property names in rendered signatures** (`src/tool.ts`): `renderSchema`
emitted raw property names, so schema properties like `foo-bar`/`@type`/`x.y`/`123`
rendered invalid TypeScript (`{ foo-bar?: string }`). Fixed with a `renderKey` helper -
bare identifiers stay bare, everything else is `JSON.stringify`-quoted - applied in the
single `field` closure both the compact and pretty renderings share. The
`identifierSegment` regex now lives in `tool.ts` (exported) and `tool-runtime.ts`'s
bracket-notation `toolExpression` imports it: one source of truth for "is this a bare
identifier" across object keys and tool paths. Tests: `signature.test.ts` +4 (compact,
pretty with JSDoc on a quoted key, JSON Schema input+output, Effect Schema struct).
- **Numeric schema unions keep their real alternatives** (`src/tool.ts`): the old
`anyOf`/`oneOf` renderer collapsed any union containing `{ type: "number" }` to just
`number`, dropping real JSON Schema alternatives (`string | number`, `number | null`,
etc.). The collapse is now restricted to Effect's number-schema artifact
(`number | "NaN" | "Infinity" | "-Infinity"`, emitted as single-value string enums),
while raw JSON Schema unions render every branch. Tests: `signature.test.ts` +3.
- **Compound assignment now matches binary-operator semantics** (`src/codemode.ts`):
`applyCompoundAssignment` did raw JS ops on interpreter wrapper objects, so `x += y`
diverged from `x = x + y` (sandbox Date `d += 1` produced `"[object Object]1"`;
`d -= 400` gave `NaN` instead of epoch arithmetic). The operator table + coercion moved
verbatim out of `evaluateBinaryExpression` into a shared `applyBinaryOperator`;
compound assignment validates against a `compoundOperators` set (`+=` ... `>>>=`) and
dispatches through it (`operator.slice(0, -1)`). Logical assignments (`&&=`/`||=`/`??=`)
keep their separate short-circuit path (`evaluateLogicalAssignment`), and both
assignment call sites still wrap results in `boundedData`. Deliberate side effect:
compound assignment now rejects opaque references, consistent with binary operators.
Tests: `parity.test.ts` +5 (Date `+=` concat parity, Date `-=`/`/=` epoch parity,
string `+=` object/array, member-target compound, 13-case operator sweep vs real JS).
Package suite: 220 pass.
---
## 4. Remaining work (detailed TODO)
### Next DSL-expansion pass (done - see the DSL-expansion pass entry in section 3)
Batch these together - per user direction: important, but deliberately deferred to one
focused interpreter-surface pass rather than picked off piecemeal.
- [x] Medium-tier JS parity items deferred from the original audit: caught errors are plain
`{ name, message }` objects, not `instanceof Error` (and `Error` isn't a value -
`x instanceof Error` is unsupported syntax); `splice` (still a
@@ -981,6 +1005,7 @@ focused interpreter-surface pass rather than picked off piecemeal.
(`console.log({ m: map })`) - could deep-format instead.
### Next iteration: text-result handling (deliberate follow-up, user-directed)
- [ ] Revisit how MCP text results reach the program. Today: `structuredContent` when the
server sends it, else joined text as a plain string (the program JSON.parses it,
guided by a workflow step). Considered and deferred: (a) conservative boundary
@@ -992,6 +1017,7 @@ focused interpreter-surface pass rather than picked off piecemeal.
revisit once real usage shows which failure modes matter.
### Next iteration: stdlib surface (prioritized)
Current instructions say "usual Array/String/Object/Math/JSON methods," but the interpreter is
intentionally a subset. Keep CodeMode focused on orchestration and data shaping, not a full host
runtime, but close the high-friction gaps models are likely to reach for.
@@ -1028,7 +1054,9 @@ Explicit non-goals for now: `structuredClone`, `WeakMap`/`WeakSet`, and timers
orchestration use case.
### Wiring-review findings (subagent code review of the OpenCode integration, triaged)
Pre-PR fixes (user-approved cut):
- [x] **Cancellation does not interrupt the interpreter** - the no-limits rationale claimed
"user cancel interrupts the execution fiber," but `tools.ts` runs tools via
`run.promise` -> `Effect.runPromise` (`effect/bridge.ts:64-66`) with NO abort wiring;
@@ -1073,6 +1101,7 @@ Pre-PR fixes (user-approved cut):
`title: "execute"` sites in `code-mode.ts` now reference `CODE_MODE_TOOL`.
Post-MVP (logged, not blocking an experimental flag):
- [x] **Plugin `tool.execute.before/after` hooks skip child calls** - legacy MCP
registration fires them per tool (`tools.ts:419-441`); under code mode only the
outer `execute` fires them, so auditing/intercepting plugins silently lose MCP
@@ -1105,6 +1134,7 @@ Post-MVP (logged, not blocking an experimental flag):
names that are no longer directly callable under code mode.
### Backlog / loose ends (non-blocking, any order)
- [ ] `evaluateUpdateExpression` (`++`/`--`) still uses raw `Number(current)`, so `d++` on a
sandbox Date yields `NaN` where `d += 1` now uses epoch semantics (and real JS `d++`
would give epoch+0 numeric). Pre-existing, out of scope of the compound-assignment
@@ -1141,7 +1171,7 @@ Post-MVP (logged, not blocking an experimental flag):
## 5. Context and gotchas for whoever picks this up
- **Motivating failure (why forgiving semantics + prompting matter):** in a real transcript,
the model wrote `me.result?.login ?? me.result` where the tool result was a JSON *string* -
the model wrote `me.result?.login ?? me.result` where the tool result was a JSON _string_ -
the old strict interpreter threw (`String property 'login' is not available`); then the
model returned a raw 105KB payload, which native truncation dumped to a file, costing a
subagent round-trip to extract one number. Interpreter forgiveness stops the crashes;
+889 -348
View File
@@ -118,9 +118,16 @@ export type CodeModeOptions<Tools extends Record<string, unknown> = {}> = Omit<E
export const ExecuteInputSchema = Schema.Struct({ code: Schema.String })
const DiagnosticKindSchema = Schema.Literals([
"ParseError", "UnsupportedSyntax", "UnknownTool", "InvalidToolInput", "InvalidToolOutput", "InvalidDataValue",
"ToolCallLimitExceeded", "TimeoutExceeded",
"ToolFailure", "ExecutionFailure",
"ParseError",
"UnsupportedSyntax",
"UnknownTool",
"InvalidToolInput",
"InvalidToolOutput",
"InvalidDataValue",
"ToolCallLimitExceeded",
"TimeoutExceeded",
"ToolFailure",
"ExecutionFailure",
])
/** Structured success or diagnostic result schema returned by CodeMode execution. */
@@ -245,7 +252,10 @@ class GlobalNamespace {
}
class GlobalMethodReference {
constructor(readonly namespace: GlobalNamespaceName | "Number" | "String", readonly name: string) {}
constructor(
readonly namespace: GlobalNamespaceName | "Number" | "String",
readonly name: string,
) {}
}
class CoercionFunction {
@@ -290,10 +300,41 @@ export type DiagnosticKind =
| "ExecutionFailure"
const arrayMethods = new Set([
"map", "filter", "find", "findIndex", "findLast", "findLastIndex", "some", "every", "includes", "join",
"reduce", "reduceRight", "flatMap", "forEach", "sort", "toSorted", "slice", "concat", "indexOf", "lastIndexOf",
"at", "flat", "reverse", "toReversed", "with", "push", "pop", "shift", "unshift",
"splice", "fill", "copyWithin", "keys", "values", "entries",
"map",
"filter",
"find",
"findIndex",
"findLast",
"findLastIndex",
"some",
"every",
"includes",
"join",
"reduce",
"reduceRight",
"flatMap",
"forEach",
"sort",
"toSorted",
"slice",
"concat",
"indexOf",
"lastIndexOf",
"at",
"flat",
"reverse",
"toReversed",
"with",
"push",
"pop",
"shift",
"unshift",
"splice",
"fill",
"copyWithin",
"keys",
"values",
"entries",
])
const mathConstants = new Set(["PI", "E", "LN2", "LN10", "LOG2E", "LOG10E", "SQRT2", "SQRT1_2"])
@@ -301,10 +342,38 @@ const mathConstants = new Set(["PI", "E", "LN2", "LN10", "LOG2E", "LOG10E", "SQR
const numberMethods = new Set(["toFixed", "toPrecision", "toExponential", "toString"])
const stringMethods = new Set([
"toLowerCase", "toUpperCase", "trim", "trimStart", "trimEnd", "trimLeft", "trimRight", "split", "slice",
"substring", "substr", "includes", "startsWith", "endsWith", "indexOf", "lastIndexOf", "replace", "replaceAll",
"repeat", "padStart", "padEnd", "charAt", "charCodeAt", "codePointAt", "at", "concat", "toString",
"match", "matchAll", "search", "localeCompare", "normalize",
"toLowerCase",
"toUpperCase",
"trim",
"trimStart",
"trimEnd",
"trimLeft",
"trimRight",
"split",
"slice",
"substring",
"substr",
"includes",
"startsWith",
"endsWith",
"indexOf",
"lastIndexOf",
"replace",
"replaceAll",
"repeat",
"padStart",
"padEnd",
"charAt",
"charCodeAt",
"codePointAt",
"at",
"concat",
"toString",
"match",
"matchAll",
"search",
"localeCompare",
"normalize",
])
const numberConstants = new Set(["MAX_SAFE_INTEGER", "MIN_SAFE_INTEGER", "MAX_VALUE", "MIN_VALUE", "EPSILON"])
@@ -317,21 +386,57 @@ const consoleMethods = new Set(["log", "info", "debug", "warn", "error", "dir",
const promiseStatics = new Set<PromiseMethodName>(["all", "allSettled", "race", "resolve", "reject"])
const errorConstructors = new Set(["Error", "TypeError", "RangeError", "SyntaxError", "ReferenceError", "EvalError", "URIError"])
const errorConstructors = new Set([
"Error",
"TypeError",
"RangeError",
"SyntaxError",
"ReferenceError",
"EvalError",
"URIError",
])
const valueConstructors = new Set(["Date", "RegExp", "Map", "Set"])
const dateMethods = new Set([
"getTime", "valueOf", "toISOString", "toJSON", "toString",
"getFullYear", "getMonth", "getDate", "getDay", "getHours", "getMinutes", "getSeconds", "getMilliseconds",
"getUTCFullYear", "getUTCMonth", "getUTCDate", "getUTCDay", "getUTCHours", "getUTCMinutes", "getUTCSeconds", "getUTCMilliseconds",
"getTime",
"valueOf",
"toISOString",
"toJSON",
"toString",
"getFullYear",
"getMonth",
"getDate",
"getDay",
"getHours",
"getMinutes",
"getSeconds",
"getMilliseconds",
"getUTCFullYear",
"getUTCMonth",
"getUTCDate",
"getUTCDay",
"getUTCHours",
"getUTCMinutes",
"getUTCSeconds",
"getUTCMilliseconds",
"getTimezoneOffset",
])
const dateStatics = new Set(["now", "parse", "UTC"])
const regexpMethods = new Set(["test", "exec", "toString"])
// Read-only host regex fields surfaced as plain values.
const regexpProperties = new Set(["source", "flags", "lastIndex", "global", "ignoreCase", "multiline", "sticky", "unicode", "dotAll"])
const regexpProperties = new Set([
"source",
"flags",
"lastIndex",
"global",
"ignoreCase",
"multiline",
"sticky",
"unicode",
"dotAll",
])
const mapMethods = new Set(["get", "set", "has", "delete", "clear", "forEach", "keys", "values", "entries"])
const setMethods = new Set(["add", "has", "delete", "clear", "forEach", "keys", "values", "entries"])
@@ -342,7 +447,12 @@ const supportedSyntaxMessage =
"Supported orchestration syntax: tools.* calls (they return promises - resolve them with await), data literals, destructuring, optional chaining, template literals, conditionals, switch, loops (incl. for...of and for...in over object/array/tools keys), arrow functions, spread, try/catch, array methods (map/filter/find/findIndex/some/every/reduce/flatMap/forEach/sort/slice/concat/indexOf/lastIndexOf/at/flat/reverse/includes/join), string methods (incl. match/matchAll/replace/split with regular expressions), Date/RegExp/Map/Set, Object/Math/JSON helpers, captured console.log/warn/error/dir/table, and Promise.all/allSettled/race/resolve/reject over arrays mixing promises and plain values for parallel tool calls (promise chaining with .then/.catch is not supported - use await with try/catch)."
const unsupportedSyntax = (kind: string, node: AstNode): InterpreterRuntimeError =>
new InterpreterRuntimeError(`Syntax '${kind}' is not supported in CodeMode. ${supportedSyntaxMessage}`, node, "UnsupportedSyntax", [supportedSyntaxMessage])
new InterpreterRuntimeError(
`Syntax '${kind}' is not supported in CodeMode. ${supportedSyntaxMessage}`,
node,
"UnsupportedSyntax",
[supportedSyntaxMessage],
)
/** How many eagerly forked tool calls may run at once. Fixed; not a configurable knob. */
const TOOL_CALL_CONCURRENCY = 8
@@ -350,7 +460,11 @@ const TOOL_CALL_CONCURRENCY = 8
/** Console formatting recursion ceiling; deeper values render as "...". Fixed; not a knob. */
const MAX_CONSOLE_DEPTH = 32
const validateLimit = <Value extends number | undefined>(name: keyof ExecutionLimits, value: Value, minimum: number): Value => {
const validateLimit = <Value extends number | undefined>(
name: keyof ExecutionLimits,
value: Value,
minimum: number,
): Value => {
if (value !== undefined && (!Number.isSafeInteger(value) || value < minimum)) {
throw new RangeError(`${name} must be a safe integer greater than or equal to ${minimum}.`)
}
@@ -377,7 +491,12 @@ class InterpreterRuntimeError extends Error {
*/
errorName: string = "Error"
constructor(message: string, node?: AstNode, readonly kind: DiagnosticKind = "ExecutionFailure", readonly suggestions?: ReadonlyArray<string>) {
constructor(
message: string,
node?: AstNode,
readonly kind: DiagnosticKind = "ExecutionFailure",
readonly suggestions?: ReadonlyArray<string>,
) {
super(message)
this.name = "InterpreterRuntimeError"
@@ -392,8 +511,7 @@ class InterpreterRuntimeError extends Error {
}
}
const isRecord = (value: unknown): value is Record<string, unknown> =>
typeof value === "object" && value !== null
const isRecord = (value: unknown): value is Record<string, unknown> => typeof value === "object" && value !== null
const asNode = (value: unknown, context: string): AstNode => {
if (!isRecord(value) || typeof value.type !== "string") {
@@ -527,7 +645,11 @@ const normalizeError = (error: unknown): Diagnostic => {
message = "a non-data value"
} else if (typeof value === "string") {
message = value
} else if (value !== null && typeof value === "object" && typeof (value as { message?: unknown }).message === "string") {
} else if (
value !== null &&
typeof value === "object" &&
typeof (value as { message?: unknown }).message === "string"
) {
message = (value as { message: string }).message
} else {
try {
@@ -572,10 +694,17 @@ const caughtErrorValue = (thrown: unknown): unknown => {
const boundedData = (value: unknown, label: string): unknown => copyIn(value, label, true)
const isRuntimeReference = (value: unknown): boolean =>
value instanceof CodeModeFunction || value instanceof ToolReference || value instanceof IntrinsicReference ||
value instanceof GlobalNamespace || value instanceof GlobalMethodReference || value instanceof PromiseNamespace ||
value instanceof PromiseMethodReference || value instanceof SandboxPromise || value instanceof CoercionFunction ||
value instanceof ErrorConstructorReference || isSandboxValue(value)
value instanceof CodeModeFunction ||
value instanceof ToolReference ||
value instanceof IntrinsicReference ||
value instanceof GlobalNamespace ||
value instanceof GlobalMethodReference ||
value instanceof PromiseNamespace ||
value instanceof PromiseMethodReference ||
value instanceof SandboxPromise ||
value instanceof CoercionFunction ||
value instanceof ErrorConstructorReference ||
isSandboxValue(value)
const containsRuntimeReference = (value: unknown, seen = new Set<object>()): boolean => {
if (isRuntimeReference(value)) return true
@@ -610,10 +739,15 @@ const containsOpaqueReference = (value: unknown, seen = new Set<object>()): bool
// like a real JS promise.
const typeofValue = (value: unknown): string => {
if (
value instanceof CodeModeFunction || value instanceof CoercionFunction || value instanceof IntrinsicReference ||
value instanceof GlobalMethodReference || value instanceof PromiseMethodReference || value instanceof PromiseNamespace ||
value instanceof CodeModeFunction ||
value instanceof CoercionFunction ||
value instanceof IntrinsicReference ||
value instanceof GlobalMethodReference ||
value instanceof PromiseMethodReference ||
value instanceof PromiseNamespace ||
value instanceof ErrorConstructorReference
) return "function"
)
return "function"
if (value instanceof ToolReference) return value.path.length > 0 ? "function" : "object"
if (value instanceof GlobalNamespace) {
return value.name === "Math" || value.name === "JSON" || value.name === "console" ? "object" : "function"
@@ -632,12 +766,18 @@ const instanceofValue = (lhs: unknown, rhs: unknown, node: AstNode): boolean =>
}
if (rhs instanceof GlobalNamespace) {
switch (rhs.name) {
case "Date": return lhs instanceof SandboxDate
case "RegExp": return lhs instanceof SandboxRegExp
case "Map": return lhs instanceof SandboxMap
case "Set": return lhs instanceof SandboxSet
case "Array": return Array.isArray(lhs)
case "Object": return lhs !== null && (typeof lhs === "object" || typeofValue(lhs) === "function")
case "Date":
return lhs instanceof SandboxDate
case "RegExp":
return lhs instanceof SandboxRegExp
case "Map":
return lhs instanceof SandboxMap
case "Set":
return lhs instanceof SandboxSet
case "Array":
return Array.isArray(lhs)
case "Object":
return lhs !== null && (typeof lhs === "object" || typeofValue(lhs) === "function")
}
}
if (rhs instanceof PromiseNamespace) return lhs instanceof SandboxPromise
@@ -703,12 +843,14 @@ const matchToValue = (match: RegExpMatchArray): Array<unknown> => {
const invokeStringMethod = (value: string, name: string, args: Array<unknown>, node: AstNode): unknown => {
const str = (index: number): string => {
const arg = args[index]
if (typeof arg !== "string") throw new InterpreterRuntimeError(`String.${name} expects argument ${index + 1} to be a string.`, node)
if (typeof arg !== "string")
throw new InterpreterRuntimeError(`String.${name} expects argument ${index + 1} to be a string.`, node)
return arg
}
const num = (index: number): number => {
const arg = args[index]
if (typeof arg !== "number") throw new InterpreterRuntimeError(`String.${name} expects argument ${index + 1} to be a number.`, node)
if (typeof arg !== "number")
throw new InterpreterRuntimeError(`String.${name} expects argument ${index + 1} to be a number.`, node)
return arg
}
const optNum = (index: number): number | undefined => (args[index] === undefined ? undefined : num(index))
@@ -716,15 +858,29 @@ const invokeStringMethod = (value: string, name: string, args: Array<unknown>, n
let result: unknown
switch (name) {
case "toLowerCase": result = value.toLowerCase(); break
case "toUpperCase": result = value.toUpperCase(); break
case "trim": result = value.trim(); break
case "toLowerCase":
result = value.toLowerCase()
break
case "toUpperCase":
result = value.toUpperCase()
break
case "trim":
result = value.trim()
break
// trimLeft/trimRight are the legacy aliases of trimStart/trimEnd, kept because models write them.
case "trimStart": case "trimLeft": result = value.trimStart(); break
case "trimEnd": case "trimRight": result = value.trimEnd(); break
case "trimStart":
case "trimLeft":
result = value.trimStart()
break
case "trimEnd":
case "trimRight":
result = value.trimEnd()
break
// Locale/options arguments are ignored: comparison runs with the host default locale, and
// the common use is a sort comparator where any consistent order works.
case "localeCompare": result = value.localeCompare(str(0)); break
case "localeCompare":
result = value.localeCompare(str(0))
break
case "normalize": {
const form = optStr(0)
try {
@@ -750,16 +906,33 @@ const invokeStringMethod = (value: string, name: string, args: Array<unknown>, n
result = value.split(str(0), requestedLimit === undefined ? undefined : requestedLimit >>> 0)
break
}
case "slice": result = value.slice(optNum(0), optNum(1)); break
case "includes": result = value.includes(str(0), optNum(1)); break
case "startsWith": result = value.startsWith(str(0), optNum(1)); break
case "endsWith": result = value.endsWith(str(0), optNum(1)); break
case "indexOf": result = value.indexOf(str(0), optNum(1)); break
case "lastIndexOf": result = value.lastIndexOf(str(0), optNum(1)); break
case "slice":
result = value.slice(optNum(0), optNum(1))
break
case "includes":
result = value.includes(str(0), optNum(1))
break
case "startsWith":
result = value.startsWith(str(0), optNum(1))
break
case "endsWith":
result = value.endsWith(str(0), optNum(1))
break
case "indexOf":
result = value.indexOf(str(0), optNum(1))
break
case "lastIndexOf":
result = value.lastIndexOf(str(0), optNum(1))
break
case "replace":
case "replaceAll": {
if (args[0] instanceof CodeModeFunction || args[1] instanceof CodeModeFunction) {
throw new InterpreterRuntimeError(`String.${name} does not support function replacers in CodeMode; use match/matchAll and rebuild the string instead.`, node, "UnsupportedSyntax", [supportedSyntaxMessage])
throw new InterpreterRuntimeError(
`String.${name} does not support function replacers in CodeMode; use match/matchAll and rebuild the string instead.`,
node,
"UnsupportedSyntax",
[supportedSyntaxMessage],
)
}
if (args[0] instanceof SandboxRegExp) {
const pattern = (args[0] as SandboxRegExp).regex
@@ -807,26 +980,46 @@ const invokeStringMethod = (value: string, name: string, args: Array<unknown>, n
}
case "repeat": {
const count = num(0)
if (!Number.isFinite(count) || count < 0) throw new InterpreterRuntimeError("String.repeat expects a finite non-negative count.", node)
if (!Number.isFinite(count) || count < 0)
throw new InterpreterRuntimeError("String.repeat expects a finite non-negative count.", node)
result = value.repeat(count)
break
}
case "padStart": result = value.padStart(num(0), optStr(1)); break
case "padEnd": result = value.padEnd(num(0), optStr(1)); break
case "charAt": result = value.charAt(optNum(0) ?? 0); break
case "at": result = value.at(optNum(0) ?? 0); break
case "substring": result = value.substring(optNum(0) ?? 0, optNum(1)); break
case "substr": result = value.substr(optNum(0) ?? 0, optNum(1)); break
case "padStart":
result = value.padStart(num(0), optStr(1))
break
case "padEnd":
result = value.padEnd(num(0), optStr(1))
break
case "charAt":
result = value.charAt(optNum(0) ?? 0)
break
case "at":
result = value.at(optNum(0) ?? 0)
break
case "substring":
result = value.substring(optNum(0) ?? 0, optNum(1))
break
case "substr":
result = value.substr(optNum(0) ?? 0, optNum(1))
break
// JS charCodeAt returns NaN out of range; NaN flows as an ordinary in-sandbox value
// (normalized to null only at the data boundary - see copyOut), so return it as-is.
case "charCodeAt": result = value.charCodeAt(optNum(0) ?? 0); break
case "codePointAt": result = value.codePointAt(optNum(0) ?? 0); break
case "toString": result = value; break
case "charCodeAt":
result = value.charCodeAt(optNum(0) ?? 0)
break
case "codePointAt":
result = value.codePointAt(optNum(0) ?? 0)
break
case "toString":
result = value
break
case "concat": {
result = value.concat(...args.map((_, index) => str(index)))
break
}
default: throw new InterpreterRuntimeError(`String method '${name}' is not available in CodeMode.`, node)
default:
throw new InterpreterRuntimeError(`String method '${name}' is not available in CodeMode.`, node)
}
return boundedData(result, `String.${name} result`)
}
@@ -840,8 +1033,12 @@ const invokeNumberMethod = (value: number, name: string, args: Array<unknown>, n
}
let result: unknown
switch (name) {
case "toFixed": result = value.toFixed(optNum(0)); break
case "toExponential": result = value.toExponential(optNum(0)); break
case "toFixed":
result = value.toFixed(optNum(0))
break
case "toExponential":
result = value.toExponential(optNum(0))
break
case "toPrecision": {
const digits = optNum(0)
result = digits === undefined ? value.toString() : value.toPrecision(digits)
@@ -855,7 +1052,8 @@ const invokeNumberMethod = (value: number, name: string, args: Array<unknown>, n
result = value.toString(radix)
break
}
default: throw new InterpreterRuntimeError(`Number method '${name}' is not available in CodeMode.`, node)
default:
throw new InterpreterRuntimeError(`Number method '${name}' is not available in CodeMode.`, node)
}
return boundedData(result, `Number.${name} result`)
}
@@ -866,7 +1064,8 @@ const coerceToString = (value: unknown): string => {
if (value === undefined) return "undefined"
// Sandbox values stringify deterministically: Date as ISO (not the host's locale/timezone
// toString), RegExp as its literal form, Map/Set with their JS Object.prototype tags.
if (value instanceof SandboxDate) return Number.isFinite(value.time) ? new Date(value.time).toISOString() : "Invalid Date"
if (value instanceof SandboxDate)
return Number.isFinite(value.time) ? new Date(value.time).toISOString() : "Invalid Date"
if (value instanceof SandboxRegExp) return `/${value.regex.source}/${value.regex.flags}`
if (value instanceof SandboxMap) return "[object Map]"
if (value instanceof SandboxSet) return "[object Set]"
@@ -903,7 +1102,8 @@ const invokeCoercion = (ref: CoercionFunction, args: Array<unknown>, node: AstNo
if (ref.name === "Boolean") return Boolean(value)
if (ref.name === "parseInt") {
const radix = args[1]
if (radix !== undefined && typeof radix !== "number") throw new InterpreterRuntimeError("parseInt expects a numeric radix.", node)
if (radix !== undefined && typeof radix !== "number")
throw new InterpreterRuntimeError("parseInt expects a numeric radix.", node)
return parseInt(coerceToString(value), radix)
}
if (ref.name === "parseFloat") return parseFloat(coerceToString(value))
@@ -938,9 +1138,12 @@ const invokeObjectMethod = (name: string, args: Array<unknown>, node: AstNode):
}
return Object.keys(value)
}
case "values": return Object.values(requireObject())
case "entries": return Object.entries(requireObject()).map(([key, item]) => [key, item])
case "hasOwn": return Object.hasOwn(requireObject(), String(args[1]))
case "values":
return Object.values(requireObject())
case "entries":
return Object.entries(requireObject()).map(([key, item]) => [key, item])
case "hasOwn":
return Object.hasOwn(requireObject(), String(args[1]))
case "assign": {
const out: Record<string, unknown> = Object.create(null)
for (const source of args) {
@@ -948,7 +1151,8 @@ const invokeObjectMethod = (name: string, args: Array<unknown>, node: AstNode):
const value = boundedData(source, "Object.assign input")
// A sandbox value source contributes nothing (no own enumerable properties in JS).
if (isSandboxValue(value)) continue
if (value === null || typeof value !== "object" || Array.isArray(value)) throw new InterpreterRuntimeError("Object.assign expects data objects.", node)
if (value === null || typeof value !== "object" || Array.isArray(value))
throw new InterpreterRuntimeError("Object.assign expects data objects.", node)
for (const [key, item] of Object.entries(value)) guardedSet(out, key, item)
}
return out
@@ -962,15 +1166,18 @@ const invokeObjectMethod = (name: string, args: Array<unknown>, node: AstNode):
return out
}
const pairs = boundedData(args[0], "Object.fromEntries input")
if (!Array.isArray(pairs)) throw new InterpreterRuntimeError("Object.fromEntries expects an array of [key, value] pairs.", node)
if (!Array.isArray(pairs))
throw new InterpreterRuntimeError("Object.fromEntries expects an array of [key, value] pairs.", node)
const out: Record<string, unknown> = Object.create(null)
for (const pair of pairs) {
if (!Array.isArray(pair)) throw new InterpreterRuntimeError("Object.fromEntries expects [key, value] pairs.", node)
if (!Array.isArray(pair))
throw new InterpreterRuntimeError("Object.fromEntries expects [key, value] pairs.", node)
guardedSet(out, String(pair[0]), pair[1])
}
return out
}
default: throw new InterpreterRuntimeError(`Object.${name} is not available in CodeMode.`, node)
default:
throw new InterpreterRuntimeError(`Object.${name} is not available in CodeMode.`, node)
}
}
@@ -981,23 +1188,40 @@ const invokeMathMethod = (name: string, args: Array<unknown>, node: AstNode): nu
})
const [a = Number.NaN, b = Number.NaN] = nums
switch (name) {
case "max": return Math.max(...nums)
case "min": return Math.min(...nums)
case "abs": return Math.abs(a)
case "floor": return Math.floor(a)
case "ceil": return Math.ceil(a)
case "round": return Math.round(a)
case "trunc": return Math.trunc(a)
case "sign": return Math.sign(a)
case "sqrt": return Math.sqrt(a)
case "cbrt": return Math.cbrt(a)
case "pow": return Math.pow(a, b)
case "hypot": return Math.hypot(...nums)
case "log": return Math.log(a)
case "log2": return Math.log2(a)
case "log10": return Math.log10(a)
case "exp": return Math.exp(a)
default: throw new InterpreterRuntimeError(`Math.${name} is not available in CodeMode.`, node)
case "max":
return Math.max(...nums)
case "min":
return Math.min(...nums)
case "abs":
return Math.abs(a)
case "floor":
return Math.floor(a)
case "ceil":
return Math.ceil(a)
case "round":
return Math.round(a)
case "trunc":
return Math.trunc(a)
case "sign":
return Math.sign(a)
case "sqrt":
return Math.sqrt(a)
case "cbrt":
return Math.cbrt(a)
case "pow":
return Math.pow(a, b)
case "hypot":
return Math.hypot(...nums)
case "log":
return Math.log(a)
case "log2":
return Math.log2(a)
case "log10":
return Math.log10(a)
case "exp":
return Math.exp(a)
default:
throw new InterpreterRuntimeError(`Math.${name} is not available in CodeMode.`, node)
}
}
@@ -1006,7 +1230,12 @@ const invokeJsonMethod = (name: string, args: Array<unknown>, node: AstNode): un
case "stringify": {
const replacer = args[1]
if (Array.isArray(replacer) || replacer instanceof CodeModeFunction) {
throw new InterpreterRuntimeError("JSON.stringify replacers are not supported in CodeMode.", node, "UnsupportedSyntax", [supportedSyntaxMessage])
throw new InterpreterRuntimeError(
"JSON.stringify replacers are not supported in CodeMode.",
node,
"UnsupportedSyntax",
[supportedSyntaxMessage],
)
}
const space = args[2]
const indent = typeof space === "number" || typeof space === "string" ? space : undefined
@@ -1029,7 +1258,8 @@ const invokeJsonMethod = (name: string, args: Array<unknown>, node: AstNode): un
}
return copyIn(parsed, "JSON.parse result")
}
default: throw new InterpreterRuntimeError(`JSON.${name} is not available in CodeMode.`, node)
default:
throw new InterpreterRuntimeError(`JSON.${name} is not available in CodeMode.`, node)
}
}
@@ -1049,12 +1279,17 @@ const invokeArrayStatic = (name: string, args: Array<unknown>, node: AstNode): u
)
}
// Map/Set materialize directly (the data checkpoint would serialize them to {}).
if (args[0] instanceof SandboxMap) return Array.from((args[0] as SandboxMap).map.entries(), ([key, item]) => [key, item])
if (args[0] instanceof SandboxMap)
return Array.from((args[0] as SandboxMap).map.entries(), ([key, item]) => [key, item])
if (args[0] instanceof SandboxSet) return Array.from((args[0] as SandboxSet).set.values())
const source = boundedData(args[0], "Array.from input")
if (typeof source === "string") return Array.from(source)
if (Array.isArray(source)) return [...source]
if (source !== null && typeof source === "object" && typeof (source as { length?: unknown }).length === "number") {
if (
source !== null &&
typeof source === "object" &&
typeof (source as { length?: unknown }).length === "number"
) {
return Array.from(source as ArrayLike<unknown>)
}
throw new InterpreterRuntimeError("Array.from expects an array, string, Map, Set, or array-like value.", node)
@@ -1067,17 +1302,24 @@ const invokeArrayStatic = (name: string, args: Array<unknown>, node: AstNode): u
const invokeNumberStatic = (name: string, args: Array<unknown>, node: AstNode): unknown => {
const value = args[0]
switch (name) {
case "isInteger": return Number.isInteger(value)
case "isFinite": return Number.isFinite(value)
case "isNaN": return Number.isNaN(value)
case "isSafeInteger": return Number.isSafeInteger(value)
case "isInteger":
return Number.isInteger(value)
case "isFinite":
return Number.isFinite(value)
case "isNaN":
return Number.isNaN(value)
case "isSafeInteger":
return Number.isSafeInteger(value)
case "parseInt": {
const radix = args[1]
if (radix !== undefined && typeof radix !== "number") throw new InterpreterRuntimeError("Number.parseInt expects a numeric radix.", node)
if (radix !== undefined && typeof radix !== "number")
throw new InterpreterRuntimeError("Number.parseInt expects a numeric radix.", node)
return parseInt(coerceToString(value), radix)
}
case "parseFloat": return parseFloat(coerceToString(value))
default: throw new InterpreterRuntimeError(`Number.${name} is not available in CodeMode.`, node)
case "parseFloat":
return parseFloat(coerceToString(value))
default:
throw new InterpreterRuntimeError(`Number.${name} is not available in CodeMode.`, node)
}
}
@@ -1087,54 +1329,82 @@ const invokeStringStatic = (name: string, args: Array<unknown>, node: AstNode):
return arg
})
switch (name) {
case "fromCharCode": return String.fromCharCode(...codes)
case "fromCodePoint": return String.fromCodePoint(...codes)
default: throw new InterpreterRuntimeError(`String.${name} is not available in CodeMode.`, node)
case "fromCharCode":
return String.fromCharCode(...codes)
case "fromCodePoint":
return String.fromCodePoint(...codes)
default:
throw new InterpreterRuntimeError(`String.${name} is not available in CodeMode.`, node)
}
}
const invokeDateStatic = (name: string, args: Array<unknown>, node: AstNode): number => {
switch (name) {
case "now": return Date.now()
case "parse": return Date.parse(coerceToString(args[0]))
case "now":
return Date.now()
case "parse":
return Date.parse(coerceToString(args[0]))
case "UTC": {
const parts = args.map((arg) => coerceToNumber(arg))
return Date.UTC(...(parts as Parameters<typeof Date.UTC>))
}
default: throw new InterpreterRuntimeError(`Date.${name} is not available in CodeMode.`, node)
default:
throw new InterpreterRuntimeError(`Date.${name} is not available in CodeMode.`, node)
}
}
const invokeDateMethod = (value: SandboxDate, name: string, node: AstNode): unknown => {
const hosted = new Date(value.time)
switch (name) {
case "getTime": case "valueOf": return value.time
case "getTime":
case "valueOf":
return value.time
case "toISOString": {
if (!Number.isFinite(value.time)) throw new InterpreterRuntimeError("Invalid time value.", node)
return hosted.toISOString()
}
// toJSON of an invalid date is null in JS (never a throw); toString stays ISO for
// determinism across host timezones/locales.
case "toJSON": return Number.isFinite(value.time) ? hosted.toISOString() : null
case "toString": return coerceToString(value)
case "getFullYear": return hosted.getFullYear()
case "getMonth": return hosted.getMonth()
case "getDate": return hosted.getDate()
case "getDay": return hosted.getDay()
case "getHours": return hosted.getHours()
case "getMinutes": return hosted.getMinutes()
case "getSeconds": return hosted.getSeconds()
case "getMilliseconds": return hosted.getMilliseconds()
case "getUTCFullYear": return hosted.getUTCFullYear()
case "getUTCMonth": return hosted.getUTCMonth()
case "getUTCDate": return hosted.getUTCDate()
case "getUTCDay": return hosted.getUTCDay()
case "getUTCHours": return hosted.getUTCHours()
case "getUTCMinutes": return hosted.getUTCMinutes()
case "getUTCSeconds": return hosted.getUTCSeconds()
case "getUTCMilliseconds": return hosted.getUTCMilliseconds()
case "getTimezoneOffset": return hosted.getTimezoneOffset()
default: throw new InterpreterRuntimeError(`Date method '${name}' is not available in CodeMode.`, node)
case "toJSON":
return Number.isFinite(value.time) ? hosted.toISOString() : null
case "toString":
return coerceToString(value)
case "getFullYear":
return hosted.getFullYear()
case "getMonth":
return hosted.getMonth()
case "getDate":
return hosted.getDate()
case "getDay":
return hosted.getDay()
case "getHours":
return hosted.getHours()
case "getMinutes":
return hosted.getMinutes()
case "getSeconds":
return hosted.getSeconds()
case "getMilliseconds":
return hosted.getMilliseconds()
case "getUTCFullYear":
return hosted.getUTCFullYear()
case "getUTCMonth":
return hosted.getUTCMonth()
case "getUTCDate":
return hosted.getUTCDate()
case "getUTCDay":
return hosted.getUTCDay()
case "getUTCHours":
return hosted.getUTCHours()
case "getUTCMinutes":
return hosted.getUTCMinutes()
case "getUTCSeconds":
return hosted.getUTCSeconds()
case "getUTCMilliseconds":
return hosted.getUTCMilliseconds()
case "getTimezoneOffset":
return hosted.getTimezoneOffset()
default:
throw new InterpreterRuntimeError(`Date method '${name}' is not available in CodeMode.`, node)
}
}
@@ -1142,26 +1412,31 @@ const invokeRegExpMethod = (value: SandboxRegExp, name: string, args: Array<unkn
switch (name) {
// test/exec run on the sandbox regex's own host instance, so `g`-flag lastIndex advances
// across calls per the spec.
case "test": return value.regex.test(coerceToString(args[0]))
case "test":
return value.regex.test(coerceToString(args[0]))
case "exec": {
const matched = value.regex.exec(coerceToString(args[0]))
if (matched === null) return null
return matchToValue(matched)
}
case "toString": return coerceToString(value)
default: throw new InterpreterRuntimeError(`RegExp method '${name}' is not available in CodeMode.`, node)
case "toString":
return coerceToString(value)
default:
throw new InterpreterRuntimeError(`RegExp method '${name}' is not available in CodeMode.`, node)
}
}
const invokeGlobalMethod = (ref: GlobalMethodReference, args: Array<unknown>, node: AstNode): unknown => {
if (ref.namespace === "console") throw new InterpreterRuntimeError(`console.${ref.name} is not available in CodeMode.`, node)
if (ref.namespace === "console")
throw new InterpreterRuntimeError(`console.${ref.name} is not available in CodeMode.`, node)
if (ref.namespace === "Object") return invokeObjectMethod(ref.name, args, node)
if (ref.namespace === "Math") return invokeMathMethod(ref.name, args, node)
if (ref.namespace === "Array") return invokeArrayStatic(ref.name, args, node)
if (ref.namespace === "Number") return invokeNumberStatic(ref.name, args, node)
if (ref.namespace === "String") return invokeStringStatic(ref.name, args, node)
if (ref.namespace === "Date") {
if (!dateStatics.has(ref.name)) throw new InterpreterRuntimeError(`Date.${ref.name} is not available in CodeMode.`, node)
if (!dateStatics.has(ref.name))
throw new InterpreterRuntimeError(`Date.${ref.name} is not available in CodeMode.`, node)
return invokeDateStatic(ref.name, args, node)
}
if (ref.namespace === "RegExp" || ref.namespace === "Map" || ref.namespace === "Set") {
@@ -1174,7 +1449,8 @@ const invokeGlobalMethod = (ref: GlobalMethodReference, args: Array<unknown>, no
const spreadItems = (spread: unknown): Array<unknown> | undefined => {
if (Array.isArray(spread)) return spread
if (typeof spread === "string") return Array.from(spread)
if (spread instanceof SandboxMap) return Array.from(spread.map.entries(), ([key, item]): Array<unknown> => [key, item])
if (spread instanceof SandboxMap)
return Array.from(spread.map.entries(), ([key, item]): Array<unknown> => [key, item])
if (spread instanceof SandboxSet) return Array.from(spread.set.values())
return undefined
}
@@ -1267,7 +1543,7 @@ class Interpreter<R> {
// top-level declarations (`let undefined = 5`, `const Object = ...`) shadow builtins like
// JS module scope, instead of colliding with the seeded globals.
this.pushScope()
return Effect.gen(function*() {
return Effect.gen(function* () {
self.hoistFunctions(program.body)
let value: unknown = undefined
let returned = false
@@ -1305,7 +1581,7 @@ class Interpreter<R> {
// diagnostic (interrupted calls, e.g. Promise.race losers, are ignored).
private drainPendingSettlements(): Effect.Effect<void, unknown, never> {
const self = this
return Effect.gen(function*() {
return Effect.gen(function* () {
for (const promise of [...self.pendingSettlements]) {
const exit = yield* self.observePromise(promise)
if (Exit.isSuccess(exit) || Cause.hasInterruptsOnly(exit.cause)) continue
@@ -1324,13 +1600,15 @@ class Interpreter<R> {
// scope teardown interrupt it) gated by the concurrency semaphore, and wraps the fiber in a
// first-class promise value. `startImmediately` makes the runtime admit the call - charging
// the tool-call budget and firing onToolCallStart - at the call site, before any await.
private createToolCallPromise(path: ReadonlyArray<string>, args: Array<unknown>): Effect.Effect<SandboxPromise, never, R> {
private createToolCallPromise(
path: ReadonlyArray<string>,
args: Array<unknown>,
): Effect.Effect<SandboxPromise, never, R> {
const self = this
return Effect.map(
Effect.forkChild(
this.callPermits.withPermit(Effect.suspend(() => self.invokeTool(path, args))),
{ startImmediately: true },
),
Effect.forkChild(this.callPermits.withPermit(Effect.suspend(() => self.invokeTool(path, args))), {
startImmediately: true,
}),
(fiber) => {
const promise = new SandboxPromise(fiber)
self.pendingSettlements.add(promise)
@@ -1354,13 +1632,22 @@ class Interpreter<R> {
return Effect.flatMap(this.observePromise(promise), (exit) => self.unwrapPromiseExit(promise, exit, node))
}
private unwrapPromiseExit(promise: SandboxPromise | undefined, exit: Exit.Exit<unknown, unknown>, node?: AstNode): Effect.Effect<unknown, unknown> {
private unwrapPromiseExit(
promise: SandboxPromise | undefined,
exit: Exit.Exit<unknown, unknown>,
node?: AstNode,
): Effect.Effect<unknown, unknown> {
if (Exit.isSuccess(exit)) return Effect.succeed(exit.value)
// A call Promise.race interrupted after losing settles as a catchable program failure;
// any other interruption is execution teardown (timeout/host) and must keep propagating
// as interruption rather than becoming program-visible data.
if (promise?.interrupted === true && Cause.hasInterruptsOnly(exit.cause)) {
return Effect.fail(new InterpreterRuntimeError("This tool call was interrupted because another value settled a Promise.race first.", node))
return Effect.fail(
new InterpreterRuntimeError(
"This tool call was interrupted because another value settled a Promise.race first.",
node,
),
)
}
return Effect.failCause(exit.cause)
}
@@ -1413,7 +1700,7 @@ class Interpreter<R> {
private evaluateBlock(node: AstNode): Effect.Effect<StatementResult, unknown, R> {
this.pushScope()
const self = this
return Effect.gen(function*() {
return Effect.gen(function* () {
const body = getArray(node, "body")
self.hoistFunctions(body)
@@ -1437,7 +1724,12 @@ class Interpreter<R> {
private createFunction(node: AstNode): CodeModeFunction {
if (node.generator === true) {
throw new InterpreterRuntimeError("Generator functions are not supported in CodeMode.", node, "UnsupportedSyntax", [supportedSyntaxMessage])
throw new InterpreterRuntimeError(
"Generator functions are not supported in CodeMode.",
node,
"UnsupportedSyntax",
[supportedSyntaxMessage],
)
}
return new CodeModeFunction(
getArray(node, "params").map((parameter, index) => asNode(parameter, `params[${index}]`)),
@@ -1462,16 +1754,25 @@ class Interpreter<R> {
const alternateNode = getOptionalNode(node, "alternate")
return Effect.flatMap(this.evaluateExpression(testNode), (test) =>
test ? this.evaluateStatement(consequentNode) : alternateNode ? this.evaluateStatement(alternateNode) : Effect.succeed({ kind: "none" }))
test
? this.evaluateStatement(consequentNode)
: alternateNode
? this.evaluateStatement(alternateNode)
: Effect.succeed({ kind: "none" }),
)
}
private evaluateSwitchStatement(node: AstNode): Effect.Effect<StatementResult, unknown, R> {
const self = this
this.pushScope()
return Effect.gen(function*() {
return Effect.gen(function* () {
const discriminant = yield* self.evaluateExpression(getNode(node, "discriminant"))
if (containsOpaqueReference(discriminant)) {
throw new InterpreterRuntimeError("Switch discriminants must be data values in CodeMode.", node, "InvalidDataValue")
throw new InterpreterRuntimeError(
"Switch discriminants must be data values in CodeMode.",
node,
"InvalidDataValue",
)
}
const cases = getArray(node, "cases").map((value, index) => asNode(value, `cases[${index}]`))
let defaultIndex: number | undefined
@@ -1484,7 +1785,11 @@ class Interpreter<R> {
}
const candidate = yield* self.evaluateExpression(test)
if (containsOpaqueReference(candidate)) {
throw new InterpreterRuntimeError("Switch case values must be data values in CodeMode.", test, "InvalidDataValue")
throw new InterpreterRuntimeError(
"Switch case values must be data values in CodeMode.",
test,
"InvalidDataValue",
)
}
if (candidate === discriminant) {
selected = index
@@ -1510,7 +1815,7 @@ class Interpreter<R> {
const bodyNode = getNode(node, "body")
const self = this
return Effect.gen(function*() {
return Effect.gen(function* () {
while (yield* self.evaluateExpression(testNode)) {
const result = yield* self.evaluateStatement(bodyNode)
@@ -1519,7 +1824,7 @@ class Interpreter<R> {
}
if (result.kind === "break") {
return { kind: "none" } satisfies StatementResult
return { kind: "none" } satisfies StatementResult
}
if (result.kind === "return") {
@@ -1540,7 +1845,7 @@ class Interpreter<R> {
const testNode = getNode(node, "test")
const self = this
return Effect.gen(function*() {
return Effect.gen(function* () {
do {
const result = yield* self.evaluateStatement(bodyNode)
@@ -1568,7 +1873,7 @@ class Interpreter<R> {
private evaluateForStatement(node: AstNode): Effect.Effect<StatementResult, unknown, R> {
this.pushScope()
const self = this
return Effect.gen(function*() {
return Effect.gen(function* () {
const initNode = getOptionalNode(node, "init")
const testNode = getOptionalNode(node, "test")
const updateNode = getOptionalNode(node, "update")
@@ -1582,23 +1887,28 @@ class Interpreter<R> {
}
}
const perIterationBindings = initNode?.type === "VariableDeclaration" && getString(initNode, "kind") !== "var"
? Array.from(self.currentScope().keys())
: []
const perIterationBindings =
initNode?.type === "VariableDeclaration" && getString(initNode, "kind") !== "var"
? Array.from(self.currentScope().keys())
: []
while (testNode ? yield* self.evaluateExpression(testNode) : true) {
let iterationScope: Map<string, Binding> | undefined
if (perIterationBindings.length > 0) {
iterationScope = new Map(perIterationBindings.map((name) => {
const binding = self.currentScope().get(name)!
return [name, { ...binding }]
}))
iterationScope = new Map(
perIterationBindings.map((name) => {
const binding = self.currentScope().get(name)!
return [name, { ...binding }]
}),
)
self.scopes.push(iterationScope)
}
const result = yield* self.evaluateStatement(bodyNode).pipe(
Effect.ensuring(Effect.sync(() => {
if (iterationScope) self.popScope()
})),
Effect.ensuring(
Effect.sync(() => {
if (iterationScope) self.popScope()
}),
),
)
if (result.kind === "return") {
@@ -1639,7 +1949,7 @@ class Interpreter<R> {
}
const self = this
return Effect.gen(function*() {
return Effect.gen(function* () {
const left = getNode(node, "left")
const right = yield* self.evaluateExpression(getNode(node, "right"))
const body = getNode(node, "body")
@@ -1677,9 +1987,11 @@ class Interpreter<R> {
}
const result = yield* self.evaluateStatement(body).pipe(
Effect.ensuring(Effect.sync(() => {
if (declaration) self.popScope()
})),
Effect.ensuring(
Effect.sync(() => {
if (declaration) self.popScope()
}),
),
)
if (result.kind === "return") {
@@ -1723,7 +2035,7 @@ class Interpreter<R> {
private evaluateForInStatement(node: AstNode): Effect.Effect<StatementResult, unknown, R> {
const self = this
return Effect.gen(function*() {
return Effect.gen(function* () {
const left = getNode(node, "left")
const right = yield* self.evaluateExpression(getNode(node, "right"))
const body = getNode(node, "body")
@@ -1768,9 +2080,11 @@ class Interpreter<R> {
}
const result = yield* self.evaluateStatement(body).pipe(
Effect.ensuring(Effect.sync(() => {
if (declaration) self.popScope()
})),
Effect.ensuring(
Effect.sync(() => {
if (declaration) self.popScope()
}),
),
)
if (result.kind === "return") {
@@ -1836,12 +2150,10 @@ class Interpreter<R> {
const caught = caughtErrorValue(Cause.squash(cause))
const parameter = getOptionalNode(handler, "param")
self.pushScope()
return Effect.gen(function*() {
return Effect.gen(function* () {
if (parameter) yield* self.declarePattern(parameter, caught, true, handler)
return yield* self.evaluateStatement(getNode(handler, "body"))
}).pipe(
Effect.ensuring(Effect.sync(() => self.popScope())),
)
}).pipe(Effect.ensuring(Effect.sync(() => self.popScope())))
},
onSuccess: Effect.succeed,
})
@@ -1856,10 +2168,12 @@ class Interpreter<R> {
cause.reasons.some(Cause.isInterruptReason)
? Effect.failCause(cause)
: Effect.flatMap(this.evaluateStatement(finalizer), (final) =>
isAbrupt(final) ? Effect.succeed(final) : Effect.failCause(cause)),
isAbrupt(final) ? Effect.succeed(final) : Effect.failCause(cause),
),
onSuccess: (result) =>
Effect.flatMap(this.evaluateStatement(finalizer), (final) =>
isAbrupt(final) ? Effect.succeed(final) : Effect.succeed(result)),
isAbrupt(final) ? Effect.succeed(final) : Effect.succeed(result),
),
})
}
@@ -1867,7 +2181,7 @@ class Interpreter<R> {
const kind = getString(node, "kind")
const declarations = getArray(node, "declarations")
const self = this
return Effect.gen(function*() {
return Effect.gen(function* () {
for (const declarationValue of declarations) {
const declaration = asNode(declarationValue, "declarations")
@@ -1882,9 +2196,14 @@ class Interpreter<R> {
})
}
private declarePattern(pattern: AstNode, value: unknown, mutable: boolean, node: AstNode): Effect.Effect<void, unknown, R> {
private declarePattern(
pattern: AstNode,
value: unknown,
mutable: boolean,
node: AstNode,
): Effect.Effect<void, unknown, R> {
const self = this
return Effect.gen(function*() {
return Effect.gen(function* () {
if (pattern.type === "Identifier") {
self.declare(getString(pattern, "name"), value, mutable, node)
return
@@ -1899,7 +2218,11 @@ class Interpreter<R> {
if (pattern.type === "ObjectPattern") {
if (value === null || typeof value !== "object" || Array.isArray(value) || isRuntimeReference(value)) {
throw new InterpreterRuntimeError("Object destructuring requires a data object value.", pattern, "InvalidDataValue")
throw new InterpreterRuntimeError(
"Object destructuring requires a data object value.",
pattern,
"InvalidDataValue",
)
}
const consumed = new Set<string>()
@@ -1916,7 +2239,11 @@ class Interpreter<R> {
continue
}
if (property.type !== "Property" || getBoolean(property, "computed") || getString(property, "kind") !== "init") {
if (
property.type !== "Property" ||
getBoolean(property, "computed") ||
getString(property, "kind") !== "init"
) {
throw new InterpreterRuntimeError("Only named object destructuring properties are supported.", property)
}
@@ -1960,7 +2287,9 @@ class Interpreter<R> {
// sandbox regex from those (the host `value` instance is never exposed).
const regex = node.regex
if (isRecord(regex) && typeof regex.pattern === "string") {
return Effect.sync(() => this.constructRegExp([regex.pattern, typeof regex.flags === "string" ? regex.flags : ""], node))
return Effect.sync(() =>
this.constructRegExp([regex.pattern, typeof regex.flags === "string" ? regex.flags : ""], node),
)
}
return Effect.sync(() => boundedData(node.value, "Literal"))
}
@@ -1983,7 +2312,8 @@ class Interpreter<R> {
return this.readMember(node)
case "ChainExpression":
return Effect.map(this.evaluateExpression(getNode(node, "expression")), (value) =>
value === OptionalShortCircuit ? undefined : value)
value === OptionalShortCircuit ? undefined : value,
)
case "ObjectExpression":
return this.evaluateObjectExpression(node)
case "ArrayExpression":
@@ -1999,7 +2329,8 @@ class Interpreter<R> {
// matching real JS semantics for non-thenables.
const self = this
return Effect.flatMap(this.evaluateExpression(getNode(node, "argument")), (value) =>
value instanceof SandboxPromise ? self.settlePromise(value, node) : Effect.succeed(value))
value instanceof SandboxPromise ? self.settlePromise(value, node) : Effect.succeed(value),
)
}
case "NewExpression":
return this.evaluateNewExpression(node)
@@ -2025,19 +2356,24 @@ class Interpreter<R> {
)
}
if (errorConstructors.has(name)) {
return Effect.gen(function*() {
const arg = argNodes.length > 0 ? yield* self.evaluateExpression(asNode(argNodes[0], "arguments[0]")) : undefined
return Effect.gen(function* () {
const arg =
argNodes.length > 0 ? yield* self.evaluateExpression(asNode(argNodes[0], "arguments[0]")) : undefined
return createErrorValue(name, arg === undefined ? "" : coerceToString(arg))
})
}
if (valueConstructors.has(name)) {
return Effect.gen(function*() {
return Effect.gen(function* () {
const args = yield* self.evaluateCallArguments(argNodes)
switch (name) {
case "Date": return self.constructDate(args)
case "RegExp": return self.constructRegExp(args, node)
case "Map": return self.constructMap(args[0], node)
default: return self.constructSet(args[0], node)
case "Date":
return self.constructDate(args)
case "RegExp":
return self.constructRegExp(args, node)
case "Map":
return self.constructMap(args[0], node)
default:
return self.constructSet(args[0], node)
}
})
}
@@ -2060,7 +2396,8 @@ class Interpreter<R> {
private constructRegExp(args: Array<unknown>, node: AstNode): SandboxRegExp {
const first = args[0]
const pattern = first instanceof SandboxRegExp ? first.regex.source : first === undefined ? "" : coerceToString(first)
const pattern =
first instanceof SandboxRegExp ? first.regex.source : first === undefined ? "" : coerceToString(first)
const flagsArg = args[1]
if (flagsArg !== undefined && typeof flagsArg !== "string") {
throw new InterpreterRuntimeError(
@@ -2094,7 +2431,10 @@ class Interpreter<R> {
? Array.from(init.map.entries(), ([key, item]): Array<unknown> => [key, item])
: undefined
if (entries === undefined) {
throw new InterpreterRuntimeError("new Map(...) expects an array of [key, value] pairs, a Map, or no argument.", node)
throw new InterpreterRuntimeError(
"new Map(...) expects an array of [key, value] pairs, a Map, or no argument.",
node,
)
}
for (const pair of entries) {
if (!Array.isArray(pair)) {
@@ -2122,12 +2462,10 @@ class Interpreter<R> {
return target
}
private evaluateBinaryExpression(node: AstNode): Effect.Effect<unknown, unknown, R> {
const operator = getString(node, "operator")
const self = this
return Effect.gen(function*() {
return Effect.gen(function* () {
const lhs = yield* self.evaluateExpression(getNode(node, "left"))
const rhs = yield* self.evaluateExpression(getNode(node, "right"))
// Like `typeof`, `instanceof` observes any value without coercing it (a promise or
@@ -2161,34 +2499,55 @@ class Interpreter<R> {
const l = coerceOperand(lhs)
const r = coerceOperand(rhs)
switch (operator) {
case "+": return (l as string) + (r as string)
case "-": return (l as number) - (r as number)
case "*": return (l as number) * (r as number)
case "/": return (l as number) / (r as number)
case "%": return (l as number) % (r as number)
case "**": return (l as number) ** (r as number)
case "+":
return (l as string) + (r as string)
case "-":
return (l as number) - (r as number)
case "*":
return (l as number) * (r as number)
case "/":
return (l as number) / (r as number)
case "%":
return (l as number) % (r as number)
case "**":
return (l as number) ** (r as number)
// Two objects compare by identity in JS (no ToPrimitive); only object-vs-primitive coerces.
case "==": return bothObjects ? lhs === rhs : l == r
case "===": return lhs === rhs
case "!=": return bothObjects ? lhs !== rhs : l != r
case "!==": return lhs !== rhs
case "<": return (l as string) < (r as string)
case "<=": return (l as string) <= (r as string)
case ">": return (l as string) > (r as string)
case ">=": return (l as string) >= (r as string)
case "&": return (l as number) & (r as number)
case "|": return (l as number) | (r as number)
case "^": return (l as number) ^ (r as number)
case "<<": return (l as number) << (r as number)
case ">>": return (l as number) >> (r as number)
case ">>>": return (l as number) >>> (r as number)
case "==":
return bothObjects ? lhs === rhs : l == r
case "===":
return lhs === rhs
case "!=":
return bothObjects ? lhs !== rhs : l != r
case "!==":
return lhs !== rhs
case "<":
return (l as string) < (r as string)
case "<=":
return (l as string) <= (r as string)
case ">":
return (l as string) > (r as string)
case ">=":
return (l as string) >= (r as string)
case "&":
return (l as number) & (r as number)
case "|":
return (l as number) | (r as number)
case "^":
return (l as number) ^ (r as number)
case "<<":
return (l as number) << (r as number)
case ">>":
return (l as number) >> (r as number)
case ">>>":
return (l as number) >>> (r as number)
case "in":
if (rhs === null || typeof rhs !== "object") {
throw new InterpreterRuntimeError("The 'in' operator requires a data object on the right-hand side.", node)
}
// Own properties only, so arrays don't leak the host Array.prototype (map/constructor/...).
return Object.hasOwn(rhs as object, coerceOperand(lhs) as PropertyKey)
default: throw new InterpreterRuntimeError(`Unsupported binary operator '${operator}'.`, node)
default:
throw new InterpreterRuntimeError(`Unsupported binary operator '${operator}'.`, node)
}
}
@@ -2197,7 +2556,10 @@ class Interpreter<R> {
return Effect.flatMap(this.evaluateExpression(getNode(node, "left")), (left) => {
if (operator === "&&") return left ? this.evaluateExpression(getNode(node, "right")) : Effect.succeed(left)
if (operator === "||") return left ? Effect.succeed(left) : this.evaluateExpression(getNode(node, "right"))
if (operator === "??") return left !== null && left !== undefined ? Effect.succeed(left) : this.evaluateExpression(getNode(node, "right"))
if (operator === "??")
return left !== null && left !== undefined
? Effect.succeed(left)
: this.evaluateExpression(getNode(node, "right"))
throw new InterpreterRuntimeError(`Unsupported logical operator '${operator}'.`, node)
})
}
@@ -2222,17 +2584,25 @@ class Interpreter<R> {
// Numeric/bitwise unary operators ToPrimitive their operand; a Date yields its time value
// (`+date` is the epoch-ms idiom), other null-prototype data objects/arrays coerce to
// their JS string form first (see evaluateBinaryExpression).
const operand = value instanceof SandboxDate
? value.time
: value !== null && typeof value === "object"
? coerceToString(value)
: value
const operand =
value instanceof SandboxDate
? value.time
: value !== null && typeof value === "object"
? coerceToString(value)
: value
let result: unknown
switch (operator) {
case "+": result = +(operand as number); break
case "-": result = -(operand as number); break
case "~": result = ~(operand as number); break
default: throw new InterpreterRuntimeError(`Unsupported unary operator '${operator}'.`, node)
case "+":
result = +(operand as number)
break
case "-":
result = -(operand as number)
break
case "~":
result = ~(operand as number)
break
default:
throw new InterpreterRuntimeError(`Unsupported unary operator '${operator}'.`, node)
}
return boundedData(result, "Unary expression result")
})
@@ -2242,7 +2612,7 @@ class Interpreter<R> {
const left = getNode(node, "left")
const operator = getString(node, "operator")
const self = this
return Effect.gen(function*() {
return Effect.gen(function* () {
if (operator === "??=" || operator === "||=" || operator === "&&=") {
return yield* self.evaluateLogicalAssignment(node, left, operator)
}
@@ -2250,13 +2620,19 @@ class Interpreter<R> {
if (left.type === "Identifier") {
const name = getString(left, "name")
if (operator === "=") return self.setIdentifierValue(name, rightValue, left)
const next = boundedData(self.applyCompoundAssignment(operator, self.getIdentifierValue(name, left), rightValue, node), "Assignment result")
const next = boundedData(
self.applyCompoundAssignment(operator, self.getIdentifierValue(name, left), rightValue, node),
"Assignment result",
)
return self.setIdentifierValue(name, next, left)
}
if (left.type === "MemberExpression") {
if (operator === "=") return yield* self.writeMember(left, rightValue)
return yield* self.modifyMember(left, (current) => {
const next = boundedData(self.applyCompoundAssignment(operator, current, rightValue, node), "Assignment result")
const next = boundedData(
self.applyCompoundAssignment(operator, current, rightValue, node),
"Assignment result",
)
return Effect.succeed({ write: true, next, result: next })
})
}
@@ -2264,13 +2640,17 @@ class Interpreter<R> {
})
}
private evaluateLogicalAssignment(node: AstNode, left: AstNode, operator: string): Effect.Effect<unknown, unknown, R> {
private evaluateLogicalAssignment(
node: AstNode,
left: AstNode,
operator: string,
): Effect.Effect<unknown, unknown, R> {
const self = this
const shouldAssign = (current: unknown): boolean =>
operator === "??=" ? current === null || current === undefined : operator === "||=" ? !current : Boolean(current)
if (left.type === "Identifier") {
const name = getString(left, "name")
return Effect.gen(function*() {
return Effect.gen(function* () {
const current = self.getIdentifierValue(name, left)
if (!shouldAssign(current)) return current
const rightValue = yield* self.evaluateExpression(getNode(node, "right"))
@@ -2281,8 +2661,13 @@ class Interpreter<R> {
// Resolve the member exactly once; evaluate the RHS only if we actually assign.
return self.modifyMember(left, (current) =>
shouldAssign(current)
? Effect.map(self.evaluateExpression(getNode(node, "right")), (rightValue) => ({ write: true, next: rightValue, result: rightValue }))
: Effect.succeed({ write: false, next: current, result: current }))
? Effect.map(self.evaluateExpression(getNode(node, "right")), (rightValue) => ({
write: true,
next: rightValue,
result: rightValue,
}))
: Effect.succeed({ write: false, next: current, result: current }),
)
}
throw new InterpreterRuntimeError("Assignment target must be an Identifier or MemberExpression.", left)
}
@@ -2324,7 +2709,7 @@ class Interpreter<R> {
const argNodes = getArray(node, "arguments")
const self = this
return Effect.gen(function*() {
return Effect.gen(function* () {
const callable = yield* self.evaluateExpression(callee)
if (callable === OptionalShortCircuit) return OptionalShortCircuit
if ((callable === null || callable === undefined) && node.optional === true) return OptionalShortCircuit
@@ -2359,7 +2744,7 @@ class Interpreter<R> {
if (callable instanceof ErrorConstructorReference) {
return createErrorValue(callable.name, args[0] === undefined ? "" : coerceToString(args[0]))
}
throw new InterpreterRuntimeError("Only tools are callable in CodeMode.", callee)
throw new InterpreterRuntimeError("Only tools are callable in CodeMode.", callee)
})
}
@@ -2379,7 +2764,8 @@ class Interpreter<R> {
}
private invokeConsole(name: string, args: Array<unknown>, node: AstNode): undefined {
if (!consoleMethods.has(name)) throw new InterpreterRuntimeError(`console.${name} is not available in CodeMode.`, node)
if (!consoleMethods.has(name))
throw new InterpreterRuntimeError(`console.${name} is not available in CodeMode.`, node)
this.logs.push(publicErrorMessage(this.formatConsoleMessage(name, args, node)))
return undefined
}
@@ -2439,7 +2825,9 @@ class Interpreter<R> {
if (Array.isArray(value)) {
return `[${value.map((item) => this.formatConsoleValue(item, seen, depth + 1)).join(",")}]`
}
return `{${Object.entries(value).map(([key, item]) => `${JSON.stringify(key)}:${this.formatConsoleValue(item, seen, depth + 1)}`).join(",")}}`
return `{${Object.entries(value)
.map(([key, item]) => `${JSON.stringify(key)}:${this.formatConsoleValue(item, seen, depth + 1)}`)
.join(",")}}`
} finally {
seen.delete(value)
}
@@ -2455,7 +2843,10 @@ class Interpreter<R> {
const rows = this.consoleTableRows(data, columns)
const keys = columns ?? Array.from(new Set(rows.flatMap((row) => Object.keys(row.values))))
const header = ["(index)", ...keys].join("\t")
return [header, ...rows.map((row) => [row.index, ...keys.map((key) => this.formatConsoleTableCell(row.values[key]))].join("\t"))].join("\n")
return [
header,
...rows.map((row) => [row.index, ...keys.map((key) => this.formatConsoleTableCell(row.values[key]))].join("\t")),
].join("\n")
}
private consoleTableColumns(value: unknown, node: AstNode): ReadonlyArray<string> | undefined {
@@ -2465,7 +2856,10 @@ class Interpreter<R> {
return Array.isArray(columns) ? columns.map((column) => String(column)) : undefined
}
private consoleTableRows(data: unknown, columns: ReadonlyArray<string> | undefined): Array<{ readonly index: string; readonly values: Record<string, unknown> }> {
private consoleTableRows(
data: unknown,
columns: ReadonlyArray<string> | undefined,
): Array<{ readonly index: string; readonly values: Record<string, unknown> }> {
if (Array.isArray(data)) {
return data.map((item, index) => ({ index: String(index), values: this.consoleTableValues(item, columns) }))
}
@@ -2492,14 +2886,18 @@ class Interpreter<R> {
private evaluateCallArguments(argNodes: Array<unknown>): Effect.Effect<Array<unknown>, unknown, R> {
const self = this
return Effect.gen(function*() {
return Effect.gen(function* () {
const args: Array<unknown> = []
for (const [index, arg] of argNodes.entries()) {
const argNode = asNode(arg, `arguments[${index}]`)
if (argNode.type === "SpreadElement") {
const spread = yield* self.evaluateExpression(getNode(argNode, "argument"))
const items = spreadItems(spread)
if (items === undefined) throw new InterpreterRuntimeError("Spread arguments require an array, string, Map, or Set in CodeMode.", argNode)
if (items === undefined)
throw new InterpreterRuntimeError(
"Spread arguments require an array, string, Map, or Set in CodeMode.",
argNode,
)
args.push(...items)
} else {
args.push(yield* self.evaluateExpression(argNode))
@@ -2514,13 +2912,19 @@ class Interpreter<R> {
// whatever - because tool calls already run eagerly on their own fibers; the combinators
// only observe settlements. Joining is therefore sequential (no extra fibers) without
// costing parallelism, and the concurrency cap stays where the work is: the fork semaphore.
private invokePromiseMethod(ref: PromiseMethodReference, args: Array<unknown>, node: AstNode): Effect.Effect<unknown, unknown, R> {
private invokePromiseMethod(
ref: PromiseMethodReference,
args: Array<unknown>,
node: AstNode,
): Effect.Effect<unknown, unknown, R> {
const self = this
if (ref.name === "resolve") {
// Promise.resolve of a promise is that promise (JS flattens); anything else is a
// promise already fulfilled with the value.
const value = args[0]
return Effect.succeed(value instanceof SandboxPromise ? value : new SandboxPromise(undefined, Effect.succeed(value)))
return Effect.succeed(
value instanceof SandboxPromise ? value : new SandboxPromise(undefined, Effect.succeed(value)),
)
}
if (ref.name === "reject") {
return Effect.sync(() => new SandboxPromise(undefined, Effect.fail(new ProgramThrow(args[0]))))
@@ -2539,8 +2943,10 @@ class Interpreter<R> {
// Mark every promise element observed up-front (Promise.all handles all of its
// members' failures, as in JS), then join in index order; the first failure rejects
// the whole call while unrelated in-flight members keep running.
const settles = items.map((item) => (item instanceof SandboxPromise ? this.settlePromise(item, node) : Effect.succeed(item)))
return Effect.gen(function*() {
const settles = items.map((item) =>
item instanceof SandboxPromise ? this.settlePromise(item, node) : Effect.succeed(item),
)
return Effect.gen(function* () {
const values: Array<unknown> = []
for (const settle of settles) values.push(yield* settle)
return values
@@ -2550,13 +2956,16 @@ class Interpreter<R> {
const observations = items.map((item) =>
item instanceof SandboxPromise
? Effect.map(this.observePromise(item), (exit) => ({ promise: item as SandboxPromise | undefined, exit }))
: Effect.succeed({ promise: undefined as SandboxPromise | undefined, exit: Exit.succeed(item as unknown) }))
return Effect.gen(function*() {
: Effect.succeed({ promise: undefined as SandboxPromise | undefined, exit: Exit.succeed(item as unknown) }),
)
return Effect.gen(function* () {
const outcomes: Array<unknown> = []
for (const observation of observations) {
const { exit, promise } = yield* observation
if (Exit.isSuccess(exit)) {
outcomes.push(Object.assign(Object.create(null) as SafeObject, { status: "fulfilled", value: exit.value }))
outcomes.push(
Object.assign(Object.create(null) as SafeObject, { status: "fulfilled", value: exit.value }),
)
continue
}
const raceInterrupted = promise?.interrupted === true && Cause.hasInterruptsOnly(exit.cause)
@@ -2565,22 +2974,34 @@ class Interpreter<R> {
return yield* Effect.failCause(exit.cause)
}
const thrown = raceInterrupted
? new InterpreterRuntimeError("This tool call was interrupted because another value settled a Promise.race first.", node)
? new InterpreterRuntimeError(
"This tool call was interrupted because another value settled a Promise.race first.",
node,
)
: Cause.squash(exit.cause)
outcomes.push(Object.assign(Object.create(null) as SafeObject, { status: "rejected", reason: caughtErrorValue(thrown) }))
outcomes.push(
Object.assign(Object.create(null) as SafeObject, {
status: "rejected",
reason: caughtErrorValue(thrown),
}),
)
}
return outcomes
})
}
case "race": {
if (items.length === 0) {
throw new InterpreterRuntimeError("Promise.race([]) would never settle; provide at least one promise or value.", node)
throw new InterpreterRuntimeError(
"Promise.race([]) would never settle; provide at least one promise or value.",
node,
)
}
const observations = items.map((item, index) =>
item instanceof SandboxPromise
? Effect.map(this.observePromise(item), (exit) => ({ index, exit }))
: Effect.succeed({ index, exit: Exit.succeed(item as unknown) }))
return Effect.gen(function*() {
: Effect.succeed({ index, exit: Exit.succeed(item as unknown) }),
)
return Effect.gen(function* () {
// First settlement (fulfilled OR rejected) wins; the observations never fail, so
// racing them yields exactly that. Losing in-flight calls are then interrupted.
const winner = yield* Effect.raceAll(observations)
@@ -2590,7 +3011,11 @@ class Interpreter<R> {
yield* Fiber.interrupt(item.fiber)
}
const winningItem = items[winner.index]
return yield* self.unwrapPromiseExit(winningItem instanceof SandboxPromise ? winningItem : undefined, winner.exit, node)
return yield* self.unwrapPromiseExit(
winningItem instanceof SandboxPromise ? winningItem : undefined,
winner.exit,
node,
)
})
}
}
@@ -2601,7 +3026,7 @@ class Interpreter<R> {
return Effect.suspend(() => {
const savedScopes = self.scopes
self.scopes = [...fn.capturedScopes, new Map<string, Binding>()]
const run = Effect.gen(function*() {
const run = Effect.gen(function* () {
// Seed every parameter name into the scope as a TDZ slot first, so a default that
// references another parameter resolves to that (uninitialized) param rather than
// silently falling through to an outer binding of the same name - matching JS.
@@ -2626,11 +3051,21 @@ class Interpreter<R> {
return yield* self.evaluateExpression(fn.body)
})
return run.pipe(Effect.ensuring(Effect.sync(() => { self.scopes = savedScopes })))
return run.pipe(
Effect.ensuring(
Effect.sync(() => {
self.scopes = savedScopes
}),
),
)
})
}
private invokeIntrinsic(ref: IntrinsicReference, args: Array<unknown>, node: AstNode): Effect.Effect<unknown, unknown, R> {
private invokeIntrinsic(
ref: IntrinsicReference,
args: Array<unknown>,
node: AstNode,
): Effect.Effect<unknown, unknown, R> {
if (typeof ref.receiver === "string") {
return Effect.succeed(invokeStringMethod(ref.receiver, ref.name, args, node))
}
@@ -2657,7 +3092,11 @@ class Interpreter<R> {
// Runs a Map/Set callback (forEach) accepting a user function or a builtin coercion,
// mirroring the array-method callback contract.
private applyCollectionCallback(callback: unknown, name: string, node: AstNode): (args: Array<unknown>) => Effect.Effect<unknown, unknown, R> {
private applyCollectionCallback(
callback: unknown,
name: string,
node: AstNode,
): (args: Array<unknown>) => Effect.Effect<unknown, unknown, R> {
if (!(callback instanceof CodeModeFunction) && !(callback instanceof CoercionFunction)) {
throw new InterpreterRuntimeError(`${name} expects a function callback.`, node)
}
@@ -2667,64 +3106,96 @@ class Interpreter<R> {
: this.invokeFunction(callback, callbackArgs)
}
private invokeMapMethod(target: SandboxMap, name: string, args: Array<unknown>, node: AstNode): Effect.Effect<unknown, unknown, R> {
private invokeMapMethod(
target: SandboxMap,
name: string,
args: Array<unknown>,
node: AstNode,
): Effect.Effect<unknown, unknown, R> {
switch (name) {
case "get": return Effect.succeed(target.map.get(args[0]))
case "has": return Effect.succeed(target.map.has(args[0]))
case "set": return Effect.sync(() => {
target.map.set(args[0], args[1])
return target
})
case "delete": return Effect.sync(() => target.map.delete(args[0]))
case "clear": return Effect.sync(() => {
target.map.clear()
return undefined
})
case "keys": return Effect.sync(() => Array.from(target.map.keys()))
case "values": return Effect.sync(() => Array.from(target.map.values()))
case "entries": return Effect.sync(() => Array.from(target.map.entries(), ([key, item]): Array<unknown> => [key, item]))
case "get":
return Effect.succeed(target.map.get(args[0]))
case "has":
return Effect.succeed(target.map.has(args[0]))
case "set":
return Effect.sync(() => {
target.map.set(args[0], args[1])
return target
})
case "delete":
return Effect.sync(() => target.map.delete(args[0]))
case "clear":
return Effect.sync(() => {
target.map.clear()
return undefined
})
case "keys":
return Effect.sync(() => Array.from(target.map.keys()))
case "values":
return Effect.sync(() => Array.from(target.map.values()))
case "entries":
return Effect.sync(() => Array.from(target.map.entries(), ([key, item]): Array<unknown> => [key, item]))
case "forEach": {
const apply = this.applyCollectionCallback(args[0], "Map.forEach", node)
return Effect.gen(function*() {
return Effect.gen(function* () {
// Snapshot iteration, matching the array-method callback contract.
for (const [key, item] of Array.from(target.map.entries())) yield* apply([item, key, target])
return undefined
})
}
default: throw new InterpreterRuntimeError(`Map method '${name}' is not available in CodeMode.`, node)
default:
throw new InterpreterRuntimeError(`Map method '${name}' is not available in CodeMode.`, node)
}
}
private invokeSetMethod(target: SandboxSet, name: string, args: Array<unknown>, node: AstNode): Effect.Effect<unknown, unknown, R> {
private invokeSetMethod(
target: SandboxSet,
name: string,
args: Array<unknown>,
node: AstNode,
): Effect.Effect<unknown, unknown, R> {
switch (name) {
case "has": return Effect.succeed(target.set.has(args[0]))
case "add": return Effect.sync(() => {
target.set.add(args[0])
return target
})
case "delete": return Effect.sync(() => target.set.delete(args[0]))
case "clear": return Effect.sync(() => {
target.set.clear()
return undefined
})
case "has":
return Effect.succeed(target.set.has(args[0]))
case "add":
return Effect.sync(() => {
target.set.add(args[0])
return target
})
case "delete":
return Effect.sync(() => target.set.delete(args[0]))
case "clear":
return Effect.sync(() => {
target.set.clear()
return undefined
})
case "keys":
case "values": return Effect.sync(() => Array.from(target.set.values()))
case "entries": return Effect.sync(() => Array.from(target.set.values(), (item): Array<unknown> => [item, item]))
case "values":
return Effect.sync(() => Array.from(target.set.values()))
case "entries":
return Effect.sync(() => Array.from(target.set.values(), (item): Array<unknown> => [item, item]))
case "forEach": {
const apply = this.applyCollectionCallback(args[0], "Set.forEach", node)
return Effect.gen(function*() {
return Effect.gen(function* () {
for (const item of Array.from(target.set.values())) yield* apply([item, item, target])
return undefined
})
}
default: throw new InterpreterRuntimeError(`Set method '${name}' is not available in CodeMode.`, node)
default:
throw new InterpreterRuntimeError(`Set method '${name}' is not available in CodeMode.`, node)
}
}
private invokeArrayMethod(target: Array<unknown>, name: string, args: Array<unknown>, node: AstNode): Effect.Effect<unknown, unknown, R> {
private invokeArrayMethod(
target: Array<unknown>,
name: string,
args: Array<unknown>,
node: AstNode,
): Effect.Effect<unknown, unknown, R> {
const optNumber = (value: unknown, label: string): number | undefined => {
if (value === undefined) return undefined
if (typeof value !== "number") throw new InterpreterRuntimeError(`Array.${name} expects ${label} to be a number.`, node)
if (typeof value !== "number")
throw new InterpreterRuntimeError(`Array.${name} expects ${label} to be a number.`, node)
return value
}
switch (name) {
@@ -2733,15 +3204,22 @@ class Interpreter<R> {
throw new InterpreterRuntimeError("Array.join expects zero arguments or one string separator.", node)
}
const input = boundedData(target, "Array.join input") as Array<unknown>
return Effect.succeed(input.map((item) => coerceToString(item ?? "")).join(args.length === 0 ? "," : args[0] as string))
return Effect.succeed(
input.map((item) => coerceToString(item ?? "")).join(args.length === 0 ? "," : (args[0] as string)),
)
}
case "includes":
if (args.length === 0 || args.length > 2) throw new InterpreterRuntimeError("Array.includes expects a value and optional start index.", node)
if (args.length === 0 || args.length > 2)
throw new InterpreterRuntimeError("Array.includes expects a value and optional start index.", node)
return Effect.succeed(target.includes(args[0], optNumber(args[1], "start index")))
case "indexOf":
return Effect.succeed(target.indexOf(args[0], optNumber(args[1], "start index")))
case "lastIndexOf":
return Effect.succeed(args[1] === undefined ? target.lastIndexOf(args[0]) : target.lastIndexOf(args[0], optNumber(args[1], "start index")))
return Effect.succeed(
args[1] === undefined
? target.lastIndexOf(args[0])
: target.lastIndexOf(args[0], optNumber(args[1], "start index")),
)
case "at":
return Effect.succeed(target.at(optNumber(args[0], "index") ?? 0))
case "slice":
@@ -2799,7 +3277,13 @@ class Interpreter<R> {
return Effect.succeed(target.fill(args[0], optNumber(args[1], "start"), optNumber(args[2], "end")))
}
case "copyWithin":
return Effect.succeed(target.copyWithin(optNumber(args[0], "target index") ?? 0, optNumber(args[1], "start") ?? 0, optNumber(args[2], "end")))
return Effect.succeed(
target.copyWithin(
optNumber(args[0], "target index") ?? 0,
optNumber(args[1], "start") ?? 0,
optNumber(args[2], "end"),
),
)
// keys/values/entries return arrays (not iterators), matching the Map/Set convention;
// they work with for...of and spread either way.
case "keys":
@@ -2822,7 +3306,7 @@ class Interpreter<R> {
callback instanceof CoercionFunction
? Effect.succeed(invokeCoercion(callback, callbackArgs, node))
: self.invokeFunction(callback, callbackArgs)
return Effect.gen(function*() {
return Effect.gen(function* () {
// Iterate a snapshot taken at call time so a callback that mutates the array can't
// self-extend the loop - matching JS, where elements appended during iteration are not visited.
const items = target.slice()
@@ -2878,7 +3362,8 @@ class Interpreter<R> {
accumulator = args[1]
start = 0
} else {
if (items.length === 0) throw new InterpreterRuntimeError("Array.reduce of an empty array with no initial value.", node)
if (items.length === 0)
throw new InterpreterRuntimeError("Array.reduce of an empty array with no initial value.", node)
accumulator = items[0]
start = 1
}
@@ -2894,7 +3379,8 @@ class Interpreter<R> {
accumulator = args[1]
start = items.length - 1
} else {
if (items.length === 0) throw new InterpreterRuntimeError("Array.reduceRight of an empty array with no initial value.", node)
if (items.length === 0)
throw new InterpreterRuntimeError("Array.reduceRight of an empty array with no initial value.", node)
accumulator = items[items.length - 1]
start = items.length - 2
}
@@ -2918,7 +3404,11 @@ class Interpreter<R> {
})
}
private sortArray(target: Array<unknown>, comparator: unknown, node: AstNode): Effect.Effect<Array<unknown>, unknown, R> {
private sortArray(
target: Array<unknown>,
comparator: unknown,
node: AstNode,
): Effect.Effect<Array<unknown>, unknown, R> {
if (comparator !== undefined && !(comparator instanceof CodeModeFunction)) {
throw new InterpreterRuntimeError("Array.sort expects an arrow function comparator.", node)
}
@@ -2928,13 +3418,14 @@ class Interpreter<R> {
const left = coerceToString(a)
const right = coerceToString(b)
return left < right ? -1 : left > right ? 1 : 0
}))
}),
)
}
const self = this
const mergeSort = (items: Array<unknown>): Effect.Effect<Array<unknown>, unknown, R> => {
if (items.length <= 1) return Effect.succeed(items)
const midpoint = Math.floor(items.length / 2)
return Effect.gen(function*() {
return Effect.gen(function* () {
const left = yield* mergeSort(items.slice(0, midpoint))
const right = yield* mergeSort(items.slice(midpoint))
const merged: Array<unknown> = []
@@ -2960,7 +3451,7 @@ class Interpreter<R> {
const objectValue: Record<string, unknown> = Object.create(null) as Record<string, unknown>
const properties = getArray(node, "properties")
const self = this
return Effect.gen(function*() {
return Effect.gen(function* () {
for (const propertyValue of properties) {
const property = asNode(propertyValue, "properties")
@@ -2971,10 +3462,15 @@ class Interpreter<R> {
// (Date/RegExp/Map/Set) have no own enumerable properties in JS, so they are no-ops too.
if (spread === null || spread === undefined || isSandboxValue(spread)) continue
if (typeof spread !== "object" || Array.isArray(spread) || isRuntimeReference(spread)) {
throw new InterpreterRuntimeError("Object spread requires a data object in CodeMode.", property, "InvalidDataValue")
throw new InterpreterRuntimeError(
"Object spread requires a data object in CodeMode.",
property,
"InvalidDataValue",
)
}
for (const [key, value] of Object.entries(spread)) {
if (isBlockedMember(key)) throw new InterpreterRuntimeError(`Property '${key}' is not available in CodeMode.`, property)
if (isBlockedMember(key))
throw new InterpreterRuntimeError(`Property '${key}' is not available in CodeMode.`, property)
objectValue[key] = value
}
continue
@@ -3019,7 +3515,7 @@ class Interpreter<R> {
const values: Array<unknown> = []
const self = this
return Effect.gen(function*() {
return Effect.gen(function* () {
for (const elementValue of elements) {
if (elementValue === null) {
values.push(undefined)
@@ -3029,7 +3525,11 @@ class Interpreter<R> {
if (element.type === "SpreadElement") {
const spread = yield* self.evaluateExpression(getNode(element, "argument"))
const items = spreadItems(spread)
if (items === undefined) throw new InterpreterRuntimeError("Array spread requires an array, string, Map, or Set in CodeMode.", element)
if (items === undefined)
throw new InterpreterRuntimeError(
"Array spread requires an array, string, Map, or Set in CodeMode.",
element,
)
values.push(...items)
} else {
values.push(yield* self.evaluateExpression(element))
@@ -3046,7 +3546,7 @@ class Interpreter<R> {
let output = ""
const self = this
return Effect.gen(function*() {
return Effect.gen(function* () {
for (let index = 0; index < quasis.length; index += 1) {
const quasi = asNode(quasis[index], "quasis")
const rawValue = quasi.value
@@ -3071,15 +3571,11 @@ class Interpreter<R> {
private evaluateConditionalExpression(node: AstNode): Effect.Effect<unknown, unknown, R> {
return Effect.flatMap(this.evaluateExpression(getNode(node, "test")), (test) =>
this.evaluateExpression(getNode(node, test ? "consequent" : "alternate")))
this.evaluateExpression(getNode(node, test ? "consequent" : "alternate")),
)
}
private applyCompoundAssignment(
operator: string,
current: unknown,
incoming: unknown,
node: AstNode,
): unknown {
private applyCompoundAssignment(operator: string, current: unknown, incoming: unknown, node: AstNode): unknown {
// `x op= y` is `x = x op y`: dispatch through the shared binary operator implementation
// so compound assignment inherits the same coercion semantics (Dates, data objects, ...).
// Only the arithmetic/bitwise operators are compoundable; logical assignments (&&=/||=/??=)
@@ -3090,13 +3586,26 @@ class Interpreter<R> {
return this.applyBinaryOperator(operator.slice(0, -1), current, incoming, node)
}
private getMemberReference(node: AstNode): Effect.Effect<MemberReference | ToolReference | PromiseMethodReference | IntrinsicReference | GlobalMethodReference | ComputedValue | typeof OptionalShortCircuit | undefined, unknown, R> {
private getMemberReference(
node: AstNode,
): Effect.Effect<
| MemberReference
| ToolReference
| PromiseMethodReference
| IntrinsicReference
| GlobalMethodReference
| ComputedValue
| typeof OptionalShortCircuit
| undefined,
unknown,
R
> {
const objectNode = getNode(node, "object")
const propertyNode = getNode(node, "property")
const computed = getBoolean(node, "computed")
const optional = node.optional === true
const self = this
return Effect.gen(function*() {
return Effect.gen(function* () {
const objectValue = yield* self.evaluateExpression(objectNode)
if (objectValue === OptionalShortCircuit) return OptionalShortCircuit
if ((objectValue === null || objectValue === undefined) && optional) return OptionalShortCircuit
@@ -3126,7 +3635,10 @@ class Interpreter<R> {
if (objectValue instanceof GlobalNamespace) {
if (typeof key !== "string" || isBlockedMember(key)) {
throw new InterpreterRuntimeError(`${objectValue.name}.${String(key)} is not available in CodeMode.`, propertyNode)
throw new InterpreterRuntimeError(
`${objectValue.name}.${String(key)} is not available in CodeMode.`,
propertyNode,
)
}
if (objectValue.name === "Math" && mathConstants.has(key)) {
return new ComputedValue((Math as unknown as Record<string, number>)[key])
@@ -3205,7 +3717,11 @@ class Interpreter<R> {
}
if (isRuntimeReference(objectValue)) {
throw new InterpreterRuntimeError("CodeMode runtime references are opaque and do not expose properties.", objectNode, "InvalidDataValue")
throw new InterpreterRuntimeError(
"CodeMode runtime references are opaque and do not expose properties.",
objectNode,
"InvalidDataValue",
)
}
if (typeof objectValue !== "object" || objectValue === null) {
@@ -3217,7 +3733,12 @@ class Interpreter<R> {
}
if (Array.isArray(objectValue)) {
if (key !== "length" && !(typeof key === "string" && arrayMethods.has(key)) && (typeof key !== "number" && !/^\d+$/.test(key))) {
if (
key !== "length" &&
!(typeof key === "string" && arrayMethods.has(key)) &&
typeof key !== "number" &&
!/^\d+$/.test(key)
) {
// Own non-index properties read through (match results carry index/groups); like JS,
// they are readable in place and dropped by JSON at data boundaries.
if (typeof key === "string" && Object.hasOwn(objectValue, key)) {
@@ -3244,7 +3765,8 @@ class Interpreter<R> {
reference instanceof PromiseMethodReference ||
reference instanceof IntrinsicReference ||
reference instanceof GlobalMethodReference
) return reference
)
return reference
if (Array.isArray(reference.target)) {
if (typeof reference.key === "string" && arrayMethods.has(reference.key)) {
return new IntrinsicReference(reference.target, reference.key)
@@ -3267,7 +3789,7 @@ class Interpreter<R> {
compute: (current: unknown) => Effect.Effect<{ write: boolean; next: unknown; result: unknown }, unknown, R>,
): Effect.Effect<unknown, unknown, R> {
const self = this
return Effect.gen(function*() {
return Effect.gen(function* () {
const reference = yield* self.getMemberReference(node)
if (
reference === OptionalShortCircuit ||
@@ -3281,7 +3803,8 @@ class Interpreter<R> {
throw new InterpreterRuntimeError("Only data fields may be assigned in CodeMode.", node)
}
if (Array.isArray(reference.target)) {
if (reference.key === "length") throw new InterpreterRuntimeError("Array length cannot be assigned in CodeMode.", node)
if (reference.key === "length")
throw new InterpreterRuntimeError("Array length cannot be assigned in CodeMode.", node)
if (typeof reference.key === "string" && arrayMethods.has(reference.key)) {
throw new InterpreterRuntimeError("Array methods cannot be assigned in CodeMode.", node)
}
@@ -3296,8 +3819,15 @@ class Interpreter<R> {
// Rejects inserting a value that (transitively) contains the container it is being inserted
// into - the mutation that would create a circular structure no later walk could survive.
private rejectCircularInsertion(container: object, value: unknown, label: string, node: AstNode, seen = new Set<object>()): void {
if (value === container) throw new InterpreterRuntimeError(`${label} contains a circular value.`, node, "InvalidDataValue")
private rejectCircularInsertion(
container: object,
value: unknown,
label: string,
node: AstNode,
seen = new Set<object>(),
): void {
if (value === container)
throw new InterpreterRuntimeError(`${label} contains a circular value.`, node, "InvalidDataValue")
if (value === null || typeof value !== "object" || isRuntimeReference(value) || seen.has(value)) return
seen.add(value)
const items = Array.isArray(value) ? value : Object.values(value)
@@ -3310,7 +3840,11 @@ class Interpreter<R> {
const target = reference.target
const index = key as number
if (!Number.isInteger(index) || index < 0) {
throw new InterpreterRuntimeError("Array assignment index must be a non-negative integer.", node, "InvalidDataValue")
throw new InterpreterRuntimeError(
"Array assignment index must be a non-negative integer.",
node,
"InvalidDataValue",
)
}
this.rejectCircularInsertion(target, next, "Array assignment result", node)
target[index] = next
@@ -3403,7 +3937,6 @@ class Interpreter<R> {
private popScope(): void {
this.scopes.pop()
}
}
/**
@@ -3426,9 +3959,14 @@ const executeWithLimits = <const Tools extends Record<string, unknown>>(
...(options.onToolCallStart === undefined ? {} : { onToolCallStart: options.onToolCallStart }),
...(options.onToolCallEnd === undefined ? {} : { onToolCallEnd: options.onToolCallEnd }),
}
const tools = ToolRuntime.make((options.tools ?? {}) as HostTools<Services<Tools>>, limits.maxToolCalls, hooks, searchIndex)
const tools = ToolRuntime.make(
(options.tools ?? {}) as HostTools<Services<Tools>>,
limits.maxToolCalls,
hooks,
searchIndex,
)
const logs: Array<string> = []
const logged = () => logs.length > 0 ? { logs: [...logs] } : {}
const logged = () => (logs.length > 0 ? { logs: [...logs] } : {})
if (options.code.trim().length === 0) {
return Effect.succeed({
@@ -3438,7 +3976,7 @@ const executeWithLimits = <const Tools extends Record<string, unknown>>(
})
}
const operation = Effect.gen(function*() {
const operation = Effect.gen(function* () {
const program = parseProgram(options.code)
const interpreter = new Interpreter<Services<Tools>>(tools.invoke, tools.keys, logs)
const value = yield* interpreter.run(program)
@@ -3449,23 +3987,22 @@ const executeWithLimits = <const Tools extends Record<string, unknown>>(
...logged(),
toolCalls: tools.calls,
} satisfies ExecuteResult
}).pipe(
(program) => {
const timeoutMs = limits.timeoutMs
if (timeoutMs === undefined) return program
return program.pipe(
Effect.timeoutOrElse({
duration: timeoutMs,
orElse: () => Effect.succeed({
}).pipe((program) => {
const timeoutMs = limits.timeoutMs
if (timeoutMs === undefined) return program
return program.pipe(
Effect.timeoutOrElse({
duration: timeoutMs,
orElse: () =>
Effect.succeed({
ok: false,
error: { kind: "TimeoutExceeded", message: `Execution timed out after ${timeoutMs}ms.` },
...logged(),
toolCalls: tools.calls,
} satisfies ExecuteResult),
}),
)
},
)
}),
)
})
return operation.pipe(
Effect.catchCause((cause) =>
@@ -3478,7 +4015,7 @@ const executeWithLimits = <const Tools extends Record<string, unknown>>(
toolCalls: tools.calls,
} satisfies ExecuteResult),
),
Effect.map((result) => limits.maxOutputBytes === undefined ? result : boundOutput(result, limits.maxOutputBytes)),
Effect.map((result) => (limits.maxOutputBytes === undefined ? result : boundOutput(result, limits.maxOutputBytes))),
)
}
@@ -3540,7 +4077,9 @@ const boundOutput = (result: ExecuteResult, maxOutputBytes: number): ExecuteResu
: { ok: false, error: result.error, ...logsPart, truncated: true, toolCalls: result.toolCalls }
}
export const execute = <const Tools extends Record<string, unknown>>(options: ExecuteOptions<Tools>): Effect.Effect<ExecuteResult, never, Services<Tools>> => {
export const execute = <const Tools extends Record<string, unknown>>(
options: ExecuteOptions<Tools>,
): Effect.Effect<ExecuteResult, never, Services<Tools>> => {
const tools = (options.tools ?? {}) as HostTools<Services<Tools>>
ToolRuntime.assertValidTools(tools)
return executeWithLimits(options, resolveExecutionLimits(options.limits), ToolRuntime.searchIndex(tools))
@@ -3558,7 +4097,9 @@ export const execute = <const Tools extends Record<string, unknown>>(options: Ex
* const code = runtime.agentTool()
* ```
*/
export const make = <const Tools extends Record<string, unknown> = {}>(options: CodeModeOptions<Tools> = {} as CodeModeOptions<Tools>): CodeModeRuntime<Services<Tools>> => {
export const make = <const Tools extends Record<string, unknown> = {}>(
options: CodeModeOptions<Tools> = {} as CodeModeOptions<Tools>,
): CodeModeRuntime<Services<Tools>> => {
const tools = (options.tools ?? {}) as HostTools<Services<Tools>>
ToolRuntime.assertValidTools(tools)
const limits = resolveExecutionLimits(options.limits)
+1 -7
View File
@@ -1,10 +1,4 @@
export {
ToolError,
CodeMode,
ExecuteInputSchema,
ExecuteResultSchema,
toolError,
} from "./codemode.js"
export { ToolError, CodeMode, ExecuteInputSchema, ExecuteResultSchema, toolError } from "./codemode.js"
export { Tool } from "./tool.js"
export type { Definition as ToolDefinition, JsonSchema, ToolSchema } from "./tool.js"
export type { ToolCallEnded, ToolCallHooks } from "./tool-runtime.js"
+153 -70
View File
@@ -21,11 +21,16 @@ export type HostTools<R = never> = {
export type Services<Tools> = Tools extends (...args: Array<unknown>) => Effect.Effect<unknown, unknown, infer R>
? R
: Tools extends { readonly _tag: "CodeModeTool"; readonly run: (input: unknown) => Effect.Effect<unknown, unknown, infer R> }
: Tools extends {
readonly _tag: "CodeModeTool"
readonly run: (input: unknown) => Effect.Effect<unknown, unknown, infer R>
}
? R
: Tools extends object
? string extends keyof Tools ? never : Services<Tools[keyof Tools]>
: never
: Tools extends object
? string extends keyof Tools
? never
: Services<Tools[keyof Tools]>
: never
/** Minimal audit record retained for each admitted tool call. */
export type ToolCall = {
@@ -68,11 +73,13 @@ export type SafeObject = Record<string, unknown>
const reservedNamespace = "$codemode"
const defaultMaxInlineCatalogTokens = 2_000
const defaultSearchLimit = 10
const searchSignature = "tools.$codemode.search({ query?: string, namespace?: string, limit?: number }): Promise<{ items: Array<{ path: string; description: string; signature: string }>; total: number }>"
const searchSignature =
"tools.$codemode.search({ query?: string, namespace?: string, limit?: number }): Promise<{ items: Array<{ path: string; description: string; signature: string }>; total: number }>"
const toolExpression = (path: string) =>
"tools" + path
"tools" +
path
.split(".")
.map((segment) => identifierSegment.test(segment) ? `.${segment}` : `[${JSON.stringify(segment)}]`)
.map((segment) => (identifierSegment.test(segment) ? `.${segment}` : `[${JSON.stringify(segment)}]`))
.join("")
export class ToolReference {
@@ -88,7 +95,12 @@ const MAX_VALUE_DEPTH = 32
export class ToolRuntimeError extends Error {
constructor(
readonly kind: "UnknownTool" | "InvalidToolInput" | "InvalidToolOutput" | "InvalidDataValue" | "ToolCallLimitExceeded",
readonly kind:
| "UnknownTool"
| "InvalidToolInput"
| "InvalidToolOutput"
| "InvalidDataValue"
| "ToolCallLimitExceeded",
message: string,
readonly suggestions: ReadonlyArray<string> = [],
) {
@@ -131,7 +143,13 @@ export const isBlockedMember = (name: string): boolean => blockedMemberNames.has
export const copyIn = (value: unknown, label: string, preserveSandboxValues = false): unknown =>
copyBounded(value, label, 0, new Set(), preserveSandboxValues)
const copyBounded = (value: unknown, label: string, depth: number, seen: Set<object>, preserveSandboxValues: boolean): unknown => {
const copyBounded = (
value: unknown,
label: string,
depth: number,
seen: Set<object>,
preserveSandboxValues: boolean,
): unknown => {
if (depth > MAX_VALUE_DEPTH) {
throw new ToolRuntimeError("InvalidDataValue", `${label} exceeds the maximum value depth of ${MAX_VALUE_DEPTH}.`)
}
@@ -166,7 +184,12 @@ const copyBounded = (value: unknown, label: string, depth: number, seen: Set<obj
// Intra-sandbox checkpoints keep sandbox value instances alive as leaves; their contents
// are never walked here (Map/Set members are validated where mutation happens, and the
// real boundary still serializes them below).
if (value instanceof SandboxDate || value instanceof SandboxRegExp || value instanceof SandboxMap || value instanceof SandboxSet) {
if (
value instanceof SandboxDate ||
value instanceof SandboxRegExp ||
value instanceof SandboxMap ||
value instanceof SandboxSet
) {
return value
}
// Host instances cannot normally reach an intra-sandbox checkpoint (tool results cross
@@ -197,8 +220,12 @@ const copyBounded = (value: unknown, label: string, depth: number, seen: Set<obj
return Number.isFinite(value.getTime()) ? value.toISOString() : null
}
if (
value instanceof SandboxRegExp || value instanceof SandboxMap || value instanceof SandboxSet ||
value instanceof RegExp || value instanceof Map || value instanceof Set
value instanceof SandboxRegExp ||
value instanceof SandboxMap ||
value instanceof SandboxSet ||
value instanceof RegExp ||
value instanceof Map ||
value instanceof Set
) {
return Object.create(null) as SafeObject
}
@@ -250,7 +277,10 @@ export const copyOut = (value: unknown, undefinedAsNull = false): unknown => {
return value
}
const definitions = <R>(tools: HostTools<R>, path: ReadonlyArray<string> = []): Array<{ path: string; definition: Definition<R> }> => {
const definitions = <R>(
tools: HostTools<R>,
path: ReadonlyArray<string> = [],
): Array<{ path: string; definition: Definition<R> }> => {
const entries: Array<{ path: string; definition: Definition<R> }> = []
for (const [name, value] of Object.entries(tools)) {
const next = [...path, name]
@@ -342,8 +372,11 @@ const toSearchEntry = <R>(path: string, definition: Definition<R>, description:
path,
definition.description,
...inputProperties(definition).flatMap(({ name, description: property }) =>
property === undefined ? [name] : [name, property]),
].join("\n").toLowerCase(),
property === undefined ? [name] : [name, property],
),
]
.join("\n")
.toLowerCase(),
})
/** The runtime search index over every described tool. Search is always registered. */
@@ -396,7 +429,8 @@ export const discoveryPlan = <R>(
namespace,
picked: new Set<ToolDescription>(),
queue: [...group].sort(
(left, right) => estimate(catalogLine(left)) - estimate(catalogLine(right)) || left.path.localeCompare(right.path),
(left, right) =>
estimate(catalogLine(left)) - estimate(catalogLine(right)) || left.path.localeCompare(right.path),
),
}))
let used = 0
@@ -472,7 +506,9 @@ export const discoveryPlan = <R>(
"- A result typed `Promise<unknown>` has no guaranteed shape - verify what actually came back before relying on its fields.",
"- Run independent calls in parallel: `await Promise.all(items.map((item) => tools.<namespace>.<tool>(item)))`.",
"- `Object.keys(tools)` lists namespaces; `Object.keys(tools.<namespace>)` lists its tools; `for...in` works on both.",
...(complete ? [] : ['- Browse one namespace: `await tools.$codemode.search({ query: "", namespace: "<name>" })`.']),
...(complete
? []
: ['- Browse one namespace: `await tools.$codemode.search({ query: "", namespace: "<name>" })`.']),
]
const syntax = [
@@ -500,26 +536,21 @@ export const discoveryPlan = <R>(
const count = `${group.length} tool${group.length === 1 ? "" : "s"}`
// Annotate only when a namespace is not fully shown, so a comprehensive
// namespace reads cleanly and a truncated one is unambiguous.
const label = picked.size === group.length ? count : picked.size === 0 ? `${count}, none shown` : `${count}, ${picked.size} shown`
const label =
picked.size === group.length
? count
: picked.size === 0
? `${count}, none shown`
: `${count}, ${picked.size} shown`
toolSection.push(`- ${namespace} (${label})`)
for (const tool of group) if (picked.has(tool)) toolSection.push(catalogLine(tool))
}
if (!complete) {
toolSection.push(
"",
"Search returns complete callable signatures:",
`- ${searchSignature}`,
)
toolSection.push("", "Search returns complete callable signatures:", `- ${searchSignature}`)
}
}
const lines = [
...intro,
...workflow,
...rules,
...syntax,
...toolSection,
]
const lines = [...intro, ...workflow, ...rules, ...syntax, ...toolSection]
return {
catalog: described,
instructions: lines.join("\n"),
@@ -534,18 +565,29 @@ export const discoveryPlan = <R>(
* function in JS). An unknown path is an `UnknownTool` error pointing at the working
* discovery idioms, mirroring how calling an unknown tool fails.
*/
const namespaceKeys = <R>(tools: HostTools<R>, path: ReadonlyArray<string>, searchEnabled: boolean): ReadonlyArray<string> => {
const namespaceKeys = <R>(
tools: HostTools<R>,
path: ReadonlyArray<string>,
searchEnabled: boolean,
): ReadonlyArray<string> => {
// The reserved discovery namespace is virtual (never present in the host tree); enumerate
// it explicitly so `Object.keys(tools.$codemode)` matches the callable surface.
if (searchEnabled && path.length === 1 && path[0] === reservedNamespace) return ["search"]
let value: HostTool<R> | Definition<R> | HostTools<R> = tools
for (const segment of path) {
if (isBlockedMember(segment) || typeof value === "function" || isDefinition(value) || !Object.hasOwn(value, segment)) {
if (
isBlockedMember(segment) ||
typeof value === "function" ||
isDefinition(value) ||
!Object.hasOwn(value, segment)
) {
throw new ToolRuntimeError(
"UnknownTool",
`Unknown tool namespace '${path.join(".")}'.`,
searchEnabled
? ["Object.keys(tools) lists the available namespaces; tools.$codemode.search({ query }) finds described tools."]
? [
"Object.keys(tools) lists the available namespaces; tools.$codemode.search({ query }) finds described tools.",
]
: ["Object.keys(tools) lists the available namespaces."],
)
}
@@ -555,12 +597,25 @@ const namespaceKeys = <R>(tools: HostTools<R>, path: ReadonlyArray<string>, sear
return Object.keys(value)
}
const resolve = <R>(tools: HostTools<R>, path: ReadonlyArray<string>, searchEnabled: boolean): HostTool<R> | Definition<R> => {
const resolve = <R>(
tools: HostTools<R>,
path: ReadonlyArray<string>,
searchEnabled: boolean,
): HostTool<R> | Definition<R> => {
let value: HostTool<R> | Definition<R> | HostTools<R> = tools
for (const segment of path) {
if (isBlockedMember(segment) || typeof value === "function" || isDefinition(value) || !Object.hasOwn(value, segment)) {
throw new ToolRuntimeError("UnknownTool", `Unknown tool '${path.join(".")}'.`, searchEnabled ? ["Use tools.$codemode.search({ query }) to find available described tools."] : [])
if (
isBlockedMember(segment) ||
typeof value === "function" ||
isDefinition(value) ||
!Object.hasOwn(value, segment)
) {
throw new ToolRuntimeError(
"UnknownTool",
`Unknown tool '${path.join(".")}'.`,
searchEnabled ? ["Use tools.$codemode.search({ query }) to find available described tools."] : [],
)
}
value = value[segment] as HostTool<R> | Definition<R> | HostTools<R>
}
@@ -602,7 +657,8 @@ export const make = <R>(
return effect.pipe(
Effect.tap(() => onEnd({ ...call, durationMs: Date.now() - startedAt, outcome: "success" })),
Effect.tapError((error) =>
onEnd({ ...call, durationMs: Date.now() - startedAt, outcome: "failure", message: failureMessage(error) })),
onEnd({ ...call, durationMs: Date.now() - startedAt, outcome: "failure", message: failureMessage(error) }),
),
)
}
@@ -624,7 +680,7 @@ export const make = <R>(
calls,
keys: (path) => namespaceKeys(tools, path, searchEnabled),
invoke: (path, args) =>
Effect.gen(function*() {
Effect.gen(function* () {
const name = path.join(".")
const externalArgs = args.map((arg) => copyOut(copyIn(arg, `Arguments for tool '${name}'`)))
const call = { name }
@@ -637,17 +693,32 @@ export const make = <R>(
if (!searchEnabled) throw new ToolRuntimeError("UnknownTool", `Unknown tool '${name}'.`)
const input = externalArgs[0]
if (externalArgs.length !== 1 || input === null || typeof input !== "object" || Array.isArray(input)) {
throw new ToolRuntimeError("InvalidToolInput", "tools.$codemode.search expects { query?: string; namespace?: string; limit?: number }.")
throw new ToolRuntimeError(
"InvalidToolInput",
"tools.$codemode.search expects { query?: string; namespace?: string; limit?: number }.",
)
}
const request = input as { query?: unknown; namespace?: unknown; limit?: unknown }
if (request.query !== undefined && typeof request.query !== "string") {
throw new ToolRuntimeError("InvalidToolInput", "tools.$codemode.search query must be a string when provided.")
throw new ToolRuntimeError(
"InvalidToolInput",
"tools.$codemode.search query must be a string when provided.",
)
}
if (request.namespace !== undefined && typeof request.namespace !== "string") {
throw new ToolRuntimeError("InvalidToolInput", "tools.$codemode.search namespace must be a string when provided.")
throw new ToolRuntimeError(
"InvalidToolInput",
"tools.$codemode.search namespace must be a string when provided.",
)
}
if (request.limit !== undefined && (typeof request.limit !== "number" || !Number.isSafeInteger(request.limit) || request.limit <= 0)) {
throw new ToolRuntimeError("InvalidToolInput", "tools.$codemode.search limit must be a positive safe integer when provided.")
if (
request.limit !== undefined &&
(typeof request.limit !== "number" || !Number.isSafeInteger(request.limit) || request.limit <= 0)
) {
throw new ToolRuntimeError(
"InvalidToolInput",
"tools.$codemode.search limit must be a positive safe integer when provided.",
)
}
const query = typeof request.query === "string" ? request.query : ""
const namespace = typeof request.namespace === "string" ? request.namespace : undefined
@@ -656,40 +727,50 @@ export const make = <R>(
Effect.try({
try: () => {
const limit = typeof request.limit === "number" ? request.limit : defaultSearchLimit
const scoped = namespace === undefined ? searchIndex : searchIndex.filter((entry) => entry.namespace === namespace)
const scoped =
namespace === undefined ? searchIndex : searchIndex.filter((entry) => entry.namespace === namespace)
// A query that names one tool path exactly (canonical path or rendered
// JavaScript expression) is a lookup, not a search: return that tool alone.
const trimmed = query.trim()
const pathQuery = trimmed.startsWith("tools.") ? trimmed.slice("tools.".length) : trimmed
const exact = pathQuery === "" ? undefined : scoped.find((entry) =>
entry.description.path === pathQuery || toolExpression(entry.description.path) === trimmed)
const exact =
pathQuery === ""
? undefined
: scoped.find(
(entry) =>
entry.description.path === pathQuery || toolExpression(entry.description.path) === trimmed,
)
const terms = tokenize(query).map(termForms)
// Additive field-weighted scoring, summed across terms: exact path or path
// segment (20) > path substring (8) > description substring (4) > any
// searchable text, incl. input parameter names/descriptions (2). Each term
// matches a field when any of its forms (the term or a singular variant)
// does. An empty query browses everything, alphabetical by path.
const ranked = exact !== undefined
? [exact]
: scoped
.map((entry) => {
const path = entry.description.path.toLowerCase()
const description = entry.description.description.toLowerCase()
const score = terms.reduce(
(total, forms) =>
total +
(forms.some((form) => path === form || path.endsWith(`.${form}`)) ? 20 : 0) +
(forms.some((form) => path.includes(form)) ? 8 : 0) +
(forms.some((form) => description.includes(form)) ? 4 : 0) +
(forms.some((form) => entry.searchText.includes(form)) ? 2 : 0),
0,
const ranked =
exact !== undefined
? [exact]
: scoped
.map((entry) => {
const path = entry.description.path.toLowerCase()
const description = entry.description.description.toLowerCase()
const score = terms.reduce(
(total, forms) =>
total +
(forms.some((form) => path === form || path.endsWith(`.${form}`)) ? 20 : 0) +
(forms.some((form) => path.includes(form)) ? 8 : 0) +
(forms.some((form) => description.includes(form)) ? 4 : 0) +
(forms.some((form) => entry.searchText.includes(form)) ? 2 : 0),
0,
)
return { entry, score }
})
.filter(({ score }) => terms.length === 0 || score > 0)
.sort(
(left, right) =>
right.score - left.score ||
left.entry.description.path.localeCompare(right.entry.description.path),
)
return { entry, score }
})
.filter(({ score }) => terms.length === 0 || score > 0)
.sort((left, right) =>
right.score - left.score || left.entry.description.path.localeCompare(right.entry.description.path))
.map(({ entry }) => entry)
.map(({ entry }) => entry)
// Result paths are rendered as JavaScript expressions so each `path` is
// directly usable as the call site (`await tools.github.list({ ... })` or
// `await tools.ns["dashed-name"]({ ... })`). The signature is the pretty,
@@ -711,10 +792,12 @@ export const make = <R>(
const tool = resolve(tools, path, searchEnabled)
let describedInput: unknown
if (isDefinition(tool)) {
if (externalArgs.length !== 1) throw new ToolRuntimeError("InvalidToolInput", `Tool '${name}' expects exactly one input object.`)
if (externalArgs.length !== 1)
throw new ToolRuntimeError("InvalidToolInput", `Tool '${name}' expects exactly one input object.`)
describedInput = yield* Effect.try({
try: () => decodeToolInput(tool, externalArgs[0]),
catch: (cause) => new ToolRuntimeError("InvalidToolInput", `Invalid input for tool '${name}': ${String(cause)}`),
catch: (cause) =>
new ToolRuntimeError("InvalidToolInput", `Invalid input for tool '${name}': ${String(cause)}`),
})
}
const input = isDefinition(tool) ? describedInput : externalArgs
@@ -722,7 +805,7 @@ export const make = <R>(
const currentCall = { index, name, input }
if (isDefinition(tool)) {
return yield* observeEnd(
Effect.gen(function*() {
Effect.gen(function* () {
const raw = yield* runHost(Effect.suspend(() => tool.run(describedInput)))
const result = yield* Effect.try({
try: () => decodeToolOutput(tool, raw),
@@ -734,7 +817,7 @@ export const make = <R>(
)
}
return yield* observeEnd(
Effect.gen(function*() {
Effect.gen(function* () {
return yield* decodeOutput(yield* runHost(Effect.suspend(() => tool(...externalArgs))), name)
}),
currentCall,
+27 -12
View File
@@ -57,8 +57,7 @@ export type Options<I extends ToolSchema, O extends ToolSchema | undefined, R =
export const isDefinition = <R = never>(value: unknown): value is Definition<R> =>
typeof value === "object" && value !== null && "_tag" in value && value._tag === "CodeModeTool"
const isEffectSchema = (schema: ToolSchema): schema is Schema.Decoder<unknown> & Schema.Top =>
Schema.isSchema(schema)
const isEffectSchema = (schema: ToolSchema): schema is Schema.Decoder<unknown> & Schema.Top => Schema.isSchema(schema)
const renderLiteral = (value: unknown): string => JSON.stringify(value) ?? "unknown"
@@ -69,10 +68,12 @@ const renderLiteral = (value: unknown): string => JSON.stringify(value) ?? "unkn
export const identifierSegment = /^[A-Za-z_$][A-Za-z0-9_$]*$/
/** Renders a property name as a valid TS object key: bare when an identifier, quoted otherwise. */
const renderKey = (name: string): string => identifierSegment.test(name) ? name : JSON.stringify(name)
const renderKey = (name: string): string => (identifierSegment.test(name) ? name : JSON.stringify(name))
const effectNumberSentinel = (schema: JsonSchema) =>
schema.type === "string" && Array.isArray(schema.enum) && schema.enum.length === 1 &&
schema.type === "string" &&
Array.isArray(schema.enum) &&
schema.enum.length === 1 &&
(schema.enum[0] === "NaN" || schema.enum[0] === "Infinity" || schema.enum[0] === "-Infinity")
/**
@@ -118,7 +119,8 @@ const docTags = (schema: JsonSchema): Array<string> => {
*/
const jsdoc = (description: string | undefined, tags: ReadonlyArray<string>, pad: string): string => {
const lines = [...(description === undefined ? [] : description.split("\n")), ...tags].map((line) =>
line.replaceAll("*/", "* /").replace(/\s+$/, ""))
line.replaceAll("*/", "* /").replace(/\s+$/, ""),
)
while (lines.length > 0 && lines[0]!.trim() === "") lines.shift()
while (lines.length > 0 && lines[lines.length - 1]!.trim() === "") lines.pop()
if (lines.length === 0) return ""
@@ -127,7 +129,12 @@ const jsdoc = (description: string | undefined, tags: ReadonlyArray<string>, pad
return `${pad}/**\n${body}\n${pad} */\n`
}
const renderSchema = (schema: JsonSchema, ctx: RenderContext, depth = 0, seen: ReadonlySet<string> = new Set()): string => {
const renderSchema = (
schema: JsonSchema,
ctx: RenderContext,
depth = 0,
seen: ReadonlySet<string> = new Set(),
): string => {
if (depth > MAX_RENDER_DEPTH) return "unknown"
if (schema.$ref) {
const name = schema.$ref.split("/").pop()
@@ -146,13 +153,16 @@ const renderSchema = (schema: JsonSchema, ctx: RenderContext, depth = 0, seen: R
if (
alternatives.some((item) => item.type === "number") &&
alternatives.every((item) => item.type === "number" || effectNumberSentinel(item))
) return "number"
)
return "number"
// An empty Schema.Struct({}) emits `anyOf: [{ type: "object" }, { type: "array" }]`
// (no properties/items); render the bare shape as {} instead of `{} | Array<unknown>`.
if (
alternatives.length === 2 &&
alternatives[0]?.type === "object" && alternatives[0].properties === undefined &&
alternatives[1]?.type === "array" && alternatives[1].items === undefined
alternatives[0]?.type === "object" &&
alternatives[0].properties === undefined &&
alternatives[1]?.type === "array" &&
alternatives[1].items === undefined
) {
return "{}"
}
@@ -170,7 +180,8 @@ const renderSchema = (schema: JsonSchema, ctx: RenderContext, depth = 0, seen: R
const required = new Set(schema.required ?? [])
const properties = Object.entries(schema.properties ?? {})
const additional = schema.additionalProperties
const indexType = additional && typeof additional === "object" ? renderSchema(additional, ctx, depth + 1, seen) : undefined
const indexType =
additional && typeof additional === "object" ? renderSchema(additional, ctx, depth + 1, seen) : undefined
const field = ([name, value]: readonly [string, JsonSchema]) =>
`${renderKey(name)}${required.has(name) ? "" : "?"}: ${renderSchema(value, ctx, depth + 1, seen)}`
@@ -183,7 +194,9 @@ const renderSchema = (schema: JsonSchema, ctx: RenderContext, depth = 0, seen: R
// Pretty: an indented block, each described field preceded by its JSDoc comment.
if (properties.length === 0 && indexType === undefined) return "{}"
const pad = " ".repeat(depth + 1)
const lines = properties.map((entry) => `${jsdoc(entry[1].description, docTags(entry[1]), pad)}${pad}${field(entry)}`)
const lines = properties.map(
(entry) => `${jsdoc(entry[1].description, docTags(entry[1]), pad)}${pad}${field(entry)}`,
)
if (indexType !== undefined) lines.push(`${pad}[key: string]: ${indexType}`)
return `{\n${lines.join("\n")}\n${" ".repeat(depth)}}`
}
@@ -262,7 +275,9 @@ export const inputProperties = <R>(definition: Definition<R>): Array<InputProper
* fields; the default stays the compact single-line form.
*/
export const inputTypeScript = <R>(definition: Definition<R>, pretty = false): string =>
isEffectSchema(definition.input) ? toTypeScript(definition.input, false, pretty) : jsonSchemaToTypeScript(definition.input, pretty)
isEffectSchema(definition.input)
? toTypeScript(definition.input, false, pretty)
: jsonSchemaToTypeScript(definition.input, pretty)
/**
* The model-visible TypeScript type of a tool's result; tools without an output schema
+4 -1
View File
@@ -28,4 +28,7 @@ export class SandboxSet {
}
export const isSandboxValue = (value: unknown): value is SandboxDate | SandboxRegExp | SandboxMap | SandboxSet =>
value instanceof SandboxDate || value instanceof SandboxRegExp || value instanceof SandboxMap || value instanceof SandboxSet
value instanceof SandboxDate ||
value instanceof SandboxRegExp ||
value instanceof SandboxMap ||
value instanceof SandboxSet
+267 -172
View File
@@ -1,6 +1,13 @@
import { describe, expect, test } from "bun:test"
import { Cause, Effect, Schema } from "effect"
import { CodeMode, ExecuteInputSchema, ExecuteResultSchema, Tool, toolError, type ExecutionLimits } from "../src/index.js"
import {
CodeMode,
ExecuteInputSchema,
ExecuteResultSchema,
Tool,
toolError,
type ExecutionLimits,
} from "../src/index.js"
import type { Definition } from "../src/tool.js"
const run = (tool: Definition<never>) =>
@@ -75,11 +82,14 @@ describe("CodeMode host failure boundary", () => {
output: Schema.Unknown,
run: () =>
Effect.succeed(
new Proxy({}, {
ownKeys: () => {
throw new Error("host-output-secret")
new Proxy(
{},
{
ownKeys: () => {
throw new Error("host-output-secret")
},
},
}),
),
),
}),
)
@@ -162,9 +172,7 @@ describe("CodeMode tool-call observation", () => {
)
expect(result.ok).toBe(true)
expect(calls).toStrictEqual([
{ index: 0, name: "context.lookup", input: { query: "deployment failure" } },
])
expect(calls).toStrictEqual([{ index: 0, name: "context.lookup", input: { query: "deployment failure" } }])
})
test("observes settled calls with outcome and duration", async () => {
@@ -173,25 +181,26 @@ describe("CodeMode tool-call observation", () => {
description: "Look up a value",
input: Schema.Struct({ query: Schema.String }),
output: Schema.String,
run: ({ query }) =>
query === "boom" ? Effect.fail(toolError("Lookup refused")) : Effect.succeed(query),
run: ({ query }) => (query === "boom" ? Effect.fail(toolError("Lookup refused")) : Effect.succeed(query)),
})
const runtime = CodeMode.make({
tools: { context: { lookup } },
onToolCallStart: (call) => Effect.sync(() => {
events.push({ phase: "start", index: call.index, name: call.name })
}),
onToolCallEnd: (call) => Effect.sync(() => {
expect(call.durationMs).toBeGreaterThanOrEqual(0)
events.push({
phase: "end",
index: call.index,
name: call.name,
outcome: call.outcome,
...(call.message === undefined ? {} : { message: call.message }),
})
}),
onToolCallStart: (call) =>
Effect.sync(() => {
events.push({ phase: "start", index: call.index, name: call.name })
}),
onToolCallEnd: (call) =>
Effect.sync(() => {
expect(call.durationMs).toBeGreaterThanOrEqual(0)
events.push({
phase: "end",
index: call.index,
name: call.name,
outcome: call.outcome,
...(call.message === undefined ? {} : { message: call.message }),
})
}),
})
const success = await Effect.runPromise(runtime.execute(`return await tools.context.lookup({ query: "ok" })`))
@@ -210,13 +219,15 @@ describe("CodeMode tool-call observation", () => {
describe("CodeMode console capture", () => {
test("captures console output as bounded result logs", async () => {
const result = await Effect.runPromise(CodeMode.execute({
code: `
const result = await Effect.runPromise(
CodeMode.execute({
code: `
const returned = console.log("Thread info:", { name: "Demo", count: 2 })
console.warn("careful")
return returned
`,
}))
}),
)
expect(result).toStrictEqual({
ok: true,
@@ -228,39 +239,45 @@ describe("CodeMode console capture", () => {
})
test("keeps logs captured before failures", async () => {
const result = await Effect.runPromise(CodeMode.execute({
code: `
const result = await Effect.runPromise(
CodeMode.execute({
code: `
console.log("before failure")
throw new Error("boom")
`,
}))
}),
)
expect(result.ok ? undefined : result.logs).toStrictEqual(["before failure"])
expect(result.ok ? undefined : result.error.message).toBe("Uncaught: boom")
})
test("prints NaN and Infinity literally instead of the JSON null", async () => {
const result = await Effect.runPromise(CodeMode.execute({
code: `
const result = await Effect.runPromise(
CodeMode.execute({
code: `
console.log(NaN)
console.log(Infinity, -Infinity)
console.log({ ratio: NaN, bounds: [Infinity] })
return null
`,
}))
}),
)
expect(result.ok).toBe(true)
expect(result.logs).toStrictEqual(["NaN", "Infinity -Infinity", '{"ratio":NaN,"bounds":[Infinity]}'])
})
test("renders sandbox values nested inside logged containers", async () => {
const result = await Effect.runPromise(CodeMode.execute({
code: `
const result = await Effect.runPromise(
CodeMode.execute({
code: `
console.log({ m: new Map([["a", 1]]), when: new Date(0), r: /ab/g, s: new Set([1, 2]) })
console.log([new Date(0)])
return null
`,
}))
}),
)
expect(result.ok).toBe(true)
expect(result.logs).toStrictEqual([
@@ -270,38 +287,40 @@ describe("CodeMode console capture", () => {
})
test("console formatting is total: cycles and opaque references render as markers", async () => {
const result = await Effect.runPromise(CodeMode.execute({
code: `
const result = await Effect.runPromise(
CodeMode.execute({
code: `
const m = new Map()
m.set("self", m)
console.log({ box: m })
console.log({ fn: (x) => x, ok: 1 })
return null
`,
}))
}),
)
expect(result.ok).toBe(true)
expect(result.logs).toStrictEqual([
'{"box":Map(1) [["self",[Circular]]]}',
'{"fn":[CodeMode reference],"ok":1}',
])
expect(result.logs).toStrictEqual(['{"box":Map(1) [["self",[Circular]]]}', '{"fn":[CodeMode reference],"ok":1}'])
})
test("console.table renders sandbox value cells", async () => {
const result = await Effect.runPromise(CodeMode.execute({
code: `
const result = await Effect.runPromise(
CodeMode.execute({
code: `
console.table([{ when: new Date(0), n: NaN }])
return null
`,
}))
}),
)
expect(result.ok).toBe(true)
expect(result.logs).toStrictEqual(["(index)\twhen\tn\n0\t1970-01-01T00:00:00.000Z\tNaN"])
})
test("captures console.dir and console.table output", async () => {
const result = await Effect.runPromise(CodeMode.execute({
code: `
const result = await Effect.runPromise(
CodeMode.execute({
code: `
console.dir({ nested: { ok: true } })
console.table([
{ name: "Kit", count: 1, hidden: "x" },
@@ -309,15 +328,13 @@ describe("CodeMode console capture", () => {
], ["name", "count"])
return "done"
`,
}))
}),
)
expect(result).toStrictEqual({
ok: true,
value: "done",
logs: [
'{"nested":{"ok":true}}',
"(index)\tname\tcount\n0\tKit\t1\n1\tOlive\t2",
],
logs: ['{"nested":{"ok":true}}', "(index)\tname\tcount\n0\tKit\t1\n1\tOlive\t2"],
toolCalls: [],
})
})
@@ -325,9 +342,11 @@ describe("CodeMode console capture", () => {
describe("CodeMode output budget", () => {
test("absent maxOutputBytes means no truncation at all", async () => {
const result = await Effect.runPromise(CodeMode.execute({
code: `console.log("z".repeat(50_000)); return "x".repeat(100_000)`,
}))
const result = await Effect.runPromise(
CodeMode.execute({
code: `console.log("z".repeat(50_000)); return "x".repeat(100_000)`,
}),
)
expect(result.ok).toBe(true)
if (!result.ok) return
@@ -338,29 +357,35 @@ describe("CodeMode output budget", () => {
test("truncates an oversized result value with a marker instead of failing", async () => {
const limits: ExecutionLimits = { maxOutputBytes: 40 }
const result = await Effect.runPromise(CodeMode.execute({
code: `return { data: "${"x".repeat(200)}" }`,
limits,
}))
const result = await Effect.runPromise(
CodeMode.execute({
code: `return { data: "${"x".repeat(200)}" }`,
limits,
}),
)
expect(result.ok).toBe(true)
if (!result.ok) return
expect(result.truncated).toBe(true)
expect(typeof result.value).toBe("string")
expect(result.value).toMatch(/^\{"data":"x+ \[result truncated: \d+ bytes exceeds the 40-byte output limit; return a smaller value\]$/)
expect(result.value).toMatch(
/^\{"data":"x+ \[result truncated: \d+ bytes exceeds the 40-byte output limit; return a smaller value\]$/,
)
expect(Schema.decodeUnknownSync(ExecuteResultSchema)(JSON.parse(JSON.stringify(result)))).toStrictEqual(result)
})
test("keeps leading logs within the remaining budget and marks the cut", async () => {
const limits: ExecutionLimits = { maxOutputBytes: 40 }
const result = await Effect.runPromise(CodeMode.execute({
code: `
const result = await Effect.runPromise(
CodeMode.execute({
code: `
console.log("first line")
console.log("${"y".repeat(200)}")
return "ok"
`,
limits,
}))
limits,
}),
)
expect(result.ok).toBe(true)
if (!result.ok) return
@@ -370,12 +395,14 @@ describe("CodeMode output budget", () => {
})
test("does not mark results within the budget", async () => {
const result = await Effect.runPromise(CodeMode.execute({
code: `
const result = await Effect.runPromise(
CodeMode.execute({
code: `
console.log("fits")
return { fits: true }
`,
}))
}),
)
expect(result).toStrictEqual({
ok: true,
value: { fits: true },
@@ -395,18 +422,21 @@ describe("CodeMode schema flexibility", () => {
properties: { id: { type: "string" }, count: { type: "number" } },
required: ["id"],
},
run: (input) => Effect.sync(() => {
observed.push(input)
return { echoed: input }
}),
run: (input) =>
Effect.sync(() => {
observed.push(input)
return { echoed: input }
}),
})
const runtime = CodeMode.make({ tools: { adapter: { call } } })
expect(runtime.catalog()).toStrictEqual([{
path: "adapter.call",
description: "Call an adapter-described tool",
signature: "tools.adapter.call(input: { id: string; count?: number }): Promise<unknown>",
}])
expect(runtime.catalog()).toStrictEqual([
{
path: "adapter.call",
description: "Call an adapter-described tool",
signature: "tools.adapter.call(input: { id: string; count?: number }): Promise<unknown>",
},
])
// JSON Schema is render-only: mistyped input passes through unvalidated.
const result = await Effect.runPromise(runtime.execute(`return await tools.adapter.call({ id: 42 })`))
@@ -421,17 +451,25 @@ describe("CodeMode schema flexibility", () => {
input: { type: "object", properties: { login: { type: "string" } }, required: ["login"] },
output: {
$ref: "#/$defs/User",
$defs: { User: { type: "object", properties: { login: { type: "string" }, id: { type: "number" } }, required: ["login", "id"] } },
$defs: {
User: {
type: "object",
properties: { login: { type: "string" }, id: { type: "number" } },
required: ["login", "id"],
},
},
},
run: () => Effect.succeed({ login: "kit", id: 7 }),
})
const runtime = CodeMode.make({ tools: { users: { lookup } } })
expect(runtime.catalog()).toStrictEqual([{
path: "users.lookup",
description: "Look up a user",
signature: "tools.users.lookup(input: { login: string }): Promise<{ login: string; id: number }>",
}])
expect(runtime.catalog()).toStrictEqual([
{
path: "users.lookup",
description: "Look up a user",
signature: "tools.users.lookup(input: { login: string }): Promise<{ login: string; id: number }>",
},
])
const result = await Effect.runPromise(runtime.execute(`return await tools.users.lookup({ login: "kit" })`))
expect(result.ok).toBe(true)
@@ -478,16 +516,20 @@ describe("CodeMode public contract", () => {
expect(agentTool.input).toBe(ExecuteInputSchema)
expect(agentTool.output).toBe(ExecuteResultSchema)
expect(agentTool.description).toBe(runtime.instructions())
expect(Schema.decodeUnknownSync(ExecuteResultSchema)(JSON.parse(JSON.stringify(projected)))).toStrictEqual(projected)
expect(Schema.decodeUnknownSync(ExecuteResultSchema)(JSON.parse(JSON.stringify(projected)))).toStrictEqual(
projected,
)
})
test("inlines a COMPLETE small catalog and keeps search registered but unadvertised", async () => {
const runtime = CodeMode.make({ tools })
expect(runtime.catalog()).toStrictEqual([{
path: "orders.lookup",
description: "Look up an order by ID",
signature: "tools.orders.lookup(input: { id: string }): Promise<{ id: string; status: string }>",
}])
expect(runtime.catalog()).toStrictEqual([
{
path: "orders.lookup",
description: "Look up an order by ID",
signature: "tools.orders.lookup(input: { id: string }): Promise<{ id: string; status: string }>",
},
])
expect(runtime.instructions()).toContain("Available tools (COMPLETE list")
expect(runtime.instructions()).toContain("- orders (1 tool)")
expect(runtime.instructions()).toContain(
@@ -502,11 +544,13 @@ describe("CodeMode public contract", () => {
expect(result.ok).toBe(true)
if (result.ok) {
expect(result.value).toStrictEqual({
items: [{
path: "tools.orders.lookup",
description: "Look up an order by ID",
signature: "tools.orders.lookup(input: {\n id: string\n}): Promise<{\n id: string\n status: string\n}>",
}],
items: [
{
path: "tools.orders.lookup",
description: "Look up an order by ID",
signature: "tools.orders.lookup(input: {\n id: string\n}): Promise<{\n id: string\n status: string\n}>",
},
],
total: 1,
})
}
@@ -521,31 +565,43 @@ describe("CodeMode public contract", () => {
})
const runtime = CodeMode.make({ tools: { context7: { "resolve-library-id": resolveLibrary } } })
expect(runtime.catalog()).toStrictEqual([{
path: "context7.resolve-library-id",
description: "Resolve a library ID",
signature: 'tools.context7["resolve-library-id"](input: { libraryName: string }): Promise<string>',
}])
expect(runtime.instructions()).toContain('tools.context7["resolve-library-id"](input: { libraryName: string }): Promise<string>')
expect(runtime.catalog()).toStrictEqual([
{
path: "context7.resolve-library-id",
description: "Resolve a library ID",
signature: 'tools.context7["resolve-library-id"](input: { libraryName: string }): Promise<string>',
},
])
expect(runtime.instructions()).toContain(
'tools.context7["resolve-library-id"](input: { libraryName: string }): Promise<string>',
)
const search = await Effect.runPromise(runtime.execute(`return await tools.$codemode.search({ query: "resolve library id" })`))
const search = await Effect.runPromise(
runtime.execute(`return await tools.$codemode.search({ query: "resolve library id" })`),
)
expect(search.ok).toBe(true)
if (search.ok) {
expect(search.value).toStrictEqual({
items: [{
path: 'tools.context7["resolve-library-id"]',
description: "Resolve a library ID",
signature: 'tools.context7["resolve-library-id"](input: {\n libraryName: string\n}): Promise<string>',
}],
items: [
{
path: 'tools.context7["resolve-library-id"]',
description: "Resolve a library ID",
signature: 'tools.context7["resolve-library-id"](input: {\n libraryName: string\n}): Promise<string>',
},
],
total: 1,
})
}
const call = await Effect.runPromise(runtime.execute(`return await tools.context7["resolve-library-id"]({ libraryName: "TypeScript" })`))
const call = await Effect.runPromise(
runtime.execute(`return await tools.context7["resolve-library-id"]({ libraryName: "TypeScript" })`),
)
expect(call.ok).toBe(true)
if (call.ok) expect(call.value).toBe("/resolved/TypeScript")
const exact = await Effect.runPromise(runtime.execute(`return await tools.$codemode.search({ query: 'tools.context7["resolve-library-id"]' })`))
const exact = await Effect.runPromise(
runtime.execute(`return await tools.$codemode.search({ query: 'tools.context7["resolve-library-id"]' })`),
)
expect(exact.ok).toBe(true)
if (exact.ok) expect((exact.value as { total: number }).total).toBe(1)
})
@@ -561,7 +617,9 @@ describe("CodeMode public contract", () => {
expect(instructions.indexOf("## Rules")).toBeLessThan(instructions.indexOf("## Syntax"))
expect(instructions.indexOf("## Syntax")).toBeLessThan(instructions.indexOf("\n## Available tools (COMPLETE list"))
// The workflow carries the result-shape guidance; Rules only add content beyond it.
expect(instructions).toContain('`const data = typeof res === "string" ? JSON.parse(res) : res` - most tools return JSON as a string')
expect(instructions).toContain(
'`const data = typeof res === "string" ? JSON.parse(res) : res` - most tools return JSON as a string',
)
expect(instructions).toContain("Return only the fields you need")
expect(instructions).toContain("raw payloads get truncated and waste context")
expect(instructions).toContain("`const res = await tools.<namespace>.<tool>(input)`")
@@ -584,8 +642,12 @@ describe("CodeMode public contract", () => {
expect(partial).toContain(
'1. Find a tool (skip when it is already listed below): `const { items } = await tools.$codemode.search({ query: "<intent + key nouns>" })` - short phrases like "list issues" work best.',
)
expect(partial).toContain("Only tools listed here or returned by `tools.$codemode.search` are available inside `tools`")
expect(partial).toContain('- Browse one namespace: `await tools.$codemode.search({ query: "", namespace: "<name>" })`.')
expect(partial).toContain(
"Only tools listed here or returned by `tools.$codemode.search` are available inside `tools`",
)
expect(partial).toContain(
'- Browse one namespace: `await tools.$codemode.search({ query: "", namespace: "<name>" })`.',
)
expect(partial).not.toContain("total_count")
expect(partial).not.toContain("tools.orders.lookup({")
})
@@ -604,7 +666,9 @@ describe("CodeMode public contract", () => {
expect(instructions).not.toContain("instanceof Error")
expect(instructions).not.toContain("splice")
// The data-boundary note survives.
expect(instructions).toContain("Dates serialize to ISO strings at data boundaries; Map/Set/RegExp serialize to `{}`.")
expect(instructions).toContain(
"Dates serialize to ISO strings at data boundaries; Map/Set/RegExp serialize to `{}`.",
)
})
test("zero tools keep minimal sections and the no-tools notice", () => {
@@ -635,18 +699,22 @@ describe("CodeMode public contract", () => {
tools: { thread: { uploadFile: upload, generateImage: generate }, orders: { lookup } },
discovery: { maxInlineCatalogTokens: 0 },
})
expect(runtime.instructions()).toContain("Available tools (PARTIAL - 0 of 3 shown; find the rest with tools.$codemode.search)")
expect(runtime.instructions()).toContain(
"Available tools (PARTIAL - 0 of 3 shown; find the rest with tools.$codemode.search)",
)
expect(runtime.instructions()).toContain("- thread (2 tools, none shown)")
expect(runtime.instructions()).toContain("- orders (1 tool, none shown)")
expect(runtime.instructions()).toMatch(/\$codemode\.search/)
expect(runtime.instructions()).not.toMatch(/tools\.thread\.uploadFile\(input/)
const result = await Effect.runPromise(runtime.execute(`
const result = await Effect.runPromise(
runtime.execute(`
return await tools.$codemode.search({
query: "send message attachment upload file to current Discord thread",
limit: 2
})
`))
`),
)
expect(result.ok).toBe(true)
if (!result.ok) return
expect(result.value).toStrictEqual({
@@ -666,19 +734,27 @@ describe("CodeMode public contract", () => {
})
expect(result.toolCalls).toStrictEqual([{ name: "$codemode.search" }])
const variants = await Effect.runPromise(runtime.execute(`
const variants = await Effect.runPromise(
runtime.execute(`
return await Promise.all([
tools.$codemode.search({ query: "file" }),
tools.$codemode.search({ query: "image" })
])
`))
`),
)
expect(variants.ok).toBe(true)
if (variants.ok) {
expect((variants.value as Array<{ items: Array<{ path: string }> }>)[0]?.items[0]?.path).toBe("tools.thread.uploadFile")
expect((variants.value as Array<{ items: Array<{ path: string }> }>)[1]?.items[0]?.path).toBe("tools.thread.generateImage")
expect((variants.value as Array<{ items: Array<{ path: string }> }>)[0]?.items[0]?.path).toBe(
"tools.thread.uploadFile",
)
expect((variants.value as Array<{ items: Array<{ path: string }> }>)[1]?.items[0]?.path).toBe(
"tools.thread.generateImage",
)
}
const removed = await Effect.runPromise(runtime.execute(`return await tools.$codemode.describe({ path: "thread.uploadFile" })`))
const removed = await Effect.runPromise(
runtime.execute(`return await tools.$codemode.describe({ path: "thread.uploadFile" })`),
)
expect(removed.ok).toBe(false)
if (!removed.ok) expect(removed.error.kind).toBe("UnknownTool")
})
@@ -706,15 +782,19 @@ describe("CodeMode public contract", () => {
}
for (const query of ["many.tool13", "tools.many.tool13"]) {
const exact = await Effect.runPromise(runtime.execute(`return await tools.$codemode.search({ query: ${JSON.stringify(query)} })`))
const exact = await Effect.runPromise(
runtime.execute(`return await tools.$codemode.search({ query: ${JSON.stringify(query)} })`),
)
expect(exact.ok).toBe(true)
if (exact.ok) {
expect(exact.value).toStrictEqual({
items: [{
path: "tools.many.tool13",
description: "Numbered tool 13",
signature: "tools.many.tool13(input: {\n id: string\n}): Promise<string>",
}],
items: [
{
path: "tools.many.tool13",
description: "Numbered tool 13",
signature: "tools.many.tool13(input: {\n id: string\n}): Promise<string>",
},
],
total: 1,
})
}
@@ -737,20 +817,23 @@ describe("CodeMode public contract", () => {
})
// Empty query + namespace browses just that namespace, alphabetical by path.
const browse = await Effect.runPromise(runtime.execute(
`return await tools.$codemode.search({ query: "", namespace: "github" })`,
))
const browse = await Effect.runPromise(
runtime.execute(`return await tools.$codemode.search({ query: "", namespace: "github" })`),
)
expect(browse.ok).toBe(true)
if (browse.ok) {
const value = browse.value as { items: Array<{ path: string }>; total: number }
expect(value.total).toBe(2)
expect(value.items.map((item) => item.path)).toStrictEqual(["tools.github.create_issue", "tools.github.list_issues"])
expect(value.items.map((item) => item.path)).toStrictEqual([
"tools.github.create_issue",
"tools.github.list_issues",
])
}
// A query + namespace ranks within that namespace only.
const scoped = await Effect.runPromise(runtime.execute(
`return await tools.$codemode.search({ query: "issues", namespace: "linear" })`,
))
const scoped = await Effect.runPromise(
runtime.execute(`return await tools.$codemode.search({ query: "issues", namespace: "linear" })`),
)
expect(scoped.ok).toBe(true)
if (scoped.ok) {
const value = scoped.value as { items: Array<{ path: string }>; total: number }
@@ -758,9 +841,9 @@ describe("CodeMode public contract", () => {
expect(value.items[0]?.path).toBe("tools.linear.list_issues")
}
const invalid = await Effect.runPromise(runtime.execute(
`return await tools.$codemode.search({ query: "issues", namespace: 7 })`,
))
const invalid = await Effect.runPromise(
runtime.execute(`return await tools.$codemode.search({ query: "issues", namespace: 7 })`),
)
expect(invalid.ok).toBe(false)
if (!invalid.ok) expect(invalid.error.kind).toBe("InvalidToolInput")
})
@@ -785,9 +868,9 @@ describe("CodeMode public contract", () => {
// "attachment" appears in neither path nor description - only in the input schema's
// property names, which the searchable text includes.
const byParameter = await Effect.runPromise(runtime.execute(
`return await tools.$codemode.search({ query: "attachment" })`,
))
const byParameter = await Effect.runPromise(
runtime.execute(`return await tools.$codemode.search({ query: "attachment" })`),
)
expect(byParameter.ok).toBe(true)
if (byParameter.ok) {
const value = byParameter.value as { items: Array<{ path: string }>; total: number }
@@ -796,9 +879,9 @@ describe("CodeMode public contract", () => {
}
// Substring matching: a partial word ("docum") still hits the description.
const bySubstring = await Effect.runPromise(runtime.execute(
`return await tools.$codemode.search({ query: "docum" })`,
))
const bySubstring = await Effect.runPromise(
runtime.execute(`return await tools.$codemode.search({ query: "docum" })`),
)
expect(bySubstring.ok).toBe(true)
if (bySubstring.ok) {
const value = bySubstring.value as { items: Array<{ path: string }>; total: number }
@@ -825,9 +908,9 @@ describe("CodeMode public contract", () => {
})
// "issues" still finds the singular-only tool (term OR singular(term) per field)...
const plural = await Effect.runPromise(runtime.execute(
`return await tools.$codemode.search({ query: "issues", namespace: "tracker" })`,
))
const plural = await Effect.runPromise(
runtime.execute(`return await tools.$codemode.search({ query: "issues", namespace: "tracker" })`),
)
expect(plural.ok).toBe(true)
if (plural.ok) {
const value = plural.value as { items: Array<{ path: string }>; total: number }
@@ -836,14 +919,15 @@ describe("CodeMode public contract", () => {
}
// ...while a true "issues" path match still outranks the singular-only description match.
const ranked = await Effect.runPromise(runtime.execute(
`return await tools.$codemode.search({ query: "issues" })`,
))
const ranked = await Effect.runPromise(runtime.execute(`return await tools.$codemode.search({ query: "issues" })`))
expect(ranked.ok).toBe(true)
if (ranked.ok) {
const value = ranked.value as { items: Array<{ path: string }>; total: number }
expect(value.total).toBe(2)
expect(value.items.map((item) => item.path)).toStrictEqual(["tools.github.list_issues", "tools.tracker.fetch_all"])
expect(value.items.map((item) => item.path)).toStrictEqual([
"tools.github.list_issues",
"tools.tracker.fetch_all",
])
}
})
@@ -882,8 +966,12 @@ describe("CodeMode public contract", () => {
run: () => Effect.succeed("ok"),
})
const expensive = Tool.make({
description: "An expensive tool whose description alone consumes far more than the remaining inline catalog byte budget for this runtime",
input: Schema.Struct({ someRatherLongParameterName: Schema.String, anotherEvenLongerParameterName: Schema.Number }),
description:
"An expensive tool whose description alone consumes far more than the remaining inline catalog byte budget for this runtime",
input: Schema.Struct({
someRatherLongParameterName: Schema.String,
anotherEvenLongerParameterName: Schema.Number,
}),
output: Schema.String,
run: () => Effect.succeed("ok"),
})
@@ -896,7 +984,9 @@ describe("CodeMode public contract", () => {
})
const instructions = runtime.instructions()
expect(instructions).toContain("Available tools (PARTIAL - 2 of 3 shown; find the rest with tools.$codemode.search)")
expect(instructions).toContain(
"Available tools (PARTIAL - 2 of 3 shown; find the rest with tools.$codemode.search)",
)
expect(instructions).toContain("- alpha (2 tools, 1 shown)")
expect(instructions).toContain(" - tools.alpha.cheap(input: { q: string }): Promise<string> // Cheap")
expect(instructions).not.toContain("tools.alpha.expensive(")
@@ -912,10 +1002,11 @@ describe("CodeMode public contract", () => {
description: "Double a number",
input: Schema.Struct({ value: Schema.NumberFromString }),
output: Schema.NumberFromString,
run: ({ value }) => Effect.sync(() => {
observed.push(value)
return String(value * 2)
}),
run: ({ value }) =>
Effect.sync(() => {
observed.push(value)
return String(value * 2)
}),
})
const runtime = CodeMode.make({
tools: { math: { double: transformed } },
@@ -934,9 +1025,11 @@ describe("CodeMode public contract", () => {
})
test("returns JSON-safe data and normalizes undefined to null", async () => {
const result = await Effect.runPromise(CodeMode.execute({
code: `return { top: undefined, nested: [1, undefined] }`,
}))
const result = await Effect.runPromise(
CodeMode.execute({
code: `return { top: undefined, nested: [1, undefined] }`,
}),
)
expect(result).toStrictEqual({
ok: true,
value: { top: null, nested: [1, null] },
@@ -947,18 +1040,20 @@ describe("CodeMode public contract", () => {
test("rejects invalid configuration and discovery limits", async () => {
expect(() => CodeMode.execute({ code: "return 1", limits: { timeoutMs: 0 } })).toThrow(RangeError)
expect(() => CodeMode.execute({ code: "return 1", limits: { timeoutMs: Number.POSITIVE_INFINITY } })).toThrow(RangeError)
expect(() => CodeMode.execute({ code: "return 1", limits: { timeoutMs: Number.POSITIVE_INFINITY } })).toThrow(
RangeError,
)
expect(() => CodeMode.execute({ code: "return 1", limits: { maxToolCalls: -1 } })).toThrow(RangeError)
expect(() => CodeMode.execute({ code: "return 1", limits: { maxOutputBytes: -1 } })).toThrow(RangeError)
expect(() => CodeMode.make({ tools, discovery: { maxInlineCatalogTokens: -1 } })).toThrow(RangeError)
const result = await Effect.runPromise(CodeMode.make({
tools,
discovery: { maxInlineCatalogTokens: 0 },
}).execute(
`return await tools.$codemode.search({ query: "order", limit: 0.5 })`,
))
const result = await Effect.runPromise(
CodeMode.make({
tools,
discovery: { maxInlineCatalogTokens: 0 },
}).execute(`return await tools.$codemode.search({ query: "order", limit: 0.5 })`),
)
expect(result.ok).toBe(false)
if (result.ok) return
expect(result.error.kind).toBe("InvalidToolInput")
@@ -979,14 +1074,16 @@ describe("CodeMode public contract", () => {
output: Schema.Number,
run: () => Effect.succeed(1),
})
const result = await Effect.runPromise(CodeMode.execute({
tools: { host: { count: counter } },
code: `
const result = await Effect.runPromise(
CodeMode.execute({
tools: { host: { count: counter } },
code: `
let total = 0
for (let i = 0; i < 150; i += 1) total += await tools.host.count({})
return total
`,
}))
}),
)
expect(result).toMatchObject({ ok: true, value: 150 })
if (result.ok) expect(result.toolCalls.length).toBe(150)
})
@@ -1008,8 +1105,6 @@ describe("CodeMode public contract", () => {
})
test("reserves the discovery namespace", () => {
expect(() => CodeMode.make({ tools: { $codemode: { lookup } } })).toThrow(
/reserved for CodeMode discovery tools/,
)
expect(() => CodeMode.make({ tools: { $codemode: { lookup } } })).toThrow(/reserved for CodeMode discovery tools/)
})
})
+24 -12
View File
@@ -36,10 +36,12 @@ const error = async (code: string) => {
describe("Object.keys over tool references", () => {
test("enumerates top-level namespaces (the transcript program)", async () => {
expect(await value(`
expect(
await value(`
const namespaces = Object.keys(tools)
return { namespaces, count: namespaces.length }
`)).toEqual({ namespaces: ["github", "memory", "playwright"], count: 3 })
`),
).toEqual({ namespaces: ["github", "memory", "playwright"], count: 3 })
})
test("enumerates tool names at a nested namespace", async () => {
@@ -92,7 +94,8 @@ describe("Object.keys over arrays", () => {
describe("for...in", () => {
test("iterates own enumerable keys of a plain object with break/continue", async () => {
expect(await value(`
expect(
await value(`
const seen = []
for (const key in { a: 1, b: 2, c: 3, d: 4 }) {
if (key === "b") continue
@@ -100,41 +103,50 @@ describe("for...in", () => {
seen.push(key)
}
return seen
`)).toEqual(["a", "c"])
`),
).toEqual(["a", "c"])
})
test("iterates index strings over arrays", async () => {
expect(await value(`
expect(
await value(`
const indexes = []
for (const i in ["x", "y", "z"]) {
if (i === "2") break
indexes.push(i)
}
return indexes
`)).toEqual(["0", "1"])
`),
).toEqual(["0", "1"])
})
test("supports let declarations and bare identifiers", async () => {
expect(await value(`
expect(
await value(`
let last = ""
for (let key in { a: 1, b: 2 }) last = key
return last
`)).toBe("b")
expect(await value(`
`),
).toBe("b")
expect(
await value(`
let key = "before"
for (key in { only: 1 }) {}
return key
`)).toBe("only")
`),
).toBe("only")
})
test("enumerates namespaces and tools from the host tool tree", async () => {
expect(await value(`
expect(
await value(`
const names = []
for (const ns in tools) {
for (const name in tools[ns]) names.push(ns + "." + name)
}
return names
`)).toEqual(["github.list_issues", "github.get_issue", "memory.search", "playwright.navigate"])
`),
).toEqual(["github.list_issues", "github.get_issue", "memory.search", "playwright.navigate"])
})
test("unsupported values fail with a hint at for...of and Object.keys", async () => {
+97 -52
View File
@@ -143,21 +143,40 @@ describe("H1: NaN/Infinity flow as intermediates and normalize to null at the bo
describe("Error values and instanceof", () => {
test("new Error carries name/message and is instanceof Error", async () => {
expect(await value(`const e = new Error("boom"); return [e instanceof Error, e.name, e.message]`)).toEqual([true, "Error", "boom"])
expect(await value(`const e = new Error("boom"); return [e instanceof Error, e.name, e.message]`)).toEqual([
true,
"Error",
"boom",
])
})
test("Error without new behaves like new Error", async () => {
expect(await value(`const e = Error("plain"); return [e instanceof Error, e.name, e.message]`)).toEqual([true, "Error", "plain"])
expect(await value(`const e = new Error(); return [e.name, e.message, e instanceof Error]`)).toEqual(["Error", "", true])
expect(await value(`const e = Error("plain"); return [e instanceof Error, e.name, e.message]`)).toEqual([
true,
"Error",
"plain",
])
expect(await value(`const e = new Error(); return [e.name, e.message, e instanceof Error]`)).toEqual([
"Error",
"",
true,
])
})
test("specific error types are instanceof themselves and Error, not each other", async () => {
expect(await value(`const e = new TypeError("t"); return [e instanceof TypeError, e instanceof Error, e instanceof RangeError]`)).toEqual([true, true, false])
expect(
await value(
`const e = new TypeError("t"); return [e instanceof TypeError, e instanceof Error, e instanceof RangeError]`,
),
).toEqual([true, true, false])
expect(await value(`return new Error("e") instanceof TypeError`)).toBe(false)
})
test("thrown errors keep instanceof through try/catch", async () => {
expect(await value(`try { throw new Error("x") } catch (e) { return [e instanceof Error, e.message] }`)).toEqual([true, "x"])
expect(await value(`try { throw new Error("x") } catch (e) { return [e instanceof Error, e.message] }`)).toEqual([
true,
"x",
])
})
test("interpreter runtime failures are caught as Error values", async () => {
@@ -168,33 +187,48 @@ describe("Error values and instanceof", () => {
test("caught failures carry the constructor name the real-JS failure would have", async () => {
// JSON.parse throws SyntaxError: name and specific-instanceof both carry through, and the
// message keeps the engine's position detail.
expect(await value(`
expect(
await value(`
try { JSON.parse("{oops") } catch (e) {
return [e.name, e instanceof SyntaxError, e instanceof Error, e instanceof TypeError, e.message.includes("JSON")]
}
`)).toEqual(["SyntaxError", true, true, false, true])
expect(await value(`try { undeclared() } catch (e) { return [e.name, e instanceof ReferenceError] }`))
.toEqual(["ReferenceError", true])
expect(await value(`try { const c = 1; c = 2 } catch (e) { return [e.name, e instanceof TypeError] }`))
.toEqual(["TypeError", true])
expect(await value(`try { "a".normalize("NOPE") } catch (e) { return [e.name, e instanceof RangeError] }`))
.toEqual(["RangeError", true])
expect(await value(`try { "a".match("(") } catch (e) { return [e.name, e instanceof SyntaxError] }`))
.toEqual(["SyntaxError", true])
expect(await value(`try { new RegExp("(") } catch (e) { return [e.name, e instanceof SyntaxError] }`))
.toEqual(["SyntaxError", true])
`),
).toEqual(["SyntaxError", true, true, false, true])
expect(await value(`try { undeclared() } catch (e) { return [e.name, e instanceof ReferenceError] }`)).toEqual([
"ReferenceError",
true,
])
expect(await value(`try { const c = 1; c = 2 } catch (e) { return [e.name, e instanceof TypeError] }`)).toEqual([
"TypeError",
true,
])
expect(await value(`try { "a".normalize("NOPE") } catch (e) { return [e.name, e instanceof RangeError] }`)).toEqual(
["RangeError", true],
)
expect(await value(`try { "a".match("(") } catch (e) { return [e.name, e instanceof SyntaxError] }`)).toEqual([
"SyntaxError",
true,
])
expect(await value(`try { new RegExp("(") } catch (e) { return [e.name, e instanceof SyntaxError] }`)).toEqual([
"SyntaxError",
true,
])
})
test("diagnostics without a specific real-JS analogue are named plain Error", async () => {
expect(await value(`try { JSON.parse(5) } catch (e) { return [e.name, e instanceof Error] }`))
.toEqual(["Error", true])
expect(await value(`try { JSON.parse(5) } catch (e) { return [e.name, e instanceof Error] }`)).toEqual([
"Error",
true,
])
})
test("Promise.allSettled rejection reasons are Error values", async () => {
expect(await value(`
expect(
await value(`
const settled = await Promise.allSettled([Promise.reject(new Error("b"))])
return [settled[0].reason instanceof Error, settled[0].reason.message]
`)).toEqual([true, "b"])
`),
).toEqual([true, "b"])
})
test("non-error thrown values are not instanceof Error", async () => {
@@ -203,7 +237,11 @@ describe("Error values and instanceof", () => {
})
test("plain data is never instanceof Error", async () => {
expect(await value(`return [({}) instanceof Error, "s" instanceof Error, null instanceof Error]`)).toEqual([false, false, false])
expect(await value(`return [({}) instanceof Error, "s" instanceof Error, null instanceof Error]`)).toEqual([
false,
false,
false,
])
})
test("error values still serialize as plain { name, message } data", async () => {
@@ -227,17 +265,29 @@ describe("Error values and instanceof", () => {
describe("array methods: splice, fill, copyWithin, keys/values/entries", () => {
test("splice removes in place and returns the removed elements", async () => {
expect(await value(`const a = [1,2,3,4]; const removed = a.splice(1, 2); return { removed, a }`)).toEqual({ removed: [2, 3], a: [1, 4] })
expect(await value(`const a = [1,2,3,4]; const removed = a.splice(1, 2); return { removed, a }`)).toEqual({
removed: [2, 3],
a: [1, 4],
})
})
test("splice inserts new elements at the cut", async () => {
expect(await value(`const a = ["a","d"]; a.splice(1, 0, "b", "c"); return a`)).toEqual(["a", "b", "c", "d"])
expect(await value(`const a = [1,2,3]; const removed = a.splice(1, 1, "x"); return { removed, a }`)).toEqual({ removed: [2], a: [1, "x", 3] })
expect(await value(`const a = [1,2,3]; const removed = a.splice(1, 1, "x"); return { removed, a }`)).toEqual({
removed: [2],
a: [1, "x", 3],
})
})
test("splice with one argument removes to the end; negative start counts back", async () => {
expect(await value(`const a = [1,2,3]; const removed = a.splice(1); return { removed, a }`)).toEqual({ removed: [2, 3], a: [1] })
expect(await value(`const a = [1,2,3]; const removed = a.splice(-1); return { removed, a }`)).toEqual({ removed: [3], a: [1, 2] })
expect(await value(`const a = [1,2,3]; const removed = a.splice(1); return { removed, a }`)).toEqual({
removed: [2, 3],
a: [1],
})
expect(await value(`const a = [1,2,3]; const removed = a.splice(-1); return { removed, a }`)).toEqual({
removed: [3],
a: [1, 2],
})
})
test("splice rejects inserting a container into itself", async () => {
@@ -258,11 +308,13 @@ describe("array methods: splice, fill, copyWithin, keys/values/entries", () => {
test("keys/values/entries return arrays usable with for...of and spread", async () => {
expect(await value(`return [...["x","y","z"].keys()]`)).toEqual([0, 1, 2])
expect(await value(`return ["x","y"].values()`)).toEqual(["x", "y"])
expect(await value(`
expect(
await value(`
const out = []
for (const [index, item] of ["a","b"].entries()) out.push(index + ":" + item)
return out
`)).toEqual(["0:a", "1:b"])
`),
).toEqual(["0:a", "1:b"])
expect(await value(`return [...[7].entries()]`)).toEqual([[0, 7]])
})
})
@@ -300,40 +352,33 @@ describe("compound assignment matches its binary operator", () => {
}
test("sandbox Date += concatenates its string form, like d = d + 1", async () => {
const result = await pair(
`let d = new Date(1000); d += 1; return d`,
`let d = new Date(1000); d = d + 1; return d`,
)
const result = await pair(`let d = new Date(1000); d += 1; return d`, `let d = new Date(1000); d = d + 1; return d`)
expect(result).toBe("1970-01-01T00:00:01.000Z1")
})
test("sandbox Date numeric compound ops use its time value", async () => {
expect(await pair(
`let d = new Date(1000); d -= 400; return d`,
`let d = new Date(1000); d = d - 400; return d`,
)).toBe(600)
expect(await pair(
`let d = new Date(1000); d /= 4; return d`,
`let d = new Date(1000); d = d / 4; return d`,
)).toBe(250)
expect(
await pair(`let d = new Date(1000); d -= 400; return d`, `let d = new Date(1000); d = d - 400; return d`),
).toBe(600)
expect(await pair(`let d = new Date(1000); d /= 4; return d`, `let d = new Date(1000); d = d / 4; return d`)).toBe(
250,
)
})
test("string += object/array matches x = x + obj", async () => {
expect(await pair(
`let x = "a"; x += { b: 1 }; return x`,
`let x = "a"; x = x + { b: 1 }; return x`,
)).toBe("a[object Object]")
expect(await pair(
`let x = "a"; x += [1, 2]; return x`,
`let x = "a"; x = x + [1, 2]; return x`,
)).toBe("a1,2")
expect(await pair(`let x = "a"; x += { b: 1 }; return x`, `let x = "a"; x = x + { b: 1 }; return x`)).toBe(
"a[object Object]",
)
expect(await pair(`let x = "a"; x += [1, 2]; return x`, `let x = "a"; x = x + [1, 2]; return x`)).toBe("a1,2")
})
test("compound assignment through a member target coerces the same way", async () => {
expect(await pair(
`const o = { s: "t" }; o.s += new Date(0); return o.s`,
`const o = { s: "t" }; o.s = o.s + new Date(0); return o.s`,
)).toBe("t1970-01-01T00:00:00.000Z")
expect(
await pair(
`const o = { s: "t" }; o.s += new Date(0); return o.s`,
`const o = { s: "t" }; o.s = o.s + new Date(0); return o.s`,
),
).toBe("t1970-01-01T00:00:00.000Z")
})
test("numeric and string compound operators sweep identically to their expansions", async () => {
+54 -27
View File
@@ -23,7 +23,7 @@ const sleepyTool = (trace: Trace) =>
input: Schema.Struct({ id: Schema.Number, ms: Schema.optionalKey(Schema.Number) }),
output: Schema.Number,
run: ({ id, ms }) =>
Effect.gen(function*() {
Effect.gen(function* () {
trace.starts.push(id)
trace.active += 1
trace.maxActive = Math.max(trace.maxActive, trace.active)
@@ -31,10 +31,14 @@ const sleepyTool = (trace: Trace) =>
trace.active -= 1
trace.completed += 1
return id
}).pipe(Effect.onInterrupt(() => Effect.sync(() => {
trace.active -= 1
trace.interrupted += 1
}))),
}).pipe(
Effect.onInterrupt(() =>
Effect.sync(() => {
trace.active -= 1
trace.interrupted += 1
}),
),
),
})
const failingTool = Tool.make({
@@ -46,11 +50,13 @@ const failingTool = Tool.make({
const run = (code: string, options: { trace?: Trace; limits?: ExecutionLimits } = {}): Promise<ExecuteResult> => {
const trace = options.trace ?? makeTrace()
return Effect.runPromise(CodeMode.execute({
tools: { host: { sleepy: sleepyTool(trace), fail: failingTool } },
code,
...(options.limits ? { limits: options.limits } : {}),
}))
return Effect.runPromise(
CodeMode.execute({
tools: { host: { sleepy: sleepyTool(trace), fail: failingTool } },
code,
...(options.limits ? { limits: options.limits } : {}),
}),
)
}
const value = async (code: string, options: { trace?: Trace; limits?: ExecutionLimits } = {}) => {
@@ -121,7 +127,8 @@ describe("first-class promise values", () => {
})
test("an awaited failure is catchable exactly like a synchronous throw", async () => {
expect(await value(`
expect(
await value(`
const p = tools.host.fail({})
try {
await p
@@ -129,7 +136,8 @@ describe("first-class promise values", () => {
} catch (e) {
return e.message
}
`)).toBe("Lookup refused")
`),
).toBe("Lookup refused")
})
test("a fire-and-forget call completes before the execution ends", async () => {
@@ -186,20 +194,24 @@ describe("promises at data boundaries", () => {
describe("Promise.all over arbitrary arrays", () => {
test("mixes promises and plain values, preserving order", async () => {
expect(await value(`
expect(
await value(`
return await Promise.all([tools.host.sleepy({ id: 1 }), "plain", tools.host.sleepy({ id: 2 }), 42])
`)).toEqual([1, "plain", 2, 42])
`),
).toEqual([1, "plain", 2, 42])
})
test("accepts arrays built beforehand, passed as identifiers, and spread elements", async () => {
expect(await value(`
expect(
await value(`
const calls = []
calls.push(tools.host.sleepy({ id: 1 }))
calls.push(7)
const more = [tools.host.sleepy({ id: 2 })]
const batch = [...calls, ...more, "x"]
return await Promise.all(batch)
`)).toEqual([1, 7, 2, "x"])
`),
).toEqual([1, 7, 2, "x"])
})
test("runs items.map tool calls in parallel", async () => {
@@ -238,14 +250,16 @@ describe("Promise.all over arbitrary arrays", () => {
})
test("rejects with the first failure, catchable in-program", async () => {
expect(await value(`
expect(
await value(`
try {
await Promise.all([tools.host.sleepy({ id: 1 }), tools.host.fail({})])
return "no"
} catch (e) {
return e.message
}
`)).toBe("Lookup refused")
`),
).toBe("Lookup refused")
})
test("a non-collection argument is a clear error", async () => {
@@ -264,14 +278,16 @@ describe("Promise.all over arbitrary arrays", () => {
describe("Promise.allSettled", () => {
test("reports fulfilled and rejected outcomes with catch-normalized reasons", async () => {
expect(await value(`
expect(
await value(`
return await Promise.allSettled([
tools.host.sleepy({ id: 5 }),
tools.host.fail({}),
"plain",
Promise.reject(new Error("boom")),
])
`)).toEqual([
`),
).toEqual([
{ status: "fulfilled", value: 5 },
{ status: "rejected", reason: { name: "Error", message: "Lookup refused" } },
{ status: "fulfilled", value: "plain" },
@@ -306,7 +322,8 @@ describe("Promise.race", () => {
})
test("awaiting an interrupted loser afterwards is a catchable program failure", async () => {
expect(await value(`
expect(
await value(`
const fast = tools.host.sleepy({ id: 1, ms: 10 })
const slow = tools.host.sleepy({ id: 2, ms: 5000 })
const winner = await Promise.race([fast, slow])
@@ -316,23 +333,31 @@ describe("Promise.race", () => {
} catch (e) {
return { winner, caught: e.message }
}
`)).toEqual({ winner: 1, caught: "This tool call was interrupted because another value settled a Promise.race first." })
`),
).toEqual({
winner: 1,
caught: "This tool call was interrupted because another value settled a Promise.race first.",
})
})
test("a rejection can win the race", async () => {
expect(await value(`
expect(
await value(`
try {
await Promise.race([tools.host.fail({}), tools.host.sleepy({ id: 1, ms: 5000 })])
return "no"
} catch (e) {
return e.message
}
`)).toBe("Lookup refused")
`),
).toBe("Lookup refused")
})
test("a plain value wins over pending promises", async () => {
const trace = makeTrace()
expect(await value(`return await Promise.race([tools.host.sleepy({ id: 1, ms: 5000 }), "immediate"])`, { trace })).toBe("immediate")
expect(
await value(`return await Promise.race([tools.host.sleepy({ id: 1, ms: 5000 }), "immediate"])`, { trace }),
).toBe("immediate")
expect(trace.interrupted).toBe(1)
})
@@ -350,14 +375,16 @@ describe("Promise.resolve / Promise.reject", () => {
})
test("reject produces a promise whose await throws the reason", async () => {
expect(await value(`
expect(
await value(`
try {
await Promise.reject("nope")
return "no"
} catch (e) {
return e
}
`)).toBe("nope")
`),
).toBe("nope")
})
})
+19 -14
View File
@@ -83,15 +83,9 @@ describe("pretty signature rendering", () => {
true,
)
expect(pretty).toBe(
[
"{",
" /** Search filter */",
" filter?: {",
" /** Issue state */",
" state?: string",
" }",
"}",
].join("\n"),
["{", " /** Search filter */", " filter?: {", " /** Issue state */", " state?: string", " }", "}"].join(
"\n",
),
)
})
@@ -119,7 +113,14 @@ describe("pretty signature rendering", () => {
expect(pretty).toContain(" /** @deprecated */\n legacy?: string")
expect(pretty).toContain(" /** @format uri */\n homepage?: string")
expect(pretty).toContain(
[" /**", ' * @default ["a","b"]', " * @minItems 2", " * @maxItems 5", " */", " tags?: Array<string>"].join("\n"),
[
" /**",
' * @default ["a","b"]',
" * @minItems 2",
" * @maxItems 5",
" */",
" tags?: Array<string>",
].join("\n"),
)
})
@@ -212,7 +213,11 @@ describe("non-identifier property names render as quoted keys", () => {
const tool = Tool.make({
description: "Adapter tool with awkward field names",
input: rawSchema,
output: { type: "object", properties: { "content-type": { type: "string" } }, required: ["content-type"] } as const,
output: {
type: "object",
properties: { "content-type": { type: "string" } },
required: ["content-type"],
} as const,
run: () => Effect.succeed({ "content-type": "text/plain" }),
})
expect(inputTypeScript(tool)).toContain('"foo-bar"?: string')
@@ -269,9 +274,9 @@ describe("pretty signatures in search results", () => {
const runtime = CodeMode.make({ tools: { github: { list_issues: listIssues }, orders: { lookup: lookupOrder } } })
const search = async (query: string) => {
const result = await Effect.runPromise(runtime.execute(
`return await tools.$codemode.search({ query: ${JSON.stringify(query)} })`,
))
const result = await Effect.runPromise(
runtime.execute(`return await tools.$codemode.search({ query: ${JSON.stringify(query)} })`),
)
expect(result.ok).toBe(true)
if (!result.ok) throw new Error("search failed")
return result.value as { items: Array<{ path: string; signature: string }>; total: number }
+116 -48
View File
@@ -40,7 +40,11 @@ describe("Date", () => {
})
test("UTC getters read calendar components", async () => {
expect(await value(`const d = new Date("2024-03-05T06:07:08.009Z"); return [d.getUTCFullYear(), d.getUTCMonth(), d.getUTCDate(), d.getUTCHours(), d.getUTCMinutes(), d.getUTCSeconds(), d.getUTCMilliseconds()]`)).toEqual([2024, 2, 5, 6, 7, 8, 9])
expect(
await value(
`const d = new Date("2024-03-05T06:07:08.009Z"); return [d.getUTCFullYear(), d.getUTCMonth(), d.getUTCDate(), d.getUTCHours(), d.getUTCMinutes(), d.getUTCSeconds(), d.getUTCMilliseconds()]`,
),
).toEqual([2024, 2, 5, 6, 7, 8, 9])
})
test("invalid dates yield NaN times, guardable in-sandbox", async () => {
@@ -49,7 +53,9 @@ describe("Date", () => {
})
test("toISOString on an invalid date is a catchable error", async () => {
expect(await value(`try { new Date("garbage").toISOString(); return "no" } catch { return "caught" }`)).toBe("caught")
expect(await value(`try { new Date("garbage").toISOString(); return "no" } catch { return "caught" }`)).toBe(
"caught",
)
})
test("template interpolation renders the ISO form", async () => {
@@ -72,14 +78,18 @@ describe("Date", () => {
})
test("sorting dates with a numeric comparator", async () => {
expect(await value(`
expect(
await value(`
const dates = [new Date(3000), new Date(1000), new Date(2000)]
return dates.sort((a, b) => a - b).map((d) => d.getTime())
`)).toEqual([1000, 2000, 3000])
`),
).toEqual([1000, 2000, 3000])
})
test("new Date(year, month, day) accepts component form", async () => {
expect(await value(`const d = new Date(2024, 0, 2); return [d.getFullYear(), d.getMonth(), d.getDate()]`)).toEqual([2024, 0, 2])
expect(await value(`const d = new Date(2024, 0, 2); return [d.getFullYear(), d.getMonth(), d.getDate()]`)).toEqual([
2024, 0, 2,
])
})
test("typeof and unknown properties are forgiving", async () => {
@@ -95,25 +105,31 @@ describe("RegExp", () => {
})
test("exec exposes captures and index", async () => {
expect(await value(`const m = /a(b+)/.exec("xxabbc"); return { full: m[0], group: m[1], index: m.index }`)).toEqual({
full: "abb",
group: "bb",
index: 2,
})
expect(await value(`const m = /a(b+)/.exec("xxabbc"); return { full: m[0], group: m[1], index: m.index }`)).toEqual(
{
full: "abb",
group: "bb",
index: 2,
},
)
expect(await value(`return /a/.exec("zzz")`)).toBeNull()
})
test("named groups read through", async () => {
expect(await value(`const m = /(?<word>[a-z]+)-(?<num>\\d+)/.exec("id ab-42"); return m.groups.word + m.groups.num`)).toBe("ab42")
expect(
await value(`const m = /(?<word>[a-z]+)-(?<num>\\d+)/.exec("id ab-42"); return m.groups.word + m.groups.num`),
).toBe("ab42")
})
test("global exec advances lastIndex across calls", async () => {
expect(await value(`
expect(
await value(`
const r = /\\d+/g
const first = r.exec("a1b22c")
const second = r.exec("a1b22c")
return [first[0], second[0]]
`)).toEqual(["1", "22"])
`),
).toEqual(["1", "22"])
})
test("string match: non-global carries index, global lists all matches", async () => {
@@ -194,34 +210,51 @@ describe("RegExp", () => {
describe("Map", () => {
test("get/set/has/size with chaining", async () => {
expect(await value(`
expect(
await value(`
const m = new Map()
m.set("a", 1).set("b", 2)
return { a: m.get("a"), b: m.get("b"), has: m.has("a"), miss: m.get("zz") === undefined, size: m.size }
`)).toEqual({ a: 1, b: 2, has: true, miss: true, size: 5 - 3 })
`),
).toEqual({ a: 1, b: 2, has: true, miss: true, size: 5 - 3 })
})
test("object keys use identity", async () => {
expect(await value(`
expect(
await value(`
const key = { id: 1 }
const m = new Map()
m.set(key, "hit")
return [m.get(key), m.get({ id: 1 }) === undefined]
`)).toEqual(["hit", true])
`),
).toEqual(["hit", true])
})
test("construction from entry pairs and another Map", async () => {
expect(await value(`const m = new Map([["a", 1], ["b", 2]]); return m.get("b")`)).toBe(2)
expect(await value(`const m = new Map([["a", 1]]); const n = new Map(m); n.set("b", 2); return [n.get("a"), n.get("b"), m.has("b")]`)).toEqual([1, 2, false])
expect(
await value(
`const m = new Map([["a", 1]]); const n = new Map(m); n.set("b", 2); return [n.get("a"), n.get("b"), m.has("b")]`,
),
).toEqual([1, 2, false])
expect((await error(`return new Map("nope")`)).message).toMatch(/\[key, value\] pairs/)
expect((await error(`return new Map(["flat"])`)).message).toMatch(/\[key, value\] pairs/)
})
test("keys/values/entries return arrays", async () => {
expect(await value(`
expect(
await value(`
const m = new Map([["a", 1], ["b", 2]])
return { keys: m.keys(), values: m.values(), entries: m.entries() }
`)).toEqual({ keys: ["a", "b"], values: [1, 2], entries: [["a", 1], ["b", 2]] })
`),
).toEqual({
keys: ["a", "b"],
values: [1, 2],
entries: [
["a", 1],
["b", 2],
],
})
})
test("Object.fromEntries(map) and Array.from(map)", async () => {
@@ -230,13 +263,15 @@ describe("Map", () => {
})
test("for...of iterates [key, value] pairs with destructuring", async () => {
expect(await value(`
expect(
await value(`
const m = new Map([["a", 1], ["b", 2]])
let total = 0
let names = ""
for (const [key, count] of m) { names += key; total += count }
return names + total
`)).toBe("ab3")
`),
).toBe("ab3")
})
test("spread produces entry pairs", async () => {
@@ -244,32 +279,38 @@ describe("Map", () => {
})
test("forEach passes (value, key)", async () => {
expect(await value(`
expect(
await value(`
const m = new Map([["a", 1], ["b", 2]])
const seen = []
m.forEach((count, key) => seen.push(key + count))
return seen
`)).toEqual(["a1", "b2"])
`),
).toEqual(["a1", "b2"])
})
test("delete and clear", async () => {
expect(await value(`
expect(
await value(`
const m = new Map([["a", 1], ["b", 2]])
const removed = m.delete("a")
const missed = m.delete("zz")
const sizeAfterDelete = m.size
m.clear()
return [removed, missed, sizeAfterDelete, m.size]
`)).toEqual([true, false, 1, 0])
`),
).toEqual([true, false, 1, 0])
})
test("counting idiom: grouped tallies", async () => {
expect(await value(`
expect(
await value(`
const words = ["a", "b", "a", "c", "a"]
const counts = new Map()
for (const word of words) counts.set(word, (counts.get(word) ?? 0) + 1)
return Object.fromEntries(counts)
`)).toEqual({ a: 3, b: 1, c: 1 })
`),
).toEqual({ a: 3, b: 1, c: 1 })
})
test("maps serialize to {} at the boundary, like JSON", async () => {
@@ -286,12 +327,14 @@ describe("Map", () => {
describe("Set", () => {
test("add/has/delete/size with chaining", async () => {
expect(await value(`
expect(
await value(`
const s = new Set()
s.add(1).add(2).add(1)
const removed = s.delete(2)
return [s.size, s.has(1), s.has(2), removed]
`)).toEqual([1, true, false, true])
`),
).toEqual([1, true, false, true])
})
test("dedupe idiom: [...new Set(items)]", async () => {
@@ -308,11 +351,13 @@ describe("Set", () => {
})
test("for...of iterates values", async () => {
expect(await value(`
expect(
await value(`
let total = 0
for (const n of new Set([1, 2, 3])) total += n
return total
`)).toBe(6)
`),
).toBe(6)
})
test("sets serialize to {} at the boundary, like JSON", async () => {
@@ -338,21 +383,32 @@ describe("stdlib integration", () => {
})
test("dates inside Map values survive in-sandbox reads", async () => {
expect(await value(`
expect(
await value(`
const m = new Map([["start", new Date(1000)]])
return m.get("start").getTime()
`)).toBe(1000)
`),
).toBe(1000)
})
test("instanceof recognizes the stdlib value types", async () => {
expect(await value(`return [new Date(0) instanceof Date, /a/ instanceof RegExp, new Map() instanceof Map, new Set() instanceof Set]`)).toEqual([true, true, true, true])
expect(await value(`return [[1] instanceof Array, [1] instanceof Object, ({}) instanceof Object, 5 instanceof Object]`)).toEqual([true, true, true, false])
expect(
await value(
`return [new Date(0) instanceof Date, /a/ instanceof RegExp, new Map() instanceof Map, new Set() instanceof Set]`,
),
).toEqual([true, true, true, true])
expect(
await value(`return [[1] instanceof Array, [1] instanceof Object, ({}) instanceof Object, 5 instanceof Object]`),
).toEqual([true, true, true, false])
expect(await value(`return [new Map() instanceof Set, "s" instanceof Date]`)).toEqual([false, false])
expect(await value(`const p = Promise.resolve(1); const isPromise = p instanceof Promise; await p; return isPromise`)).toBe(true)
expect(
await value(`const p = Promise.resolve(1); const isPromise = p instanceof Promise; await p; return isPromise`),
).toBe(true)
})
test("realistic pipeline: parse, extract with regex, dedupe, count by day", async () => {
expect(await value(`
expect(
await value(`
const raw = '[{"at":"2024-01-01T05:00:00Z","tag":"a b"},{"at":"2024-01-01T09:00:00Z","tag":"b c"},{"at":"2024-01-02T01:00:00Z","tag":"a"}]'
const rows = JSON.parse(raw)
const tags = new Set()
@@ -363,27 +419,34 @@ describe("stdlib integration", () => {
byDay.set(day, (byDay.get(day) ?? 0) + 1)
}
return { tags: [...tags].sort((a, b) => (a < b ? -1 : 1)), byDay: Object.fromEntries(byDay) }
`)).toEqual({ tags: ["a", "b", "c"], byDay: { "2024-01-01": 2, "2024-01-02": 1 } })
`),
).toEqual({ tags: ["a", "b", "c"], byDay: { "2024-01-01": 2, "2024-01-02": 1 } })
})
})
describe("sandbox values at intra-sandbox checkpoints", () => {
test("Object.values/entries keep Dates usable", async () => {
expect(await value(`return Object.values({ d: new Date(0) })[0].getTime()`)).toBe(0)
expect(await value(`const [key, d] = Object.entries({ d: new Date(0) })[0]; return key + ":" + d.getTime()`)).toBe("d:0")
expect(await value(`const [key, d] = Object.entries({ d: new Date(0) })[0]; return key + ":" + d.getTime()`)).toBe(
"d:0",
)
})
test("Object.assign keeps Maps usable", async () => {
expect(await value(`const merged = Object.assign({}, { m: new Map([["a", 1]]) }); return merged.m.get("a")`)).toBe(1)
expect(await value(`const merged = Object.assign({}, { m: new Map([["a", 1]]) }); return merged.m.get("a")`)).toBe(
1,
)
})
test("object and array spread keep sandbox values usable", async () => {
expect(await value(`
expect(
await value(`
const src = { m: new Map([["a", 1]]) }
const copy = { ...src }
copy.m.set("b", 2)
return [copy.m.get("a"), src.m.get("b")]
`)).toEqual([1, 2])
`),
).toEqual([1, 2])
expect(await value(`const list = [new Date(1000)]; const copy = [...list]; return copy[0].getTime()`)).toBe(1000)
})
@@ -404,7 +467,10 @@ describe("sandbox values at intra-sandbox checkpoints", () => {
})
test("the host boundary still serializes JSON forms: results, JSON.stringify, and tool arguments", async () => {
expect(await value(`return { d: new Date(0), m: new Map([["a", 1]]) }`)).toEqual({ d: "1970-01-01T00:00:00.000Z", m: {} })
expect(await value(`return { d: new Date(0), m: new Map([["a", 1]]) }`)).toEqual({
d: "1970-01-01T00:00:00.000Z",
m: {},
})
expect(await value(`return JSON.stringify({ d: new Date(0) })`)).toBe('{"d":"1970-01-01T00:00:00.000Z"}')
const observed: Array<unknown> = []
@@ -417,10 +483,12 @@ describe("sandbox values at intra-sandbox checkpoints", () => {
return "ok"
}),
})
const result = await Effect.runPromise(CodeMode.execute({
tools: { host: { capture } },
code: `return await tools.host.capture({ when: new Date(0), tags: new Map([["a", 1]]) })`,
}))
const result = await Effect.runPromise(
CodeMode.execute({
tools: { host: { capture } },
code: `return await tools.host.capture({ when: new Date(0), tags: new Map([["a", 1]]) })`,
}),
)
expect(result.ok).toBe(true)
expect(observed).toStrictEqual([{ when: "1970-01-01T00:00:00.000Z", tags: {} }])
})
@@ -43,7 +43,6 @@ export class Service extends ConfigService.Service<Service>()("@opencode/Runtime
experimentalBackgroundSubagents: enabledByExperimental("OPENCODE_EXPERIMENTAL_BACKGROUND_SUBAGENTS"),
experimentalLspTy: bool("OPENCODE_EXPERIMENTAL_LSP_TY"),
experimentalLspTool: enabledByExperimental("OPENCODE_EXPERIMENTAL_LSP_TOOL"),
experimentalCodeMode: enabledByExperimental("OPENCODE_EXPERIMENTAL_CODE_MODE"),
experimentalOxfmt: enabledByExperimental("OPENCODE_EXPERIMENTAL_OXFMT"),
experimentalPlanMode: enabledByExperimental("OPENCODE_EXPERIMENTAL_PLAN_MODE"),
experimentalEventSystem: enabledByExperimental("OPENCODE_EXPERIMENTAL_EVENT_SYSTEM"),
+12 -25
View File
@@ -1,7 +1,6 @@
import path from "node:path"
import { pathToFileURL } from "node:url"
import { LayerNode } from "@opencode-ai/core/effect/layer-node"
import { type Tool } from "ai"
import { ConfigV1 } from "@opencode-ai/core/v1/config/config"
import { serviceUse } from "@opencode-ai/core/effect/service-use"
import { Client, type ClientOptions } from "@modelcontextprotocol/sdk/client/index.js"
@@ -154,17 +153,19 @@ export interface ServerInstructions {
tools: string[]
}
/** An MCP tool in its native shape; consumers adapt it to their own tool format. */
export interface McpTool {
/** Shared cached definition; consumers must copy rather than mutate it. */
readonly def: MCPToolDef
readonly client: MCPClient
readonly timeout?: number
}
export interface Interface {
readonly status: () => Effect.Effect<Record<string, Status>>
readonly clients: () => Effect.Effect<Record<string, MCPClient>>
readonly instructions: () => Effect.Effect<ServerInstructions[]>
readonly tools: () => Effect.Effect<Record<string, Tool>>
/**
* Raw MCP tool definitions keyed identically to {@link tools} (`toolName(client, name)`).
* Unlike {@link tools}, these retain the original `inputSchema`/`outputSchema`, which code
* mode uses to render tool signatures (including return types) to the model.
*/
readonly defs: () => Effect.Effect<Record<string, MCPToolDef>>
readonly tools: () => Effect.Effect<Record<string, McpTool>>
readonly prompts: () => Effect.Effect<Record<string, PromptInfo & { client: string }>>
readonly resources: (clientName?: string) => Effect.Effect<Record<string, ResourceInfo & { client: string }>>
readonly resourceTemplates: (
@@ -662,7 +663,7 @@ const layer = Layer.effect(
}
const tools = Effect.fn("MCP.tools")(function* () {
const result: Record<string, Tool> = {}
const result: Record<string, McpTool> = {}
const s = yield* InstanceState.get(state)
const cfg = yield* cfgSvc.get()
@@ -678,21 +679,8 @@ const layer = Layer.effect(
continue
}
const timeout = requestTimeout(s, clientName, mcpConfig, defaultTimeout)
for (const mcpTool of listed) {
const key = McpCatalog.toolName(clientName, mcpTool.name)
result[key] = McpCatalog.convertTool(mcpTool, client, timeout)
}
}
return result
})
const defs = Effect.fn("MCP.defs")(function* () {
const result: Record<string, MCPToolDef> = {}
const s = yield* InstanceState.get(state)
for (const [clientName, listed] of Object.entries(s.defs)) {
if (s.status[clientName]?.status !== "connected") continue
for (const mcpTool of listed) {
result[McpCatalog.toolName(clientName, mcpTool.name)] = mcpTool
for (const def of listed) {
result[McpCatalog.toolName(clientName, def.name)] = { def, client, timeout }
}
}
return result
@@ -1000,7 +988,6 @@ const layer = Layer.effect(
clients,
instructions,
tools,
defs,
prompts,
resources,
resourceTemplates,
-65
View File
@@ -1,65 +0,0 @@
import type { ToolExecutionOptions } from "ai"
import { Effect } from "effect"
import type { Plugin } from "@/plugin"
import type { Tool } from "@/tool/tool"
/**
* The shared middle of every raw MCP tool invocation: plugin `tool.execute.before`
* hook permission ask dispatch through the ai-sdk tool's execute inside the
* `Tool.execute` tracing span plugin `tool.execute.after` hook. Used by both the
* legacy per-tool registration in `SessionTools.resolve` and code-mode child calls,
* so MCP tools execute identically on either path.
*
* Returns the RAW result the ai-sdk execute resolved with callers own their
* shaping edge (model-facing text/attachment shaping + truncation on the legacy
* path, `toSandboxResult` for code-mode child calls). The after hook fires here
* with that same raw result, which is exactly what the legacy loop always passed
* (the raw MCP result, not the shaped `{title, output, metadata}`), so the hook
* payload cannot drift between callers.
*
* `callID` is the hook/span identity an opaque string nothing parses. Legacy
* passes the ai-sdk `toolCallId`; code-mode child calls pass a synthetic
* `${parentCallID}/${n}`. `options.toolCallId` is what the ai-sdk execute sees and
* stays each caller's existing value. Failure semantics belong to the caller: hook
* failures, permission denials, and tool failures all propagate the legacy path
* lets them fail the tool call as before; code mode converts them into catchable
* in-program tool errors at its edge.
*/
export const invoke = Effect.fn("McpInvoke.invoke")(function* <R>(input: {
plugin: Plugin.Interface
key: string
execute: (args: any, options: ToolExecutionOptions) => R | PromiseLike<R>
args: any
callID: string
options: ToolExecutionOptions
sessionID: string
messageID: string
ask: Tool.Context["ask"]
}) {
yield* input.plugin.trigger(
"tool.execute.before",
{ tool: input.key, sessionID: input.sessionID, callID: input.callID },
{ args: input.args },
)
const result: R = yield* Effect.gen(function* () {
yield* input.ask({ permission: input.key, metadata: {}, patterns: ["*"], always: ["*"] })
return yield* Effect.promise(() => Promise.resolve(input.execute(input.args, input.options)))
}).pipe(
Effect.withSpan("Tool.execute", {
attributes: {
"tool.name": input.key,
"tool.call_id": input.callID,
"session.id": input.sessionID,
"message.id": input.messageID,
},
}),
)
yield* input.plugin.trigger(
"tool.execute.after",
{ tool: input.key, sessionID: input.sessionID, callID: input.callID, args: input.args },
result,
)
return result
})
export * as McpInvoke from "./invoke"
@@ -213,13 +213,6 @@ export function disabled(tools: string[], ruleset: PermissionV1.Ruleset): Set<st
)
}
/**
* The shared tool-visibility predicate: drop every tool a hard deny hides
* ({@link disabled} semantics a matching `deny` rule with pattern `"*"`).
* Ask-level rules leave a tool fully visible and callable (it prompts at call
* time). Used both when preparing the LLM tool list (request prep) and when
* building/dispatching the code-mode MCP catalog, so the two cannot drift.
*/
export function visibleTools<T>(tools: Record<string, T>, ruleset: PermissionV1.Ruleset): Record<string, T> {
const hidden = disabled(Object.keys(tools), ruleset)
return Object.fromEntries(Object.entries(tools).filter(([name]) => !hidden.has(name)))
+3 -3
View File
@@ -206,11 +206,11 @@ export const prepare = Effect.fn("LLMRequestPrep.prepare")(function* (input: Pre
})
function resolveTools(input: Pick<PrepareInput, "tools" | "agent" | "permission" | "user">) {
const visible = Permission.visibleTools(
input.tools,
const disabled = Permission.disabled(
Object.keys(input.tools),
Permission.merge(input.agent.permission, input.permission ?? []),
)
return Record.filter(visible, (_, k) => input.user.tools?.[k] !== false)
return Record.filter(input.tools, (_, k) => input.user.tools?.[k] !== false && !disabled.has(k))
}
export function hasToolCalls(messages: ModelMessage[]): boolean {
-2
View File
@@ -1237,8 +1237,6 @@ const layer = Layer.effect(
Effect.provideService(ToolRegistry.Service, registry),
Effect.provideService(MCP.Service, mcp),
Effect.provideService(Truncate.Service, truncate),
Effect.provideService(Agent.Service, agents),
Effect.provideService(RuntimeFlags.Service, flags),
)
if (lastUser.format?.type === "json_schema") {
+28 -30
View File
@@ -3,7 +3,7 @@ import { SessionV1 } from "@opencode-ai/core/v1/session"
import { Provider } from "@/provider/provider"
import { ProviderTransform } from "@/provider/transform"
import { MCP } from "@/mcp"
import { McpInvoke } from "@/mcp/invoke"
import { McpCatalog } from "@/mcp/catalog"
import { Permission } from "@/permission"
import { Tool } from "@/tool/tool"
import { ToolJsonSchema } from "@/tool/json-schema"
@@ -22,7 +22,6 @@ import { EffectBridge } from "@/effect/bridge"
import { ProviderV2 } from "@opencode-ai/core/provider"
import { ModelV2 } from "@opencode-ai/core/model"
import { isRecord } from "@/util/record"
import { RuntimeFlags } from "@/effect/runtime-flags"
const MCP_RESOURCE_TOOLS = {
list: "list_mcp_resources",
@@ -54,7 +53,6 @@ export const resolve = Effect.fn("SessionTools.resolve")(function* (input: {
const registry = yield* ToolRegistry.Service
const mcp = yield* MCP.Service
const truncate = yield* Truncate.Service
const flags = yield* RuntimeFlags.Service
const context = (args: Record<string, unknown>, options: ToolExecutionOptions): Tool.Context => ({
sessionID: input.session.id,
@@ -89,20 +87,11 @@ export const resolve = Effect.fn("SessionTools.resolve")(function* (input: {
.pipe(Effect.orDie),
})
const mcpTools = yield* mcp.tools()
// When code mode is enabled and MCP tools are present, the registry exposes them
// through the single code-mode `execute` tool (ToolRegistry.tools), so raw per-MCP
// registration is suppressed via the early return below. Code mode is experimental
// and off by default.
const codeMode = flags.experimentalCodeMode && Object.keys(mcpTools).length > 0
const registryTools = yield* registry.tools({
for (const item of yield* registry.tools({
modelID: ModelV2.ID.make(input.model.api.id),
providerID: input.model.providerID,
agent: input.agent,
permission: input.session.permission,
})
for (const item of registryTools) {
})) {
const schema = ProviderTransform.schema(input.model, ToolJsonSchema.fromTool(item))
tools[item.id] = tool({
description: item.description,
@@ -393,9 +382,8 @@ export const resolve = Effect.fn("SessionTools.resolve")(function* (input: {
})
}
if (codeMode) return tools
for (const [key, item] of Object.entries(mcpTools)) {
for (const [key, entry] of Object.entries(yield* mcp.tools())) {
const item = McpCatalog.convertTool(entry.def, entry.client, entry.timeout)
const execute = item.execute
if (!execute) continue
@@ -406,19 +394,29 @@ export const resolve = Effect.fn("SessionTools.resolve")(function* (input: {
run.promise(
Effect.gen(function* () {
const ctx = context(args, opts)
// Shared MCP middle (before hook → permission ask → Tool.execute span →
// dispatch → after hook); this caller keeps the model-facing shaping edge below.
const result: Awaited<ReturnType<NonNullable<typeof execute>>> = yield* McpInvoke.invoke({
plugin,
key,
execute,
args,
callID: opts.toolCallId,
options: opts,
sessionID: ctx.sessionID,
messageID: input.processor.message.id,
ask: ctx.ask,
})
yield* plugin.trigger(
"tool.execute.before",
{ tool: key, sessionID: ctx.sessionID, callID: opts.toolCallId },
{ args },
)
const result: Awaited<ReturnType<NonNullable<typeof execute>>> = yield* Effect.gen(function* () {
yield* ctx.ask({ permission: key, metadata: {}, patterns: ["*"], always: ["*"] })
return yield* Effect.promise(() => execute(args, opts))
}).pipe(
Effect.withSpan("Tool.execute", {
attributes: {
"tool.name": key,
"tool.call_id": opts.toolCallId,
"session.id": ctx.sessionID,
"message.id": input.processor.message.id,
},
}),
)
yield* plugin.trigger(
"tool.execute.after",
{ tool: key, sessionID: ctx.sessionID, callID: opts.toolCallId, args },
result,
)
const textParts: string[] = []
const attachments: Omit<SessionV1.FilePart, "id" | "sessionID" | "messageID">[] = []
+141 -247
View File
@@ -1,6 +1,5 @@
import * as Tool from "./tool"
import type { Tool as AITool } from "ai"
import type { Tool as MCPToolDef } from "@modelcontextprotocol/sdk/types.js"
import { CallToolResultSchema, type CallToolResult } from "@modelcontextprotocol/sdk/types.js"
import { Cause, Effect, Schema } from "effect"
import {
CodeMode,
@@ -12,7 +11,6 @@ import {
} from "@opencode-ai/codemode"
import { MCP } from "@/mcp"
import { McpCatalog } from "@/mcp/catalog"
import { McpInvoke } from "@/mcp/invoke"
import { Agent } from "@/agent/agent"
import { Session } from "@/session/session"
import { Permission } from "@/permission"
@@ -20,18 +18,6 @@ import { Plugin } from "@/plugin"
export const CODE_MODE_TOOL = "execute"
// OpenCode sets NO execution limits: no timeout, no tool-call cap, and no CodeMode output
// truncation. Cancelling the tool call aborts `ctx.abort`, which wins the race below and
// interrupts the execution fiber — structured concurrency takes the program and its
// in-flight child calls down with it; every child call is permission-gated anyway. Output
// bounding is OpenCode's native tool-output truncation (Tool.define's shared wrapper),
// which applies to `execute` like any other tool and dumps the full output to a file when
// it triggers.
// The static base description. The full usage guide and the grouped, permission-filtered
// tool catalog are appended per agent by the registry (`describeCodeMode`, the same
// composition point `describeTask` uses), so `plugin.trigger("tool.definition")` sees this
// base first, exactly like the task tool.
const DESCRIPTION = [
"Execute a JavaScript/TypeScript program that orchestrates the connected MCP tools inside a confined runtime.",
"The full usage guide and the catalog of available tools follow below.",
@@ -47,119 +33,51 @@ export const Parameters = Schema.Struct({
}),
})
/** One child tool call, surfaced live so the UI can render a per-call line that
* updates as the program runs. `tool` is the dotted path (e.g. `github.create_issue`). */
export type CallEntry = { tool: string; status: "running" | "completed" | "error"; input?: Record<string, unknown> }
type CallEntry = { tool: string; status: "running" | "completed" | "error"; input?: Record<string, unknown> }
type Metadata = {
toolCalls: CallEntry[]
error?: boolean
}
/**
* A tool-result attachment: identical to a session `FilePart` (minus the ids) and
* carrying the actual bytes (`url`, often a base64 `data:` URL), so it lowers 1:1 into
* `Tool.ExecuteResult.attachments`. Attachments never enter the sandbox media stripped
* from child tool results is accumulated host-side and returned on the outer `execute`
* result, where the existing attachment plumbing turns it into visible images/files.
*/
export type Attachment = NonNullable<Tool.ExecuteResult["attachments"]>[number]
type Attachment = NonNullable<Tool.ExecuteResult["attachments"]>[number]
/** One MCP tool in the grouped catalog: the flat `server_tool` key split into its
* namespace (`server`) and local name, with the raw JSON Schemas used for rendering. */
export type CatalogEntry = {
type CatalogEntry = {
path: string
key: string
server: string
local: string
description: string
tool: AITool
inputSchema: JsonSchema
outputSchema?: JsonSchema
tool: MCP.McpTool
}
/** Render-only cast: MCP definitions carry JSON Schema documents already. */
const toJsonSchema = (schema: unknown): JsonSchema => schema as JsonSchema
/** The input schema for entries without a cached MCP definition, recovered from the
* ai-sdk tool when possible so signatures stay informative. */
function fallbackInputSchema(tool: AITool): JsonSchema {
const schema = (tool.inputSchema as { jsonSchema?: unknown } | undefined)?.jsonSchema
if (schema && typeof schema === "object") return toJsonSchema(schema)
return { type: "object", properties: {} }
}
/**
* Group the flat `server_tool` catalog into per-server namespaces. `servers` are
* the sanitized MCP client names; the longest matching prefix wins so a server
* named `a_b` beats `a` for the key `a_b_tool`. `mcpDefs` carries the raw MCP
* definitions (keyed identically) so each entry retains its original
* `inputSchema`/`outputSchema` for signature rendering.
*/
export function groupByServer(
mcpTools: Record<string, AITool>,
servers: readonly string[],
mcpDefs: Record<string, MCPToolDef> = {},
): Map<string, CatalogEntry[]> {
function groupByServer(mcpTools: Record<string, MCP.McpTool>, servers: readonly string[]): Map<string, CatalogEntry[]> {
const byLongest = [...servers].sort((a, b) => b.length - a.length)
const groups = new Map<string, CatalogEntry[]>()
for (const key of Object.keys(mcpTools).sort((a, b) => a.localeCompare(b))) {
const server = byLongest.find((name) => key.startsWith(name + "_")) ?? (key.includes("_") ? key.slice(0, key.indexOf("_")) : key)
const server =
byLongest.find((name) => key.startsWith(name + "_")) ?? (key.includes("_") ? key.slice(0, key.indexOf("_")) : key)
const local = server && key.startsWith(server + "_") ? key.slice(server.length + 1) : key
const def = mcpDefs[key]
const entry: CatalogEntry = {
path: `${server}.${local}`,
key,
server,
local,
description: mcpTools[key]!.description ?? def?.description ?? "",
tool: mcpTools[key]!,
inputSchema: def?.inputSchema ? toJsonSchema(def.inputSchema) : fallbackInputSchema(mcpTools[key]!),
...(def?.outputSchema ? { outputSchema: toJsonSchema(def.outputSchema) } : {}),
}
groups.set(server, [...(groups.get(server) ?? []), entry])
}
return groups
}
/** The executable catalog for a (already permission-filtered) MCP tool set: grouped
* entries, minus any without an ai-sdk execute function. */
export function buildCatalog(
mcpTools: Record<string, AITool>,
mcpDefs: Record<string, MCPToolDef>,
servers: readonly string[],
): CatalogEntry[] {
return [...groupByServer(mcpTools, servers, mcpDefs).values()].flat().filter((entry) => entry.tool.execute !== undefined)
}
/**
* The model-facing usage guide plus grouped catalog for the given MCP tool set: the
* CodeMode instructions for this tool tree (syntax guide + tool signatures, or the
* namespace overview + search for large catalogs). Callers pass an already
* permission-filtered tool set hard-denied tools never enter the catalog. The preview
* tree's runs are placeholders rendering never invokes them.
*/
export function catalogInstructions(
mcpTools: Record<string, AITool>,
mcpDefs: Record<string, MCPToolDef>,
servers: readonly string[],
): string {
const catalog = buildCatalog(mcpTools, mcpDefs, servers)
export function describeCatalog(mcpTools: Record<string, MCP.McpTool>, servers: readonly string[]): string {
return CodeMode.make({
tools: toolTree(catalog, () => () => Effect.fail(toolError("Tool preview is not executable."))),
tools: toolTree(
[...groupByServer(mcpTools, servers).values()].flat(),
() => () => Effect.fail(toolError("Tool preview is not executable.")),
),
}).instructions()
}
function displayInput(input: unknown): Record<string, unknown> | undefined {
if (input === null || input === undefined) return
if (typeof input === "object" && !Array.isArray(input)) {
const value = input as Record<string, unknown>
if (Object.keys(value).length > 0) return value
return
}
return { input }
}
const lastSegment = (uri: string) => {
const trimmed = uri.split(/[?#]/, 1)[0]!.replace(/\/+$/, "")
const segment = trimmed.slice(trimmed.lastIndexOf("/") + 1)
@@ -168,24 +86,7 @@ const lastSegment = (uri: string) => {
const dataUrl = (mime: string, base64: string) => `data:${mime};base64,${base64}`
/** The stand-in payload for a media-only tool result, so the program knows the call
* succeeded even though the media itself never enters the sandbox. */
const mediaMarker = (files: number, images: number) => {
const noun = files === images ? "image" : "file"
return `[${files} ${noun}${files === 1 ? "" : "s"} attached to the result]`
}
/**
* Reduce a raw MCP tool result to the value the sandbox sees. Structured content is
* preferred; otherwise text blocks are joined. Media blocks (image/audio/resource
* blob/resource_link) NEVER enter the sandbox: they are stripped into `collect`, the
* per-execution attachment accumulator, and a tool that returned ONLY media yields a
* small text marker instead. Lenient never throws on unexpected shapes.
*/
export function toSandboxResult(raw: unknown, collect: (attachment: Attachment) => void): unknown {
if (raw === null || typeof raw !== "object") return raw
const record = raw as { structuredContent?: unknown; content?: unknown }
const content = Array.isArray(record.content) ? record.content : []
function projectMcpResult(result: CallToolResult, collect: (attachment: Attachment) => void): unknown {
const text: string[] = []
let files = 0
let images = 0
@@ -194,105 +95,109 @@ export function toSandboxResult(raw: unknown, collect: (attachment: Attachment)
if (attachment.mime.startsWith("image/")) images += 1
collect(attachment)
}
for (const item of content) {
if (!item || typeof item !== "object") continue
const block = item as Record<string, unknown>
for (const block of result.content) {
switch (block.type) {
case "text":
if (typeof block.text === "string") text.push(block.text)
text.push(block.text)
break
case "image":
case "audio":
if (typeof block.data === "string" && typeof block.mimeType === "string") {
push({ type: "file", mime: block.mimeType, url: dataUrl(block.mimeType, block.data) })
}
push({ type: "file", mime: block.mimeType, url: dataUrl(block.mimeType, block.data) })
break
case "resource": {
const res = block.resource as Record<string, unknown> | undefined
if (res && typeof res === "object") {
const mime = typeof res.mimeType === "string" ? res.mimeType : "application/octet-stream"
const uri = typeof res.uri === "string" ? res.uri : undefined
if (typeof res.blob === "string") {
push({ type: "file", mime, url: dataUrl(mime, res.blob), filename: uri ? lastSegment(uri) : undefined })
} else if (typeof res.text === "string") {
text.push(res.text)
}
if ("text" in block.resource) {
text.push(block.resource.text)
break
}
const mime = block.resource.mimeType ?? "application/octet-stream"
push({ type: "file", mime, url: dataUrl(mime, block.resource.blob), filename: lastSegment(block.resource.uri) })
break
}
case "resource_link":
if (typeof block.uri === "string") {
push({
type: "file",
mime: typeof block.mimeType === "string" ? block.mimeType : "application/octet-stream",
url: block.uri,
filename: typeof block.name === "string" ? block.name : lastSegment(block.uri),
})
}
// A link is a reference, not fetchable media; hand it to the program instead of the attachment channel.
text.push(`${block.name}: ${block.uri}`)
break
}
}
if (record.structuredContent !== undefined && record.structuredContent !== null) return record.structuredContent
if (result.structuredContent !== undefined && result.structuredContent !== null) return result.structuredContent
if (text.length > 0) return text.join("\n")
if (files > 0) return mediaMarker(files, images)
if (Array.isArray(record.content)) return null // MCP-shaped result with nothing extractable
return raw
}
/**
* Append captured `console.*` output to the model-facing text as a trailing `Logs:` section,
* so a program's diagnostics ride back alongside its result on success AND on error.
* Returns the text unchanged when nothing was logged. This is the sandbox's only
* stdout-like channel it goes to the model, not the user.
*/
export function withLogs(output: string, logs: ReadonlyArray<string> = []): string {
if (logs.length === 0) return output
const section = "Logs:\n" + logs.join("\n")
return output.length > 0 ? `${output}\n\n${section}` : section
}
/** Coerce the program's return value to model-facing text without ever failing on shape. */
export function formatValue(value: unknown): string {
if (typeof value === "string") return value
if (value === undefined) return "undefined"
try {
return JSON.stringify(value, null, 2) ?? String(value)
} catch {
return String(value)
if (files > 0) {
const noun = files === images ? "image" : "file"
return `[${files} ${noun}${files === 1 ? "" : "s"} attached to the result]`
}
return null
}
type Run = (input: unknown) => Effect.Effect<unknown, unknown>
/** Build the `tools.<server>.<tool>` tree CodeMode executes against, one
* `Tool.make` definition per MCP tool with its render-only JSON Schemas. */
function toolTree(catalog: readonly CatalogEntry[], run: (entry: CatalogEntry) => Run) {
const tree: Record<string, Record<string, ToolDefinition>> = {}
for (const entry of catalog) {
const namespace = (tree[entry.server] ??= {})
namespace[entry.local] = SandboxTool.make({
description: entry.description,
input: entry.inputSchema,
output: entry.outputSchema,
description: entry.tool.def.description ?? "",
input: entry.tool.def.inputSchema as JsonSchema,
output: entry.tool.def.outputSchema as JsonSchema | undefined,
run: run(entry),
})
}
return tree
}
/** Failures inside a child call plugin hook failures, permission denials, and tool
* failures alike become safe, catchable in-program errors via toolError, so a
* program can try/catch one call without the whole execution dying. Interruption
* (user cancel) keeps propagating as interruption. */
const toCatchable = <A, E, R>(effect: Effect.Effect<A, E, R>) =>
effect.pipe(
Effect.catchCause((cause) => {
if (Cause.hasInterruptsOnly(cause)) return Effect.interrupt
const error = Cause.squash(cause)
return Effect.fail(toolError(error instanceof Error ? error.message : String(error), error))
const invokeChildTool = Effect.fn("CodeMode.invokeChildTool")(function* (input: {
plugin: Plugin.Interface
entry: CatalogEntry
args: Record<string, unknown>
callID: string
ctx: Tool.Context
}) {
yield* input.plugin.trigger(
"tool.execute.before",
{ tool: input.entry.key, sessionID: input.ctx.sessionID, callID: input.callID },
{ args: input.args },
)
const result: CallToolResult = yield* Effect.gen(function* () {
yield* input.ctx.ask({ permission: input.entry.key, metadata: {}, patterns: ["*"], always: ["*"] })
// Deliberately mirrors McpCatalog.convertTool's transport call so the MCP service stays free of tool-loop concerns.
return yield* Effect.promise(async () => {
const raw = await input.entry.tool.client.callTool(
{ name: input.entry.tool.def.name, arguments: input.args },
CallToolResultSchema,
{
resetTimeoutOnProgress: true,
signal: input.ctx.abort,
timeout: input.entry.tool.timeout,
// The MCP SDK only sends a progress token when this hook is present, enabling timeout resets.
onprogress: () => {},
},
)
if (raw.isError)
throw new Error(
raw.content
.flatMap((item) => (item.type === "text" ? [item.text] : []))
.filter((text) => text.trim())
.join("\n\n") || "MCP tool returned an error",
)
return raw
})
}).pipe(
Effect.withSpan("Tool.execute", {
attributes: {
"tool.name": input.entry.key,
"tool.call_id": input.callID,
"session.id": input.ctx.sessionID,
"message.id": input.ctx.messageID,
},
}),
)
yield* input.plugin.trigger(
"tool.execute.after",
{ tool: input.entry.key, sessionID: input.ctx.sessionID, callID: input.callID, args: input.args },
result,
)
return result
})
export const CodeModeTool = Tool.define(
CODE_MODE_TOOL,
@@ -306,8 +211,6 @@ export const CodeModeTool = Tool.define(
description: DESCRIPTION,
parameters: Parameters,
execute: Effect.fn("CodeMode.execute")(function* (params, ctx) {
// Already cancelled: don't start the program at all. (The mid-flight case is the
// race below; racing alone would still let the program run its first steps.)
if (ctx.abort.aborted) {
return {
title: CODE_MODE_TOOL,
@@ -315,53 +218,35 @@ export const CodeModeTool = Tool.define(
output: "Execution cancelled.",
} satisfies Tool.ExecuteResult<Metadata>
}
// A fresh MCP snapshot per execution, so the runtime tracks live tool-list
// changes, filtered with the same merged agent+session ruleset that gates
// `ctx.ask` (see SessionTools.context). A hard-denied tool never enters the
// tree, so it is not dispatchable even if the model guesses its name — the
// program gets the normal unknown-tool diagnostic, not a permission error.
const agent = yield* agents.get(ctx.agent)
const session = yield* sessions.get(ctx.sessionID).pipe(Effect.orDie)
const ruleset = Permission.merge(agent.permission, session.permission ?? [])
const mcpTools = Permission.visibleTools(yield* mcp.tools(), ruleset)
const servers = Object.keys(yield* mcp.clients()).map(McpCatalog.sanitize)
const catalog = buildCatalog(mcpTools, yield* mcp.defs(), servers)
const catalog = [...groupByServer(mcpTools, servers).values()].flat()
const calls: CallEntry[] = []
// Media stripped from child tool results accumulates here for the life of the
// call; the bytes never enter the sandbox (see toSandboxResult).
const attachments: Attachment[] = []
const collect = (attachment: Attachment) => void attachments.push(attachment)
// Stream the current call list to the UI. Sent on every status change so the
// tool part shows each child call appearing and resolving while the program runs.
const publish = () => ctx.metadata({ title: CODE_MODE_TOOL, metadata: { toolCalls: calls.map((c) => ({ ...c })) } })
const publish = () =>
ctx.metadata({ title: CODE_MODE_TOOL, metadata: { toolCalls: calls.map((c) => ({ ...c })) } })
// One CodeMode tool per MCP tool, running the same shared middle as legacy
// per-tool registration (McpInvoke.invoke: plugin before hook → permission
// ask → Tool.execute span → dispatch through the ai-sdk wrapper, which owns
// callTool timeouts/progress and turns an MCP isError into a thrown Error →
// plugin after hook), so plugins observe child calls too. Each child gets a
// synthetic hook/span callID `${parentCallID}/${n}` (per-execution counter,
// opaque — nothing parses it); the ai-sdk toolCallId is unchanged. Failures —
// hook, denial, or tool — fail only that child call as a safe, catchable
// in-program error (toCatchable); the raw result is then shaped for the sandbox.
let childCalls = 0
const callTool = (entry: CatalogEntry) => (input: unknown) =>
toCatchable(
Effect.gen(function* () {
childCalls += 1
const raw = yield* McpInvoke.invoke({
plugin,
key: entry.key,
execute: entry.tool.execute!,
args: input ?? {},
callID: `${ctx.callID ?? entry.key}/${childCalls}`,
options: { toolCallId: ctx.callID ?? entry.key, abortSignal: ctx.abort, messages: [] },
sessionID: ctx.sessionID,
messageID: ctx.messageID,
ask: ctx.ask,
})
return toSandboxResult(raw, collect)
Effect.gen(function* () {
childCalls += 1
const result = yield* invokeChildTool({
plugin,
entry,
args: (input ?? {}) as Record<string, unknown>,
callID: `${ctx.callID ?? entry.key}/${childCalls}`,
ctx,
})
return projectMcpResult(result, (attachment: Attachment) => void attachments.push(attachment))
}).pipe(
Effect.catchCause((cause) => {
if (Cause.hasInterruptsOnly(cause)) return Effect.interrupt
const error = Cause.squash(cause)
return Effect.fail(toolError(error instanceof Error ? error.message : String(error), error))
}),
)
@@ -369,7 +254,14 @@ export const CodeModeTool = Tool.define(
tools: toolTree(catalog, callTool),
onToolCallStart: ({ index, name, input }) =>
Effect.suspend(() => {
const shown = displayInput(input)
const shown = (() => {
if (input === null || input === undefined) return
if (typeof input === "object" && !Array.isArray(input)) {
const value = input as Record<string, unknown>
return Object.keys(value).length > 0 ? value : undefined
}
return { input }
})()
calls[index] = { tool: name, status: "running", ...(shown ? { input: shown } : {}) }
return publish()
}),
@@ -381,44 +273,46 @@ export const CodeModeTool = Tool.define(
}),
})
// The shared tool runner does not wire ctx.abort to fiber interruption (it runs
// tools via Effect.runPromise with no abort handling), so without this race the
// program would keep running after the user cancels. The abort signal winning the
// race interrupts the execution fiber; the cancelled result keeps the runner's
// post-abort bookkeeping (completeToolCall) on its normal path.
const cancelled = Effect.callback<ExecuteResult>((resume) => {
const onAbort = () =>
resume(
Effect.succeed<ExecuteResult>({
ok: false,
error: { kind: "ExecutionFailure", message: "Execution cancelled." },
toolCalls: calls.map((call) => ({ name: call.tool })),
}),
)
if (ctx.abort.aborted) return onAbort()
ctx.abort.addEventListener("abort", onAbort, { once: true })
return Effect.sync(() => ctx.abort.removeEventListener("abort", onAbort))
const abort = Effect.callback<void>((resume) => {
if (ctx.abort.aborted) return resume(Effect.void)
const handler = () => resume(Effect.void)
ctx.abort.addEventListener("abort", handler, { once: true })
return Effect.sync(() => ctx.abort.removeEventListener("abort", handler))
})
const cancelled = (): ExecuteResult => ({
ok: false,
error: { kind: "ExecutionFailure", message: "Execution cancelled." },
toolCalls: calls.map((call) => ({ name: call.tool })),
})
const result = yield* Effect.raceFirst(runtime.execute(params.code), cancelled)
const result = yield* Effect.raceFirst(runtime.execute(params.code), abort.pipe(Effect.map(cancelled)))
const logs = result.logs ?? []
const attached = attachments.length > 0 ? { attachments } : {}
const hints = result.ok
? []
: (result.error.suggestions ?? []).filter((hint) => !result.error.message.includes(hint))
const metadata: Metadata = result.ok ? { toolCalls: calls } : { toolCalls: calls, error: true }
let output: string
if (result.ok) {
return {
title: CODE_MODE_TOOL,
metadata: { toolCalls: calls },
output: withLogs(formatValue(result.value), logs),
...attached,
} satisfies Tool.ExecuteResult<Metadata>
if (typeof result.value === "string") output = result.value
else if (result.value === undefined) output = "undefined"
else {
try {
output = JSON.stringify(result.value, null, 2) ?? String(result.value)
} catch {
output = String(result.value)
}
}
} else {
output = [result.error.message, ...hints].join("\n")
}
// Diagnostics may carry suggestions (e.g. pointing an unknown tool at
// discovery); append the ones the message doesn't already contain.
const hints = (result.error.suggestions ?? []).filter((hint) => !result.error.message.includes(hint))
if (logs.length > 0)
output = output.length > 0 ? `${output}\n\nLogs:\n${logs.join("\n")}` : `Logs:\n${logs.join("\n")}`
return {
title: CODE_MODE_TOOL,
metadata: { toolCalls: calls, error: true },
output: withLogs([result.error.message, ...hints].join("\n"), logs),
metadata,
output,
...attached,
} satisfies Tool.ExecuteResult<Metadata>
}),
+1 -30
View File
@@ -26,9 +26,6 @@ import { Plugin } from "../plugin"
import { Provider } from "@/provider/provider"
import { WebSearchTool } from "./websearch"
import { CodeModeTool, catalogInstructions } from "./code-mode"
import { MCP } from "@/mcp"
import { McpCatalog } from "@/mcp/catalog"
import { LspTool } from "./lsp"
import * as Truncate from "./truncate"
import { ApplyPatchTool } from "./apply_patch"
@@ -54,7 +51,6 @@ import { BackgroundJob } from "@/background/job"
import { RuntimeFlags } from "@/effect/runtime-flags"
import { ProviderV2 } from "@opencode-ai/core/provider"
import { ModelV2 } from "@opencode-ai/core/model"
import type { PermissionV1 } from "@opencode-ai/core/v1/permission"
export function webSearchEnabled(providerID: ProviderV2.ID, flags = { exa: false, parallel: false }) {
return providerID === ProviderV2.ID.opencode || flags.exa || flags.parallel
@@ -78,7 +74,6 @@ export interface Interface {
providerID: ProviderV2.ID
modelID: ModelV2.ID
agent: Agent.Info
permission?: PermissionV1.Ruleset
}) => Effect.Effect<Tool.Def[]>
}
@@ -92,7 +87,6 @@ const layer = Layer.effect(
const agents = yield* Agent.Service
const truncate = yield* Truncate.Service
const flags = yield* RuntimeFlags.Service
const mcp = yield* MCP.Service
const invalid = yield* InvalidTool
const task = yield* TaskTool
@@ -110,7 +104,6 @@ const layer = Layer.effect(
const greptool = yield* GrepTool
const patchtool = yield* ApplyPatchTool
const skilltool = yield* SkillTool
const codemode = yield* CodeModeTool
const agent = yield* Agent.Service
const state = yield* InstanceState.make<State>(
@@ -218,7 +211,6 @@ const layer = Layer.effect(
question: Tool.init(question),
lsp: Tool.init(lsptool),
plan: Tool.init(plan),
codemode: Tool.init(codemode),
})
return {
@@ -240,7 +232,6 @@ const layer = Layer.effect(
tool.patch,
...(flags.experimentalLspTool ? [tool.lsp] : []),
...(flags.experimentalPlanMode && flags.client === "cli" ? [tool.plan] : []),
...(flags.experimentalCodeMode ? [tool.codemode] : []),
],
task: tool.task,
read: tool.read,
@@ -272,25 +263,11 @@ const layer = Layer.effect(
return ["Available agent types and the tools they have access to:", description].join("\n")
})
// The grouped MCP-tool catalog appended to the code-mode base description, built
// fresh per turn so it tracks live tool-list changes. Hard-denied tools (the shared
// Permission.visibleTools predicate over the agent's ruleset) never enter the
// catalog, its inlined signatures, or the in-program search index.
const describeCodeMode = Effect.fn("ToolRegistry.describeCodeMode")(function* (agent: Agent.Info, permission?: PermissionV1.Ruleset) {
const visible = Permission.visibleTools(yield* mcp.tools(), Permission.merge(agent.permission, permission ?? []))
const servers = Object.keys(yield* mcp.clients()).map(McpCatalog.sanitize)
return catalogInstructions(visible, yield* mcp.defs(), servers)
})
const tools: Interface["tools"] = Effect.fn("ToolRegistry.tools")(function* (input) {
// Consulted once before the synchronous filter: code mode registers only when the
// experimental flag is on AND at least one MCP tool is connected.
const mcpToolCount = flags.experimentalCodeMode ? Object.keys(yield* mcp.tools()).length : 0
const filtered = (yield* all()).filter((tool) => {
if (tool.id === WebSearchTool.id) {
return webSearchEnabled(input.providerID, { exa: flags.enableExa, parallel: flags.enableParallel })
}
if (tool.id === CodeModeTool.id) return flags.experimentalCodeMode && mcpToolCount > 0
const usePatch =
input.modelID.includes("gpt-") && !input.modelID.includes("oss") && !input.modelID.includes("gpt-4")
@@ -315,11 +292,7 @@ const layer = Layer.effect(
: undefined
return {
id: tool.id,
description: [
output.description,
tool.id === TaskTool.id ? yield* describeTask(input.agent) : undefined,
tool.id === CodeModeTool.id ? yield* describeCodeMode(input.agent, input.permission) : undefined,
]
description: [output.description, tool.id === TaskTool.id ? yield* describeTask(input.agent) : undefined]
.filter(Boolean)
.join("\n"),
parameters: output.parameters,
@@ -341,7 +314,6 @@ const layer = Layer.effect(
}),
)
function isZodType(value: unknown): value is z.ZodType {
return typeof value === "object" && value !== null && "_zod" in value
}
@@ -434,7 +406,6 @@ export const node = LayerNode.make({
LSP.node,
Instruction.node,
FSUtil.node,
MCP.node,
EventV2Bridge.node,
httpClient,
CrossSpawnSpawner.node,
@@ -117,7 +117,6 @@ function makeMcp(instructions: MCP.ServerInstructions[] = []) {
clients: () => Effect.succeed({}),
instructions: () => Effect.succeed(instructions),
tools: () => Effect.succeed({}),
defs: () => Effect.succeed({}),
prompts: () => Effect.succeed({}),
resources: () => Effect.succeed({}),
resourceTemplates: () => Effect.succeed({}),
@@ -39,7 +39,6 @@ const mcp = Layer.succeed(
clients: () => Effect.succeed({}),
instructions: () => Effect.succeed([]),
tools: () => Effect.succeed({}),
defs: () => Effect.succeed({}),
prompts: () => Effect.succeed({}),
resources: () => Effect.succeed({}),
resourceTemplates: () => Effect.succeed({}),
@@ -1,164 +0,0 @@
import { afterEach, describe, expect } from "bun:test"
import path from "path"
import { Effect, Layer } from "effect"
import { LayerNode } from "@opencode-ai/core/effect/layer-node"
import { SessionTools } from "@/session/tools"
import { ToolRegistry } from "@/tool/registry"
import { Agent } from "@/agent/agent"
import { Config } from "@/config/config"
import { MCP } from "@/mcp"
import { McpCatalog } from "@/mcp/catalog"
import { Permission } from "@/permission"
import { Plugin } from "@/plugin"
import * as Truncate from "@/tool/truncate"
import { RuntimeFlags } from "@/effect/runtime-flags"
import { InstanceState } from "@/effect/instance-state"
import { MessageID, SessionID } from "@/session/schema"
import { disposeAllInstances } from "../fixture/fixture"
import { testEffect } from "../lib/effect"
import { TestConfig } from "../fixture/config"
import type { Tool as AITool } from "ai"
const configLayer = TestConfig.layer({
directories: () => InstanceState.directory.pipe(Effect.map((dir) => [path.join(dir, ".opencode")])),
})
// Fake MCP.Service built through the same `convertTool` path the real service uses,
// so the raw registration loop dispatches through a genuine ai-sdk execute. The fake
// client answers every call with a small text result.
function fakeMcpLayer(tools: Record<string, { description: string }>) {
const client = { callTool: async () => ({ content: [{ type: "text", text: "ok" }] }) }
const converted: Record<string, AITool> = Object.fromEntries(
Object.entries(tools).map(([key, def]) => [
key,
McpCatalog.convertTool(
{
name: key,
description: def.description,
inputSchema: { type: "object", properties: {} },
} as any,
client as any,
),
]),
)
return Layer.mock(MCP.Service, {
tools: () => Effect.succeed(converted),
defs: () => Effect.succeed({}),
clients: () => Effect.succeed({ github: { getServerCapabilities: () => undefined } } as any),
})
}
const mcpToolsLayer = fakeMcpLayer({
github_create_issue: { description: "Create an issue" },
github_list_issues: { description: "List issues" },
})
const root = LayerNode.group([ToolRegistry.node, Agent.node, MCP.node, RuntimeFlags.node])
const withCodeMode = testEffect(
LayerNode.compile(root, [
[Config.node, configLayer],
[RuntimeFlags.node, RuntimeFlags.layer({ experimentalCodeMode: true })],
[MCP.node, mcpToolsLayer],
]),
)
const withFlagOff = testEffect(
LayerNode.compile(root, [
[Config.node, configLayer],
[RuntimeFlags.node, RuntimeFlags.layer()],
[MCP.node, mcpToolsLayer],
]),
)
// Resolve the session tool record with the smallest honest stand-ins for the pieces
// resolve reads: a fabricated model (only providerID/api.id are consulted by the
// schema transform for these fixtures), a pass-through Truncate, an always-allow
// Permission, and an optionally-observing Plugin.trigger. Registry, MCP, and flags
// come from the compiled instance context.
function resolveTools(trigger?: Plugin.Interface["trigger"]) {
return Effect.gen(function* () {
const agents = yield* Agent.Service
const agent = yield* agents.defaultInfo()
return yield* SessionTools.resolve({
agent,
model: { providerID: "opencode", api: { id: "test" } } as any,
session: { id: SessionID.make("ses_session-tools"), permission: [] } as any,
processor: {
message: { id: MessageID.make("msg_session-tools") } as any,
updateToolCall: () => Effect.succeed(undefined),
completeToolCall: () => Effect.void,
},
bypassAgentCheck: false,
messages: [],
promptOps: {} as any,
})
}).pipe(
Effect.provide(
Layer.mergeAll(
Layer.mock(Permission.Service, { ask: () => Effect.void }),
Layer.mock(Plugin.Service, {
trigger:
trigger ?? (((_name, _input, output) => Effect.succeed(output)) as Plugin.Interface["trigger"]),
}),
Layer.mock(Truncate.Service, {
output: (text: string) => Effect.succeed({ content: text, truncated: false as const }),
}),
),
),
)
}
afterEach(async () => {
await disposeAllInstances()
})
describe("session.tools resolve", () => {
withCodeMode.instance("code mode suppresses raw per-MCP registration behind the execute tool", () =>
Effect.gen(function* () {
const tools = yield* resolveTools()
const ids = Object.keys(tools)
expect(ids).toContain("execute")
expect(ids).not.toContain("github_create_issue")
expect(ids).not.toContain("github_list_issues")
}),
)
withFlagOff.instance("without code mode the raw MCP tools register and execute is absent", () =>
Effect.gen(function* () {
const tools = yield* resolveTools()
const ids = Object.keys(tools)
expect(ids).not.toContain("execute")
expect(ids).toContain("github_create_issue")
expect(ids).toContain("github_list_issues")
}),
)
withFlagOff.instance("legacy raw MCP execution fires plugin hooks keyed by the ai-sdk toolCallId", () =>
Effect.gen(function* () {
const events: { name: string; input: any; output: any }[] = []
const trigger = ((name: unknown, input: unknown, output: unknown) =>
Effect.sync(() => {
events.push({ name: name as string, input, output })
return output
})) as Plugin.Interface["trigger"]
const tools = yield* resolveTools(trigger)
const raw = tools["github_create_issue"]!
const result = yield* Effect.promise(() =>
Promise.resolve(
raw.execute!(
{ title: "x" },
{ toolCallId: "call_legacy", abortSignal: new AbortController().signal, messages: [] },
),
),
)
expect((result as any).output).toBe("ok")
expect(events.map((e) => [e.name, e.input.tool, e.input.callID])).toEqual([
["tool.execute.before", "github_create_issue", "call_legacy"],
["tool.execute.after", "github_create_issue", "call_legacy"],
])
expect(events[0]!.output).toEqual({ args: { title: "x" } })
// The after hook receives the raw MCP result, before model-facing shaping.
expect(events[1]!.output).toMatchObject({ content: [{ type: "text", text: "ok" }] })
}),
)
})
@@ -1,5 +1,5 @@
import { beforeAll, describe, expect, test } from "bun:test"
import { CodeModeTool, catalogInstructions } from "@/tool/code-mode"
import { CodeModeTool, describeCatalog } from "@/tool/code-mode"
import { McpCatalog } from "@/mcp/catalog"
import { Agent } from "@/agent/agent"
import { MCP } from "@/mcp"
@@ -17,12 +17,9 @@ import {
ListToolsRequestSchema,
type Tool as MCPToolDef,
} from "@modelcontextprotocol/sdk/types.js"
import type { Tool as AITool } from "ai"
import { Effect, Layer } from "effect"
// A 1x1 transparent PNG, base64-encoded, used to exercise image attachments.
const PNG =
"iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR42mNkYPhfDwAChwGA60e6kgAAAABJRU5ErkJggg=="
const PNG = "iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR42mNkYPhfDwAChwGA60e6kgAAAABJRU5ErkJggg=="
const SERVER = "fixtures"
@@ -37,16 +34,7 @@ const ctx: Tool.Context = {
ask: () => Effect.void,
}
/**
* A minimal JSON-RPC MCP client speaking the real protocol over the in-memory transport.
* Deliberately NOT the SDK `Client`: `test/mcp/lifecycle.test.ts` (and friends) replace
* `@modelcontextprotocol/sdk/client/index.js` via bun's `mock.module`, which is
* process-global and irreversible in a full `bun test` run (CI) any later import of the
* SDK Client gets the mock, whose `listTools` returns a canned `test_tool`. Speaking raw
* JSON-RPC keeps this suite's "real server, real transport, real protocol" property while
* being immune to that contamination. Only the surface `convertTool`/this file use is
* implemented: connect handshake, tools/list, tools/call.
*/
// Avoid the SDK Client here; other MCP tests mock it process-globally.
class RawJsonRpcClient {
private nextId = 1
private pending = new Map<number, { resolve: (value: any) => void; reject: (error: Error) => void }>()
@@ -56,7 +44,7 @@ class RawJsonRpcClient {
async connect() {
this.transport.onmessage = (message) => {
const msg = message as { id?: number; result?: unknown; error?: { message: string } }
if (msg.id === undefined) return // notifications/requests from the server are not needed here
if (msg.id === undefined) return
const entry = this.pending.get(msg.id)
if (!entry) return
this.pending.delete(msg.id)
@@ -83,16 +71,11 @@ class RawJsonRpcClient {
return this.request("tools/list", {})
}
/** The `convertTool` surface: schema/options (timeouts, progress) are SDK-client
* concerns and are ignored here the server never sees them. */
callTool(params: { name: string; arguments?: Record<string, unknown> }, _schema?: unknown, _options?: unknown) {
return this.request("tools/call", params)
}
}
// A real MCP server, exposed over an in-memory transport, with a representative mix
// of tools: plain text, structured data (with an outputSchema), an image, and a
// failing tool. Tools are defined with raw JSON Schema so outputSchema is exact.
const TOOL_DEFS: MCPToolDef[] = [
{
name: "get_text",
@@ -150,21 +133,15 @@ async function buildTool() {
await client.connect()
const listed = (await client.listTools()).tools as MCPToolDef[]
const mcpTools: Record<string, AITool> = {}
const mcpDefs: Record<string, MCPToolDef> = {}
const mcpTools: Record<string, MCP.McpTool> = {}
for (const def of listed) {
const key = McpCatalog.toolName(SERVER, def.name)
mcpDefs[key] = def
mcpTools[key] = McpCatalog.convertTool(def, client as unknown as Client)
mcpTools[McpCatalog.toolName(SERVER, def.name)] = { def, client: client as unknown as Client }
}
// Truncate echoes its input so assertions read the exact program output; Agent/Session
// supply the (empty) permission rulesets the execute path merges; MCP serves the tools
// this real in-memory server listed — the same snapshot shape the live service returns.
const layer = Layer.mergeAll(
Layer.mock(Plugin.Service, {
trigger: (((_name: unknown, _input: unknown, output: unknown) =>
Effect.succeed(output)) as Plugin.Interface["trigger"]),
trigger: ((_name: unknown, _input: unknown, output: unknown) =>
Effect.succeed(output)) as Plugin.Interface["trigger"],
}),
Layer.mock(Truncate.Service, {
output: (text: string) => Effect.succeed({ content: text, truncated: false as const }),
@@ -173,14 +150,12 @@ async function buildTool() {
Layer.mock(Session.Service, { get: () => Effect.succeed({ permission: [] } as any) }),
Layer.mock(MCP.Service, {
tools: () => Effect.succeed(mcpTools),
defs: () => Effect.succeed(mcpDefs),
clients: () => Effect.succeed({ [SERVER]: {} as any }),
}),
)
return {
tool: await Effect.runPromise(CodeModeTool.pipe(Effect.flatMap(Tool.init), Effect.provide(layer))),
// The catalog section the registry appends to the base description (describeCodeMode).
description: catalogInstructions(mcpTools, mcpDefs, [SERVER]),
description: describeCatalog(mcpTools, [SERVER]),
}
}
@@ -196,14 +171,10 @@ describe("code mode integration (real MCP server)", () => {
test("the appended catalog inlines full signatures with real MCP schemas", () => {
expect(description).toContain("Available tools (COMPLETE list")
expect(description).toContain("- fixtures (4 tools)")
expect(description).toContain(
"tools.fixtures.add(input: { a: number; b: number }): Promise<{ sum: number }>",
)
expect(description).toContain("tools.fixtures.add(input: { a: number; b: number }): Promise<{ sum: number }>")
expect(description).toContain("tools.fixtures.get_text(input: { name: string }): Promise<unknown>")
expect(description).toContain("// Add two numbers and return the structured sum")
// Small catalog: everything is inline, so no discovery tool is advertised.
expect(description).not.toContain("$codemode")
// The workflow section is present with placeholder-only call forms.
expect(description).toContain("## Workflow")
expect(description).toContain("`const res = await tools.<namespace>.<tool>(input)`")
expect(description).not.toContain("total_count")
@@ -252,7 +223,6 @@ describe("code mode integration (real MCP server)", () => {
value: "[1 image attached to the result]",
})
expect(out.output).not.toContain(PNG)
// The stripped image still arrives as a real attachment.
expect(out.attachments).toHaveLength(1)
})
+125 -271
View File
@@ -1,15 +1,5 @@
import { describe, expect, test } from "bun:test"
import {
CODE_MODE_TOOL,
CodeModeTool,
Parameters,
catalogInstructions,
formatValue,
groupByServer,
toSandboxResult,
withLogs,
type Attachment,
} from "@/tool/code-mode"
import { CODE_MODE_TOOL, CodeModeTool, Parameters, describeCatalog } from "@/tool/code-mode"
import type { Tool as MCPToolDef } from "@modelcontextprotocol/sdk/types.js"
import type { PermissionV1 } from "@opencode-ai/core/v1/permission"
import { Agent } from "@/agent/agent"
@@ -19,9 +9,7 @@ import { Plugin } from "@/plugin"
import { Session } from "@/session/session"
import { Tool } from "@/tool/tool"
import * as Truncate from "@/tool/truncate"
import { McpCatalog } from "@/mcp/catalog"
import { MessageID, SessionID } from "@/session/schema"
import type { Tool as AITool } from "ai"
import { Effect, Layer, Schema } from "effect"
const ctx: Tool.Context = {
@@ -35,28 +23,22 @@ const ctx: Tool.Context = {
ask: () => Effect.void,
}
// Build a real MCP-derived AI SDK tool over a fake transport, so the adapter exercises
// the same `convertTool` execution path that `mcp.tools()` produces at runtime.
function mcpTool(
name: string,
handler: (args: Record<string, unknown>) => unknown,
inputSchema: Record<string, unknown> = { type: "object", properties: {} },
): AITool {
const client = {
callTool: async (params: { arguments?: Record<string, unknown> }) => handler(params.arguments ?? {}),
outputSchema?: Record<string, unknown>,
): MCP.McpTool {
return {
def: { name, description: name, inputSchema, ...(outputSchema ? { outputSchema } : {}) } as MCPToolDef,
client: {
callTool: async (params: { arguments?: Record<string, unknown> }) => handler(params.arguments ?? {}),
} as unknown as MCP.McpTool["client"],
}
return McpCatalog.convertTool({ name, description: name, inputSchema } as any, client as any)
}
// Truncate echoes its input so assertions read the exact program output. Agent.get is
// consulted by the shared wrapper during truncation AND at execute time for the
// permission ruleset that filters the dispatchable tool tree; Session.get supplies the
// (empty) session-level ruleset half of that merge. Plugin.trigger defaults to the
// pass-through the real service uses when no plugin implements a hook; tests observing
// or failing hooks override it.
function harness(input: {
mcpTools: Record<string, AITool>
defs?: Record<string, MCPToolDef>
mcpTools: Record<string, MCP.McpTool>
servers: string[]
permission?: PermissionV1.Rule[]
trigger?: Plugin.Interface["trigger"]
@@ -76,21 +58,17 @@ function harness(input: {
}),
Layer.mock(MCP.Service, {
tools: () => Effect.succeed(input.mcpTools),
defs: () => Effect.succeed(input.defs ?? {}),
clients: () => Effect.succeed(Object.fromEntries(input.servers.map((name) => [name, {} as any]))),
}),
)
}
// Derive sanitized server namespaces from the catalog keys, mirroring how the registry
// passes `Object.keys(mcp.clients()).map(sanitize)`.
function serverNames(mcpTools: Record<string, AITool>, servers?: string[]) {
function serverNames(mcpTools: Record<string, MCP.McpTool>, servers?: string[]) {
return servers ?? [...new Set(Object.keys(mcpTools).map((key) => key.split("_")[0]!))]
}
function build(
mcpTools: Record<string, AITool>,
defs: Record<string, MCPToolDef> = {},
mcpTools: Record<string, MCP.McpTool>,
servers?: string[],
permission?: PermissionV1.Rule[],
trigger?: Plugin.Interface["trigger"],
@@ -99,20 +77,13 @@ function build(
return Effect.runPromise(
CodeModeTool.pipe(
Effect.flatMap(Tool.init),
Effect.provide(harness({ mcpTools, defs, servers: names, permission, trigger })),
Effect.provide(harness({ mcpTools, servers: names, permission, trigger })),
),
)
}
// The agent-facing description, as the registry composes it (`describeCodeMode`):
// permission-filtered tool set → grouped catalog → CodeMode instructions.
function describeFor(
mcpTools: Record<string, AITool>,
defs: Record<string, MCPToolDef> = {},
servers?: string[],
permission: PermissionV1.Rule[] = [],
) {
return catalogInstructions(Permission.visibleTools(mcpTools, permission), defs, serverNames(mcpTools, servers))
function describeFor(mcpTools: Record<string, MCP.McpTool>, servers?: string[], permission: PermissionV1.Rule[] = []) {
return describeCatalog(Permission.visibleTools(mcpTools, permission), serverNames(mcpTools, servers))
}
describe("code mode execute", () => {
@@ -123,38 +94,30 @@ describe("code mode execute", () => {
})
test("groups multi-underscore server names by longest matching prefix", () => {
const groups = groupByServer({ my_server_do_thing: mcpTool("do_thing", () => "") }, ["my_server"])
expect([...groups.keys()]).toEqual(["my_server"])
expect(groups.get("my_server")![0]).toMatchObject({
path: "my_server.do_thing",
local: "do_thing",
key: "my_server_do_thing",
})
const description = describeFor({ my_server_do_thing: mcpTool("do_thing", () => "") }, ["my_server"])
expect(description).toContain("- my_server (1 tool)")
expect(description).toContain("tools.my_server.do_thing(")
})
test("groupByServer uses the whole key as the server name when it has no underscore", () => {
const groups = groupByServer({ standalone: mcpTool("standalone", () => "") }, [])
expect([...groups.keys()]).toEqual(["standalone"])
expect(groups.get("standalone")![0]).toMatchObject({
path: "standalone.standalone",
server: "standalone",
local: "standalone",
key: "standalone",
})
const description = describeFor({ standalone: mcpTool("standalone", () => "") }, [])
expect(description).toContain("- standalone (1 tool)")
expect(description).toContain("tools.standalone.standalone(")
})
test("groupByServer carries the raw MCP schemas for rendering", () => {
const defs: Record<string, MCPToolDef> = {
weather_current: {
name: "current",
inputSchema: { type: "object", properties: { city: { type: "string" } }, required: ["city"] },
outputSchema: { type: "object", properties: { tempC: { type: "number" } }, required: ["tempC"] },
} as any,
}
const groups = groupByServer({ weather_current: mcpTool("current", () => "") }, ["weather"], defs)
const entry = groups.get("weather")![0]!
expect(entry.inputSchema).toEqual(defs.weather_current!.inputSchema as any)
expect(entry.outputSchema).toEqual(defs.weather_current!.outputSchema as any)
test("describeCatalog carries the raw MCP schemas for rendering", () => {
const description = describeFor(
{
weather_current: mcpTool(
"current",
() => "",
{ type: "object", properties: { city: { type: "string" } }, required: ["city"] },
{ type: "object", properties: { tempC: { type: "number" } }, required: ["tempC"] },
),
},
["weather"],
)
expect(description).toContain("tools.weather.current(input: { city: string }): Promise<{ tempC: number }>")
})
test("the static base description carries no catalog; the registry appends it", async () => {
@@ -184,90 +147,76 @@ describe("code mode execute", () => {
)
expect(description).toContain("tools.github.list_issues(")
expect(description).toContain("tools.linear.search(")
// A schema with no properties renders as an empty object, not `{ }`.
expect(description).toContain("tools.linear.search(input: {}): Promise<unknown>")
// Fully inlined catalog: no discovery round-trip is needed or advertised.
expect(description).not.toContain("$codemode")
expect(description).not.toContain("Browse one namespace")
// The workflow/rules sections use placeholder call forms only — the example machinery
// never cherry-picks a catalog tool or fabricates result fields.
expect(description).toContain("## Workflow")
expect(description).toContain("1. Pick a tool from the list under `## Available tools`")
expect(description).toContain('`const data = typeof res === "string" ? JSON.parse(res) : res` - most tools return JSON as a string')
expect(description).toContain(
'`const data = typeof res === "string" ? JSON.parse(res) : res` - most tools return JSON as a string',
)
expect(description).toContain("Return only the fields you need")
expect(description).not.toContain("total_count")
})
test("signatures render the declared outputSchema as the return type", () => {
const defs: Record<string, MCPToolDef> = {
weather_current: {
name: "current",
inputSchema: { type: "object", properties: { city: { type: "string" } }, required: ["city"] },
outputSchema: {
const description = describeFor({
weather_current: mcpTool(
"current",
() => "",
{ type: "object", properties: { city: { type: "string" } }, required: ["city"] },
{
type: "object",
properties: { tempC: { type: "number" }, summary: { type: "string" } },
required: ["tempC"],
},
} as any,
}
const description = describeFor({ weather_current: mcpTool("current", () => "") }, defs)
),
})
expect(description).toContain(
"tools.weather.current(input: { city: string }): Promise<{ tempC: number; summary?: string }>",
)
})
test("large catalogs inline a budgeted PARTIAL list plus runtime search", async () => {
const tools: Record<string, AITool> = {}
const tools: Record<string, MCP.McpTool> = {}
const filler = "a searchable description of this operation that consumes catalog budget ".repeat(3)
for (let i = 0; i < 150; i++) {
const client = { callTool: async () => ({ content: [] }) }
tools[`alpha_op_${i}`] = McpCatalog.convertTool(
{
tools[`alpha_op_${i}`] = {
def: {
name: `op_${i}`,
description: `${filler}${i}`,
inputSchema: { type: "object", properties: { value: { type: "string" }, count: { type: "number" } } },
} as any,
client as any,
)
} as MCPToolDef,
client: { callTool: async () => ({ content: [] }) } as unknown as MCP.McpTool["client"],
}
}
tools["zeta_only_tool"] = mcpTool("only_tool", () => "", {
type: "object",
properties: { topic: { type: "string", description: "Subject to look up" } },
required: ["topic"],
})
const description = describeFor(tools, {}, ["alpha", "zeta"])
const description = describeFor(tools, ["alpha", "zeta"])
// Every namespace is listed with counts; signatures inline round-robin across
// namespaces (cheapest-first within each) until the budget runs out, and the
// description states exactly how comprehensive the list is. Round-robin fairness:
// the small zeta namespace is fully shown even though alpha alone could exhaust
// the whole budget.
expect(description).toContain("Available tools (PARTIAL - ")
expect(description).toMatch(/- alpha \(150 tools, \d+ shown\)/)
expect(description).toContain("- zeta (1 tool)\n")
expect(description).toContain("tools.zeta.only_tool(input: { topic: string }): Promise<unknown>")
expect(description).toContain("tools.$codemode.search(")
// PARTIAL catalogs put search first in the workflow and advertise namespace browsing.
expect(description).toContain("1. Find a tool (skip when it is already listed below)")
expect(description).toContain('- Browse one namespace: `await tools.$codemode.search({ query: "", namespace: "<name>" })`.')
expect(description).toContain(
'- Browse one namespace: `await tools.$codemode.search({ query: "", namespace: "<name>" })`.',
)
expect(description).not.toContain("total_count")
// All op lines cost the same estimated tokens (chars/4 rounds away the 1- vs 3-digit
// name difference), so the path tiebreak decides: the lexicographically-first ops made
// the cut and the lexicographic tail (op_99 is maximal) did not.
expect(description).toContain("tools.alpha.op_0(")
expect(description).not.toContain("tools.alpha.op_99(")
// The runtime search tool works in-program and returns complete signatures.
const tool = await build(tools, {}, ["alpha", "zeta"])
const tool = await build(tools, ["alpha", "zeta"])
const out = await Effect.runPromise(
tool.execute({ code: "return await tools.$codemode.search({ query: 'only tool', limit: 3 })" }, ctx),
)
const result = JSON.parse(out.output)
// Search-result paths carry the `tools.` prefix so each is directly usable as a call site.
expect(result.items.map((i: any) => i.path)).toContain("tools.zeta.only_tool")
expect(result.items[0].signature).toContain("tools.")
// Search results render the pretty multiline signature: MCP input-property
// descriptions ride along as JSDoc field comments. The inline catalog stays compact.
const signature = result.items.find((i: any) => i.path === "tools.zeta.only_tool").signature
expect(signature).toContain("tools.zeta.only_tool(input: {\n")
expect(signature).toContain(" /** Subject to look up */\n topic: string")
@@ -290,7 +239,10 @@ describe("code mode execute", () => {
linear_search: mcpTool("search", () => ""),
})
const output = await Effect.runPromise(
tool.execute({ code: "const namespaces = Object.keys(tools); return { namespaces, count: namespaces.length }" }, ctx),
tool.execute(
{ code: "const namespaces = Object.keys(tools); return { namespaces, count: namespaces.length }" },
ctx,
),
)
expect(JSON.parse(output.output)).toEqual({ namespaces: ["github", "linear"], count: 2 })
})
@@ -414,7 +366,6 @@ describe("code mode execute", () => {
expect(output.output).toBe("denied: permission denied by user")
expect(output.metadata.error).toBeUndefined()
// The MCP tool itself never ran.
expect(called).toEqual([])
expect(output.metadata.toolCalls).toEqual([{ tool: "a.tool", status: "error" }])
})
@@ -431,7 +382,6 @@ describe("code mode execute", () => {
a_tool: mcpTool("a", () => ({ content: [{ type: "text", text: "one" }] })),
b_tool: mcpTool("b", () => ({ content: [{ type: "text", text: "two" }] })),
},
{},
undefined,
undefined,
trigger,
@@ -442,7 +392,6 @@ describe("code mode execute", () => {
)
expect(out.output).toBe("done")
// callID is synthetic and per-execution: `${parentCallID}/${n}`, n starting at 1.
expect(events.map((e) => [e.name, e.input.tool, e.input.callID])).toEqual([
["tool.execute.before", "a_tool", "call_code_mode/1"],
["tool.execute.after", "a_tool", "call_code_mode/1"],
@@ -453,7 +402,6 @@ describe("code mode execute", () => {
expect(before!.input.sessionID).toBe(ctx.sessionID)
expect(before!.output).toEqual({ args: { x: 1 } })
expect(after!.input.args).toEqual({ x: 1 })
// The after hook sees the raw MCP result — the same payload the legacy path passes.
expect(after!.output).toEqual({ content: [{ type: "text", text: "one" }] })
})
@@ -469,7 +417,6 @@ describe("code mode execute", () => {
}
const tool = await build(
{ a_tool: mcpTool("a", record("a")), b_tool: mcpTool("b", record("b")) },
{},
undefined,
undefined,
trigger,
@@ -489,10 +436,8 @@ describe("code mode execute", () => {
),
)
// The program handled the hook failure; the rest ran and the outer result is ok.
expect(out.metadata.error).toBeUndefined()
expect(out.output).toBe("hook exploded / ok")
// The before hook gates dispatch: the failed child's tool never executed.
expect(called).toEqual(["b"])
})
@@ -508,7 +453,6 @@ describe("code mode execute", () => {
tool.execute({ code: "await tools.greeter.hello({ name: 'Ada' }); return 'done'" }, recordingCtx),
)
// The UI sees the call appear as running, then resolve to completed.
expect(snapshots).toContainEqual({
toolCalls: [{ tool: "greeter.hello", status: "running", input: { name: "Ada" } }],
})
@@ -546,7 +490,6 @@ describe("code mode execute", () => {
})
const out = await Effect.runPromise(tool.execute({ code: "return await tools.shot.take({})" }, ctx))
// The program received the structured content; the media rode along host-side.
expect(JSON.parse(out.output)).toEqual({ name: "shot.png" })
expect(out.attachments).toEqual([{ type: "file", mime: "image/png", url: "data:image/png;base64,PNGDATA" }])
expect(out.output).not.toContain("PNGDATA")
@@ -561,13 +504,72 @@ describe("code mode execute", () => {
expect(out.attachments).toEqual([{ type: "file", mime: "image/png", url: "data:image/png;base64,PNGDATA" }])
})
test("media-only markers distinguish all-image from mixed attachments", async () => {
const tool = await build({
media_images: mcpTool("images", () => ({
content: [
{ type: "image", data: "PNG1", mimeType: "image/png" },
{ type: "image", data: "PNG2", mimeType: "image/png" },
],
})),
media_mixed: mcpTool("mixed", () => ({
content: [
{ type: "image", data: "PNG3", mimeType: "image/png" },
{ type: "resource", resource: { uri: "file:///tmp/report.pdf", mimeType: "application/pdf", blob: "PDF1" } },
],
})),
})
const out = await Effect.runPromise(
tool.execute(
{
code: `
const images = await tools.media.images({})
const mixed = await tools.media.mixed({})
return { images, mixed }
`,
},
ctx,
),
)
expect(JSON.parse(out.output)).toEqual({
images: "[2 images attached to the result]",
mixed: "[2 files attached to the result]",
})
expect(out.output).not.toContain("PNG")
expect(out.attachments).toEqual([
{ type: "file", mime: "image/png", url: "data:image/png;base64,PNG1" },
{ type: "file", mime: "image/png", url: "data:image/png;base64,PNG2" },
{ type: "file", mime: "image/png", url: "data:image/png;base64,PNG3" },
{ type: "file", mime: "application/pdf", url: "data:application/pdf;base64,PDF1", filename: "report.pdf" },
])
})
test("resource links flow to the program as text, never as attachments", async () => {
const tool = await build({
docs_find: mcpTool("find", () => ({
content: [
{
type: "resource_link",
uri: "https://example.com/guide.pdf",
name: "guide.pdf",
mimeType: "application/pdf",
},
{ type: "resource_link", uri: "file:///tmp/notes.md", name: "notes.md" },
],
})),
})
const out = await Effect.runPromise(tool.execute({ code: "return await tools.docs.find({})" }, ctx))
expect(out.output).toBe("guide.pdf: https://example.com/guide.pdf\nnotes.md: file:///tmp/notes.md")
expect(out.attachments).toBeUndefined()
})
test("attachments still flow when the program returns something else entirely", async () => {
const tool = await build({
shot_take: mcpTool("take", () => ({ content: [{ type: "image", data: "PNGDATA", mimeType: "image/png" }] })),
})
const out = await Effect.runPromise(
tool.execute({ code: "await tools.shot.take({}); return 'captured'" }, ctx),
)
const out = await Effect.runPromise(tool.execute({ code: "await tools.shot.take({}); return 'captured'" }, ctx))
expect(out.output).toBe("captured")
expect(out.attachments).toHaveLength(1)
})
@@ -579,9 +581,6 @@ describe("code mode execute", () => {
})
test("cancelling via ctx.abort interrupts the running program", async () => {
// The child call triggers the abort itself, deterministically, while the program
// swallows the call's failure and heads into an infinite loop. If abort did not
// interrupt the execution fiber, this test would hang on the busy loop.
const controller = new AbortController()
const tool = await build({
host_trigger: mcpTool("trigger", () => {
@@ -613,8 +612,6 @@ describe("code mode execute", () => {
})
test("leaves oversized results to OpenCode's native tool-output truncation", async () => {
// No CodeMode output limit is set, so the full result reaches the shared Tool.define
// wrapper intact (the harness Truncate fake passes it through un-truncated).
const tool = await build({})
const output = await Effect.runPromise(tool.execute({ code: "return 'x'.repeat(40000)" }, ctx))
expect(output.metadata.error).toBeUndefined()
@@ -649,7 +646,7 @@ describe("code mode permission visibility", () => {
github_create_issue: mcpTool("create_issue", ok),
github_list_issues: mcpTool("list_issues", ok),
}
const description = describeFor(mcpTools, {}, ["github"], [deny("github_create_issue")])
const description = describeFor(mcpTools, ["github"], [deny("github_create_issue")])
expect(description).toContain("tools.github.list_issues(")
expect(description).not.toContain("create_issue")
expect(description).toContain("- github (1 tool)")
@@ -660,7 +657,7 @@ describe("code mode permission visibility", () => {
github_create_issue: mcpTool("create_issue", ok),
github_list_issues: mcpTool("list_issues", ok),
}
const description = describeFor(mcpTools, {}, ["github"], [askRule("github_create_issue")])
const description = describeFor(mcpTools, ["github"], [askRule("github_create_issue")])
expect(description).toContain("tools.github.create_issue(")
expect(description).toContain("tools.github.list_issues(")
expect(description).toContain("- github (2 tools)")
@@ -676,7 +673,6 @@ describe("code mode permission visibility", () => {
}),
github_list_issues: mcpTool("list_issues", ok),
},
{},
["github"],
[deny("github_create_issue")],
)
@@ -689,10 +685,7 @@ describe("code mode permission visibility", () => {
expect(denied.output).not.toContain("permission")
expect(called).toEqual([])
// The rest of the namespace still works.
const allowed = await Effect.runPromise(
tool.execute({ code: "return await tools.github.list_issues({})" }, ctx),
)
const allowed = await Effect.runPromise(tool.execute({ code: "return await tools.github.list_issues({})" }, ctx))
expect(allowed.metadata.error).toBeUndefined()
expect(allowed.output).toBe("ok")
})
@@ -702,13 +695,10 @@ describe("code mode permission visibility", () => {
const askCtx: Tool.Context = { ...ctx, ask: (req) => Effect.sync(() => void asked.push(req.permission)) }
const tool = await build(
{ github_list_issues: mcpTool("list_issues", ok) },
{},
["github"],
[askRule("github_list_issues")],
)
const out = await Effect.runPromise(
tool.execute({ code: "return await tools.github.list_issues({})" }, askCtx),
)
const out = await Effect.runPromise(tool.execute({ code: "return await tools.github.list_issues({})" }, askCtx))
expect(out.output).toBe("ok")
expect(asked).toEqual(["github_list_issues"])
})
@@ -718,144 +708,8 @@ describe("code mode permission visibility", () => {
const visible = Permission.visibleTools(tools, [
deny("a_tool"),
askRule("b_tool"),
// A scoped (non-"*") deny does not hide the tool from the catalog.
{ permission: "c_tool", pattern: "something", action: "deny" },
])
expect(Object.keys(visible)).toEqual(["b_tool", "c_tool"])
})
})
describe("toSandboxResult", () => {
const collector = () => {
const attachments: Attachment[] = []
return { attachments, collect: (a: Attachment) => void attachments.push(a) }
}
test("prefers structuredContent over text", () => {
const { collect } = collector()
expect(toSandboxResult({ structuredContent: { x: 1 }, content: [{ type: "text", text: "hi" }] }, collect)).toEqual(
{ x: 1 },
)
})
test("joins text content when no structured content is present", () => {
const { collect } = collector()
expect(
toSandboxResult(
{ content: [{ type: "text", text: "one" }, { type: "text", text: "two" }] },
collect,
),
).toBe("one\ntwo")
})
test("passes non-MCP values through untouched", () => {
const { collect } = collector()
expect(toSandboxResult("raw", collect)).toBe("raw")
expect(toSandboxResult(42, collect)).toBe(42)
expect(toSandboxResult(null, collect)).toBeNull()
expect(toSandboxResult({ some: "object" }, collect)).toEqual({ some: "object" })
})
test("strips media into the accumulator; text stays the sandbox value", () => {
const { attachments, collect } = collector()
const value = toSandboxResult(
{
content: [
{ type: "text", text: "see image" },
{ type: "image", data: "AAAA", mimeType: "image/png" },
],
},
collect,
)
expect(value).toBe("see image")
expect(attachments).toEqual([{ type: "file", mime: "image/png", url: "data:image/png;base64,AAAA" }])
})
test("a media-only result yields a marker; counts and nouns follow the content", () => {
const one = collector()
expect(toSandboxResult({ content: [{ type: "image", data: "A", mimeType: "image/png" }] }, one.collect)).toBe(
"[1 image attached to the result]",
)
const two = collector()
expect(
toSandboxResult(
{
content: [
{ type: "image", data: "A", mimeType: "image/png" },
{ type: "image", data: "B", mimeType: "image/jpeg" },
],
},
two.collect,
),
).toBe("[2 images attached to the result]")
expect(two.attachments).toHaveLength(2)
const mixed = collector()
expect(
toSandboxResult(
{
content: [
{ type: "image", data: "A", mimeType: "image/png" },
{ type: "audio", data: "B", mimeType: "audio/wav" },
],
},
mixed.collect,
),
).toBe("[2 files attached to the result]")
})
test("extracts embedded resources: text inline, blobs as attachments with filenames", () => {
const { attachments, collect } = collector()
const value = toSandboxResult(
{
content: [
{ type: "resource", resource: { uri: "file:///tmp/notes.txt", mimeType: "text/plain", text: "note text" } },
{ type: "resource", resource: { uri: "file:///tmp/doc.pdf", mimeType: "application/pdf", blob: "PDF" } },
],
},
collect,
)
expect(value).toBe("note text")
expect(attachments).toEqual([
{ type: "file", mime: "application/pdf", url: "data:application/pdf;base64,PDF", filename: "doc.pdf" },
])
})
test("collects resource_link blocks as external-URL attachments", () => {
const { attachments, collect } = collector()
const value = toSandboxResult(
{ content: [{ type: "resource_link", uri: "https://example.com/report.csv", mimeType: "text/csv" }] },
collect,
)
expect(value).toBe("[1 file attached to the result]")
expect(attachments).toEqual([
{ type: "file", mime: "text/csv", url: "https://example.com/report.csv", filename: "report.csv" },
])
})
test("an MCP-shaped result with nothing extractable becomes null", () => {
const { collect } = collector()
expect(toSandboxResult({ content: [] }, collect)).toBeNull()
expect(toSandboxResult({ content: [{ type: "mystery" }] }, collect)).toBeNull()
})
})
describe("formatting helpers", () => {
test("formatValue", () => {
expect(formatValue("text")).toBe("text")
expect(formatValue({ a: 1 })).toBe(JSON.stringify({ a: 1 }, null, 2))
expect(formatValue(null)).toBe("null")
expect(formatValue(undefined)).toBe("undefined")
})
test("withLogs", () => {
// No logs: output is returned untouched.
expect(withLogs("result", [])).toBe("result")
expect(withLogs("result")).toBe("result")
// Logs are appended as a trailing section, one line each.
expect(withLogs("result", ["a", "[warn] b"])).toBe("result\n\nLogs:\na\n[warn] b")
// Empty output still gets the section (no leading blank lines).
expect(withLogs("", ["[error] boom"])).toBe("Logs:\n[error] boom")
})
})
@@ -17,8 +17,6 @@ import { InstanceState } from "@/effect/instance-state"
import { ToolJsonSchema } from "@/tool/json-schema"
import { MessageID, SessionID } from "@/session/schema"
import { RuntimeFlags } from "@/effect/runtime-flags"
import { MCP } from "@/mcp"
import { McpCatalog } from "@/mcp/catalog"
import { ProviderV2 } from "@opencode-ai/core/provider"
import { ModelV2 } from "@opencode-ai/core/model"
@@ -59,45 +57,6 @@ const replacements = [
const it = testEffect(LayerNode.compile(root, replacements))
const withBrokenPlugin = testEffect(LayerNode.compile(root, [...replacements, [Plugin.node, brokenPluginLayer]]))
// Fake MCP.Service serving two tools on one server, built through the same
// `convertTool` path the real service uses so the code-mode catalog renders
// genuine signatures.
function fakeMcpLayer(tools: Record<string, { description: string }>) {
const client = { callTool: async () => ({ content: [] }) }
const converted = Object.fromEntries(
Object.entries(tools).map(([key, def]) => [
key,
McpCatalog.convertTool(
{
name: key,
description: def.description,
inputSchema: { type: "object", properties: {} },
} as any,
client as any,
),
]),
)
return Layer.mock(MCP.Service, {
tools: () => Effect.succeed(converted),
defs: () => Effect.succeed({}),
clients: () => Effect.succeed(Object.keys(tools).length > 0 ? ({ github: {} } as any) : {}),
})
}
const mcpToolsLayer = fakeMcpLayer({
github_create_issue: { description: "Create an issue" },
github_list_issues: { description: "List issues" },
})
const codeModeFlags = [RuntimeFlags.node, RuntimeFlags.layer({ experimentalCodeMode: true })] as const
const withCodeMode = testEffect(
LayerNode.compile(root, [[Config.node, configLayer], codeModeFlags, [MCP.node, mcpToolsLayer]]),
)
const withCodeModeNoMcp = testEffect(
LayerNode.compile(root, [[Config.node, configLayer], codeModeFlags, [MCP.node, fakeMcpLayer({})]]),
)
const withFlagOff = testEffect(LayerNode.compile(root, [...replacements, [MCP.node, mcpToolsLayer]]))
afterEach(async () => {
await disposeAllInstances()
})
@@ -532,82 +491,3 @@ describe("tool.registry", () => {
}),
)
})
describe("tool.registry code mode", () => {
const model = { providerID: ProviderV2.ID.opencode, modelID: ModelV2.ID.make("test") }
withCodeMode.instance("registers the code-mode execute tool when the flag is on and MCP tools exist", () =>
Effect.gen(function* () {
const registry = yield* ToolRegistry.Service
const agents = yield* Agent.Service
const tools = yield* registry.tools({ ...model, agent: yield* agents.defaultInfo() })
const execute = tools.find((tool) => tool.id === "execute")
if (!execute) throw new Error("code-mode execute tool was not registered")
// The registry appends the grouped catalog to the static base description.
expect(execute.description).toContain("confined runtime")
expect(execute.description).toContain("## Available tools")
expect(execute.description).toContain("tools.github.create_issue(")
expect(execute.description).toContain("tools.github.list_issues(")
}),
)
withCodeModeNoMcp.instance("does not register code mode when no MCP tools are connected", () =>
Effect.gen(function* () {
const registry = yield* ToolRegistry.Service
const agents = yield* Agent.Service
const tools = yield* registry.tools({ ...model, agent: yield* agents.defaultInfo() })
expect(tools.map((tool) => tool.id)).not.toContain("execute")
}),
)
withFlagOff.instance("does not register code mode when the flag is off, even with MCP tools", () =>
Effect.gen(function* () {
const registry = yield* ToolRegistry.Service
const agents = yield* Agent.Service
const tools = yield* registry.tools({ ...model, agent: yield* agents.defaultInfo() })
expect(tools.map((tool) => tool.id)).not.toContain("execute")
}),
)
withCodeMode.instance("hard-denied MCP tools are dropped from the catalog; ask-level ones stay", () =>
Effect.gen(function* () {
const registry = yield* ToolRegistry.Service
const agents = yield* Agent.Service
const base = yield* agents.defaultInfo()
const agent = {
...base,
permission: [
...base.permission,
{ permission: "github_create_issue", pattern: "*" as const, action: "deny" as const },
{ permission: "github_list_issues", pattern: "*" as const, action: "ask" as const },
],
}
const tools = yield* registry.tools({ ...model, agent })
const execute = tools.find((tool) => tool.id === "execute")
if (!execute) throw new Error("code-mode execute tool was not registered")
expect(execute.description).not.toContain("create_issue")
expect(execute.description).toContain("tools.github.list_issues(")
expect(execute.description).toContain("- github (1 tool)")
}),
)
withCodeMode.instance("session-level hard-denied MCP tools are dropped from the code-mode catalog", () =>
Effect.gen(function* () {
const registry = yield* ToolRegistry.Service
const agents = yield* Agent.Service
const tools = yield* registry.tools({
...model,
agent: yield* agents.defaultInfo(),
permission: [
{ permission: "github_create_issue", pattern: "*" as const, action: "deny" as const },
{ permission: "github_list_issues", pattern: "*" as const, action: "ask" as const },
],
})
const execute = tools.find((tool) => tool.id === "execute")
if (!execute) throw new Error("code-mode execute tool was not registered")
expect(execute.description).not.toContain("create_issue")
expect(execute.description).toContain("tools.github.list_issues(")
expect(execute.description).toContain("- github (1 tool)")
}),
)
})
@@ -230,6 +230,7 @@
line-height: var(--line-height-large);
letter-spacing: var(--letter-spacing-normal);
text-transform: capitalize;
color: var(--task-agent-legacy-color, var(--text-strong));
}
[data-slot="basic-tool-tool-subtitle"] {
@@ -247,6 +248,57 @@
}
}
body[data-new-layout] {
[data-component="task-tool-card"] {
padding: 8px 12px 8px 10px;
border-radius: 8px;
border: 0;
border-color: transparent;
background: var(--task-agent-background, light-dark(#fafafa, rgba(250, 250, 250, 0.15)));
box-shadow: inset 0 0 0 0.5px var(--task-agent-border, var(--v2-border-border-base));
[data-component="task-tool-spinner"],
[data-component="task-tool-title"] {
color: var(--task-agent-color, var(--text-strong));
}
[data-component="task-tool-title"] {
height: 20px;
font-family: var(--v2-font-family-sans, "Inter", sans-serif);
font-style: normal;
font-weight: 530;
font-size: 13px;
line-height: 20px;
letter-spacing: -0.04px;
flex: none;
flex-grow: 0;
}
[data-slot="basic-tool-tool-subtitle"] {
font-family: var(--v2-font-family-sans, "Inter", sans-serif);
font-style: normal;
font-weight: 440;
font-size: 13px;
line-height: 20px;
letter-spacing: -0.04px;
color: light-dark(var(--v2-text-text-base), #fafafa);
font-variation-settings: "slnt" 0;
flex: none;
flex-grow: 0;
}
[data-component="session-progress-indicator-v2"] {
color: var(--task-agent-color, light-dark(var(--v2-text-text-base), #ffffff));
}
&:hover,
&:focus-visible {
border-color: transparent;
background: var(--task-agent-background, light-dark(#fafafa, rgba(250, 250, 250, 0.15)));
}
}
}
body:not([data-new-layout]) {
[data-component="tool-trigger"] {
[data-slot="basic-tool-tool-spinner"] {
@@ -62,6 +62,7 @@ import { animate } from "motion"
import { useLocation } from "@solidjs/router"
import { attached, inline, kind } from "./message-file"
import { readPartText } from "./message-part-text"
import { SessionProgressIndicatorV2 } from "../v2/components/session-progress-indicator-v2"
async function writeClipboard(text: string): Promise<boolean> {
const body = typeof document === "undefined" ? undefined : document.body
@@ -370,6 +371,44 @@ const agentTones: Record<string, string> = {
plan: "var(--icon-agent-plan-base)",
}
const v2AgentTones: Record<string, { color: string; border: string; background: string }> = {
build: {
color: "var(--v2-agent-build-solid)",
border: "var(--v2-agent-build-border)",
background: "var(--v2-agent-build-background)",
},
explore: {
color: "var(--v2-agent-explore-solid)",
border: "var(--v2-agent-explore-border)",
background: "var(--v2-agent-explore-background)",
},
plan: {
color: "var(--v2-agent-plan-solid)",
border: "var(--v2-agent-plan-border)",
background: "var(--v2-agent-plan-background)",
},
}
const agentThemeColors: Record<string, string> = {
primary: "var(--text-interactive-base)",
secondary: "var(--text-base)",
accent: "var(--icon-info-base)",
success: "var(--icon-success-base)",
warning: "var(--icon-warning-base)",
error: "var(--icon-critical-base)",
info: "var(--icon-info-base)",
}
const v2AgentThemeColors: Record<string, string> = {
primary: "var(--v2-text-text-accent)",
secondary: "var(--v2-text-text-muted)",
accent: "var(--v2-icon-icon-accent)",
success: "var(--v2-state-fg-success)",
warning: "var(--v2-state-fg-warning)",
error: "var(--v2-state-fg-danger)",
info: "var(--v2-state-fg-info)",
}
const agentPalette = [
"var(--icon-agent-ask-base)",
"var(--icon-agent-build-base)",
@@ -394,16 +433,31 @@ function tone(name: string) {
function taskAgent(
raw: unknown,
list?: readonly { name: string; color?: string }[],
): { name?: string; color?: string } {
): { name?: string; color?: string; v2Color?: string; border?: string; background?: string } {
if (typeof raw !== "string" || !raw) return {}
const key = raw.toLowerCase()
const item = list?.find((entry) => entry.name === raw || entry.name.toLowerCase() === key)
const v2Tone = item?.color ? undefined : v2AgentTones[key]
const color = agentColor(item?.color, agentThemeColors) ?? agentTones[key] ?? tone(key)
const v2Color = agentColor(item?.color, v2AgentThemeColors) ?? v2Tone?.color ?? color
return {
name: item?.name ?? `${raw[0]!.toUpperCase()}${raw.slice(1)}`,
color: item?.color ?? agentTones[key] ?? tone(key),
color,
v2Color,
border: v2Tone?.border ?? `color-mix(in srgb, ${v2Color} 48%, transparent)`,
background: v2Tone?.background ?? `color-mix(in srgb, ${v2Color} 12%, transparent)`,
}
}
function agentColor(value: string | undefined, themeColors: Record<string, string>) {
if (!value) return
return themeColors[value] ?? value
}
function newLayout() {
return typeof document !== "undefined" && document.body.hasAttribute("data-new-layout")
}
function webSearchProviderLabel(provider: unknown) {
if (provider === "parallel") return "Parallel Web Search"
if (provider === "exa") return "Exa Web Search"
@@ -1882,6 +1936,9 @@ ToolRegistry.register({
const agent = createMemo(() => taskAgent(props.input.subagent_type, data.store.agent))
const title = createMemo(() => agent().name ?? i18n.t("ui.tool.agent.default"))
const tone = createMemo(() => agent().color)
const v2Tone = createMemo(() => agent().v2Color)
const border = createMemo(() => agent().border)
const background = createMemo(() => agent().background)
const subtitle = createMemo(() => {
const value =
typeof props.input.description === "string" && props.input.description
@@ -1921,17 +1978,27 @@ ToolRegistry.register({
}
const trigger = () => (
<div data-component="task-tool-card">
<div
data-component="task-tool-card"
style={{
"--task-agent-color": v2Tone(),
"--task-agent-legacy-color": tone(),
"--task-agent-border": border(),
"--task-agent-background": background(),
}}
>
<div data-slot="basic-tool-tool-info-structured">
<div data-slot="basic-tool-tool-info-main">
<Show when={running()}>
<span data-component="task-tool-spinner" style={{ color: tone() ?? "var(--icon-interactive-base)" }}>
<Spinner />
<Show when={newLayout()} fallback={<Spinner />}>
<SessionProgressIndicatorV2
style={{ color: v2Tone() ?? "light-dark(var(--v2-text-text-base), #ffffff)" }}
/>
</Show>
</span>
</Show>
<span data-component="task-tool-title" style={{ color: tone() ?? "var(--text-strong)" }}>
{title()}
</span>
<span data-component="task-tool-title">{title()}</span>
<Show when={subtitle()}>
<span data-slot="basic-tool-tool-subtitle">{subtitle()}</span>
</Show>
+18
View File
@@ -281,6 +281,15 @@
--icon-agent-docs-base: #fcb239;
--icon-agent-ask-base: #2090f5;
--icon-agent-build-base: #034cff;
--v2-agent-plan-solid: var(--v2-pink-900);
--v2-agent-plan-border: rgba(200, 61, 139, 0.5);
--v2-agent-plan-background: rgba(253, 236, 243, 0.33);
--v2-agent-build-solid: var(--v2-blue-900);
--v2-agent-build-border: rgba(59, 92, 246, 0.5);
--v2-agent-build-background: rgba(236, 241, 254, 0.33);
--v2-agent-explore-solid: var(--v2-yellow-900);
--v2-agent-explore-border: rgba(142, 114, 49, 0.5);
--v2-agent-explore-background: rgba(254, 250, 236, 0.33);
--icon-on-success-base: rgba(18, 201, 5, 0.9);
--icon-on-success-hover: rgba(45, 186, 38, 0.9);
--icon-on-success-selected: rgba(7, 137, 1, 0.9);
@@ -541,6 +550,15 @@
--icon-agent-docs-base: #fbb73c;
--icon-agent-ask-base: #2090f5;
--icon-agent-build-base: #9dbefe;
--v2-agent-plan-solid: var(--v2-pink-400);
--v2-agent-plan-border: rgba(250, 188, 216, 0.5);
--v2-agent-plan-background: rgba(247, 213, 228, 0.1);
--v2-agent-build-solid: var(--v2-blue-400);
--v2-agent-build-border: rgba(215, 226, 252, 0.5);
--v2-agent-build-background: rgba(215, 226, 252, 0.1);
--v2-agent-explore-solid: var(--v2-yellow-400);
--v2-agent-explore-border: rgba(247, 229, 181, 0.5);
--v2-agent-explore-background: rgba(252, 239, 208, 0.1);
--icon-on-success-base: rgba(18, 201, 5, 0.9);
--icon-on-success-hover: rgba(53, 192, 45, 0.9);
--icon-on-success-selected: rgba(77, 225, 68, 0.9);
+26
View File
@@ -3,6 +3,30 @@ import { V2_AVATAR_DARK, V2_AVATAR_LIGHT } from "./avatar"
const ref = (name: string): V2ColorValue => `var(--${name})`
const lightAgentTokens: Record<string, V2ColorValue> = {
"v2-agent-plan-solid": ref("v2-pink-900"),
"v2-agent-plan-border": "rgba(200, 61, 139, 0.5)",
"v2-agent-plan-background": "rgba(253, 236, 243, 0.33)",
"v2-agent-build-solid": ref("v2-blue-900"),
"v2-agent-build-border": "rgba(59, 92, 246, 0.5)",
"v2-agent-build-background": "rgba(236, 241, 254, 0.33)",
"v2-agent-explore-solid": ref("v2-yellow-900"),
"v2-agent-explore-border": "rgba(142, 114, 49, 0.5)",
"v2-agent-explore-background": "rgba(254, 250, 236, 0.33)",
}
const darkAgentTokens: Record<string, V2ColorValue> = {
"v2-agent-plan-solid": ref("v2-pink-400"),
"v2-agent-plan-border": "rgba(250, 188, 216, 0.5)",
"v2-agent-plan-background": "rgba(247, 213, 228, 0.1)",
"v2-agent-build-solid": ref("v2-blue-400"),
"v2-agent-build-border": "rgba(215, 226, 252, 0.5)",
"v2-agent-build-background": "rgba(215, 226, 252, 0.1)",
"v2-agent-explore-solid": ref("v2-yellow-400"),
"v2-agent-explore-border": "rgba(247, 229, 181, 0.5)",
"v2-agent-explore-background": "rgba(252, 239, 208, 0.1)",
}
const light: Record<string, V2ColorValue> = {
"v2-background-bg-base": ref("v2-grey-100"),
"v2-background-bg-deep": ref("v2-grey-200"),
@@ -46,6 +70,7 @@ const light: Record<string, V2ColorValue> = {
"v2-state-bg-info": ref("v2-blue-100"),
"v2-state-fg-info": ref("v2-blue-800"),
"v2-state-border-info": ref("v2-blue-300"),
...lightAgentTokens,
...V2_AVATAR_LIGHT,
"v2-elevation-raised":
"0px 2px 4px 0px var(--v2-alpha-dark-4), 0px 1px 2px -1px var(--v2-alpha-dark-8), 0px 0px 0px 0.5px var(--v2-alpha-dark-12), 0px 0px 0px 0px var(--v2-alpha-dark-0)",
@@ -110,6 +135,7 @@ const dark: Record<string, V2ColorValue> = {
"v2-state-bg-info": ref("v2-blue-1200"),
"v2-state-fg-info": ref("v2-blue-500"),
"v2-state-border-info": ref("v2-blue-900"),
...darkAgentTokens,
...V2_AVATAR_DARK,
"v2-elevation-raised":
"0px 2px 4px 0px var(--v2-alpha-dark-30), 0px 1px 2px 0px var(--v2-alpha-dark-30), 0px 0px 0px 0.5px var(--v2-alpha-light-16), 0px -0.5px 0px 0px var(--v2-alpha-light-6)",
@@ -81,6 +81,13 @@
--project-avatar-border: var(--v2-avatar-border-gray);
}
[data-slot="project-avatar-surface"][data-variant="outline"] {
--project-avatar-bg: var(--v2-background-bg-base);
--project-avatar-border: var(--v2-border-border-strong);
color: var(--v2-text-text-muted);
text-shadow: none;
}
[data-slot="project-avatar-surface"][data-has-image] {
background: var(--project-avatar-bg);
}
@@ -11,6 +11,7 @@ Saturated 16px project avatar with color variants and optional unread dot.
### Variants
- Color: orange, yellow, cyan, green, red, pink, blue, purple, gray.
- Outline: neutral, muted style for de-emphasized projects (e.g. recently closed).
- Image vs initial content state.
- Unread dot with corner mask when \`unread\` is set.
@@ -33,7 +34,7 @@ export default {
argTypes: {
variant: {
control: "select",
options: [...PROJECT_AVATAR_VARIANTS],
options: [...PROJECT_AVATAR_VARIANTS, "outline"],
},
},
args: {
@@ -62,6 +63,13 @@ export const AllVariants = {
),
}
export const Outline = {
args: {
fallback: "O",
variant: "outline",
},
}
export const Unread = {
args: {
fallback: "O",
@@ -26,10 +26,13 @@ export const PROJECT_AVATAR_VARIANTS = [
export type ProjectAvatarVariant = (typeof PROJECT_AVATAR_VARIANTS)[number]
// "outline" is a neutral, muted style (e.g. recently closed projects) and is not part of the color rotation.
export type ProjectAvatarStyle = ProjectAvatarVariant | "outline"
export interface ProjectAvatarProps extends ComponentProps<"div"> {
fallback: string
src?: string
variant?: ProjectAvatarVariant
variant?: ProjectAvatarStyle
unread?: boolean
}
+40
View File
@@ -77,6 +77,16 @@
--v2-state-fg-info: var(--v2-blue-800);
--v2-state-border-info: var(--v2-blue-300);
--v2-agent-plan-solid: var(--v2-pink-900);
--v2-agent-plan-border: rgba(200, 61, 139, 0.5);
--v2-agent-plan-background: rgba(253, 236, 243, 0.33);
--v2-agent-build-solid: var(--v2-blue-900);
--v2-agent-build-border: rgba(59, 92, 246, 0.5);
--v2-agent-build-background: rgba(236, 241, 254, 0.33);
--v2-agent-explore-solid: var(--v2-yellow-900);
--v2-agent-explore-border: rgba(142, 114, 49, 0.5);
--v2-agent-explore-background: rgba(254, 250, 236, 0.33);
/* ── Project avatar (fixed; theme-independent) ── */
--v2-avatar-fg: #ffffffff;
--v2-avatar-bg-orange: #ee7330ff;
@@ -193,6 +203,16 @@
--v2-state-fg-info: var(--v2-blue-500);
--v2-state-border-info: var(--v2-blue-900);
--v2-agent-plan-solid: var(--v2-pink-400);
--v2-agent-plan-border: rgba(250, 188, 216, 0.5);
--v2-agent-plan-background: rgba(247, 213, 228, 0.1);
--v2-agent-build-solid: var(--v2-blue-400);
--v2-agent-build-border: rgba(215, 226, 252, 0.5);
--v2-agent-build-background: rgba(215, 226, 252, 0.1);
--v2-agent-explore-solid: var(--v2-yellow-400);
--v2-agent-explore-border: rgba(247, 229, 181, 0.5);
--v2-agent-explore-background: rgba(252, 239, 208, 0.1);
--v2-elevation-raised:
0px 2px 4px 0px var(--v2-alpha-dark-30),
0px 1px 2px 0px var(--v2-alpha-dark-30),
@@ -296,6 +316,16 @@
--v2-state-fg-info: var(--v2-blue-800);
--v2-state-border-info: var(--v2-blue-300);
--v2-agent-plan-solid: var(--v2-pink-900);
--v2-agent-plan-border: rgba(200, 61, 139, 0.5);
--v2-agent-plan-background: rgba(253, 236, 243, 0.33);
--v2-agent-build-solid: var(--v2-blue-900);
--v2-agent-build-border: rgba(59, 92, 246, 0.5);
--v2-agent-build-background: rgba(236, 241, 254, 0.33);
--v2-agent-explore-solid: var(--v2-yellow-900);
--v2-agent-explore-border: rgba(142, 114, 49, 0.5);
--v2-agent-explore-background: rgba(254, 250, 236, 0.33);
--v2-elevation-raised:
0px 2px 4px 0px var(--v2-alpha-dark-4), 0px 1px 2px -1px var(--v2-alpha-dark-8),
0px 0px 0px 0.5px var(--v2-alpha-dark-12), 0px 0px 0px 0px var(--v2-alpha-dark-0);
@@ -406,6 +436,16 @@
--v2-state-fg-info: var(--v2-blue-500);
--v2-state-border-info: var(--v2-blue-900);
--v2-agent-plan-solid: var(--v2-pink-400);
--v2-agent-plan-border: rgba(250, 188, 216, 0.5);
--v2-agent-plan-background: rgba(247, 213, 228, 0.1);
--v2-agent-build-solid: var(--v2-blue-400);
--v2-agent-build-border: rgba(215, 226, 252, 0.5);
--v2-agent-build-background: rgba(215, 226, 252, 0.1);
--v2-agent-explore-solid: var(--v2-yellow-400);
--v2-agent-explore-border: rgba(247, 229, 181, 0.5);
--v2-agent-explore-background: rgba(252, 239, 208, 0.1);
--v2-avatar-fg: #ffffffff;
--v2-avatar-bg-orange: #723d22ff;
--v2-avatar-border-orange: #ff8648ff;