be20ea0f42
use SessionProcessor.Service, Session.Service, Agent.Service directly in layer
...
Yield effectified services instead of going through async facades. Eliminates
Effect→Promise→Effect double-bounce for processor.create, processor.process,
Session.get/touch/setPermission/updateMessage, and Agent.get/list. Await cancel
in session route. Remove redundant InstanceState.get in shellE ensuring block.
2026-03-29 12:21:57 -04:00
76373f0c73
use Session.Service and Agent.Service directly instead of Effect.promise wrappers
...
Yield the effectified services in the layer and call their methods
directly, eliminating the double Effect→Promise→Effect bounce through
async facades. Layer.unwrap(Effect.sync(...)) breaks the circular
import. Also improves the assertNotBusy test with a proper gate/spy
so it deterministically catches the busy state.
2026-03-29 12:21:56 -04:00
5719a13cbf
refactor(session): effectify SessionPrompt service
...
Migrate SessionPrompt to the Effect service pattern (Interface, Service,
Layer, InstanceState, makeRuntime + async facades).
Key design decisions:
- Fiber-based cancellation replaces manual AbortController management.
Effect.promise((signal) => ...) derives AbortSignals automatically;
cancel() interrupts fibers and signals propagate to the AI SDK,
shell processes, and tool execution.
- Deferred queue replaces Promise callback queue. Concurrent loop()
callers get a Deferred that resolves when the running fiber finishes.
On cancel or error, queued callers now receive proper errors instead
of hanging forever.
- Separate loops/shells maps in InstanceState replace the single shared
state object, with shell-to-loop handoff preserved: if callers queue
a loop while a shell is running, shellE cleanup starts the loop.
- Heavy helper functions (createUserMessage, handleSubtask, shellImpl,
resolveCommand, insertReminders, ensureTitle) stay as plain async
functions called via Effect.promise, keeping the migration incremental.
- resolveTools and createStructuredOutputTool are unchanged (deeply tied
to AI SDK tool callbacks).
2026-03-29 12:21:56 -04:00
860531c275
refactor(session): effectify session processor ( #19485 )
2026-03-28 12:09:47 -04:00
2145d97f18
refactor(session): effectify SessionCompaction service ( #19459 )
2026-03-28 01:09:14 +00:00
c33d9996f0
feat: AI SDK v6 support ( #18433 )
2026-03-27 15:24:30 -05:00
7123aad5a8
fix(opencode): classify ZlibError from Bun fetch as retryable instead of unknown ( #19104 )
...
Co-authored-by: Aiden Cline <63023139+rekram1-node@users.noreply.github.com >
2026-03-25 13:08:40 -05:00
b0017bf1b9
feat(core): initial implementation of syncing ( #17814 )
2026-03-25 10:47:40 -04:00
814a515a8a
fix: improve plugin system robustness — agent/command resolution, async errors, hook timing, two-phase init ( #18280 )
...
Co-authored-by: Aiden Cline <63023139+rekram1-node@users.noreply.github.com >
2026-03-24 12:50:55 -05:00
2dbcd79fd2
fix: stabilize agent and skill ordering in prompt descriptions ( #18261 )
...
Co-authored-by: Aiden Cline <aidenpcline@gmail.com >
Co-authored-by: Aiden Cline <63023139+rekram1-node@users.noreply.github.com >
2026-03-19 15:04:03 -05:00
84e62fc662
fix(session): preserve tagged error messages ( #18165 )
2026-03-18 20:36:53 -04:00
e718db624f
fix(core): consider code: context_length_exceeded as context overflow in API call errors ( #17748 )
2026-03-16 10:25:24 -05:00
59c530cc6c
fix(opencode): teach Kit's test what an ID is ( #17745 )
2026-03-16 18:08:27 +10:00
c2ca1494e5
fix(opencode): preserve prompt tool enables with empty agent permissions ( #17064 )
...
Co-authored-by: jquense <jquense@ramp.com >
2026-03-15 23:01:46 -05:00
9457493696
perf(server): paginate session history ( #17134 )
2026-03-13 15:48:43 +05:30
1cb7df7159
refactor(provider): flow branded ProviderID/ModelID through internal signatures ( #17182 )
2026-03-12 14:48:17 +00:00
c45467964c
feat(id): brand ProviderID and ModelID ( #17110 )
2026-03-12 09:27:52 -04:00
090f636354
feat(id): brand PartID through Drizzle and Zod schemas ( #16966 )
2026-03-11 23:40:50 +00:00
16a6d6feba
feat(id): brand WorkspaceID through Drizzle and Zod schemas ( #16964 )
2026-03-11 23:30:17 +00:00
cb67465675
feat(id): brand SessionID through Drizzle and Zod schemas ( #16953 )
2026-03-11 23:16:56 +00:00
5acfdd1c5d
chore: kill old copilot 403 message that was used for old plugin migration ( #16904 )
2026-03-10 16:20:41 -05:00
3ebba02d04
refactor: replace Bun.sleep with node timers ( #15013 )
publish-github-action / publish (push) Has been cancelled
2026-03-05 21:54:06 -05:00
e41b53504f
chore: generate
2026-03-03 03:53:50 +00:00
fd6f7133c5
fix(opencode): clone part data in Bus event to preserve token values ( #15780 )
2026-03-02 22:52:43 -05:00
04a634a80d
test: merge test files into a single file ( #14366 )
2026-02-19 20:32:01 -06:00
cb5a0de42f
core: remove User-Agent header assertion from LLM test to fix failing test
2026-02-19 18:37:55 -06:00
02a9495063
Remove use of Bun.file ( #14215 )
2026-02-19 11:32:32 -05:00
e35a4131d0
core: keep message part order stable when files resolve asynchronously ( #13915 )
2026-02-16 18:45:11 -06:00
e269788a8f
feat: support claude agent SDK-style structured outputs in the OpenCode SDK ( #8161 )
...
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com >
Co-authored-by: Dax Raad <d@ironbay.co >
2026-02-12 04:54:05 +00:00
0fd6f365be
fix(core): ensure compaction is more reliable, add reserve token buffer to ensure that input window has enough room to compact ( #12924 )
...
Co-authored-by: James Lal <james@littlebearlabs.io >
2026-02-10 19:55:22 -06:00
f74c0339cc
test: fix failing prompt test ( #12847 )
2026-02-09 11:25:22 -06:00
99ea1351ce
tweak: add new ContextOverflowError type ( #12777 )
2026-02-08 23:54:01 -06:00
62f38087b8
fix: parse mid stream openai responses style errors to prevent infinite retries for errors that should STOP execution ( #12768 )
2026-02-08 20:55:41 -06:00
fde0b39b7c
fix: properly encode file URLs with special characters ( #12424 )
2026-02-06 16:16:56 -06:00
1721c6efdf
fix(core): session errors when attachment file not found
2026-02-04 10:21:03 -06:00
6daa962aaa
fix: prioritize OPENCODE_CONFIG_DIR for AGENTS.md ( #11536 )
2026-02-04 01:35:22 +00:00
aa6b552c39
Revert pr that was mistakenly merged ( #11844 )
2026-02-02 15:28:02 -06:00
5db089070a
test: add unit test
2026-02-01 19:26:23 -05:00
f15755684f
fix(opencode): scope agent variant to model ( #11410 )
2026-02-01 14:12:30 -06:00
16145af480
fix: prevent duplicate AGENTS.md injection when reading instruction files ( #11581 )
...
Co-authored-by: Aiden Cline <aidenpcline@gmail.com >
2026-02-01 13:46:08 -06:00
da7c874808
tweak: show actual retry error message instead of generic error msg ( #11520 )
2026-01-31 13:15:42 -06:00
f834915d3f
test: fix flaky test ( #11427 )
2026-01-31 01:07:22 -06:00
9d3f32065b
test: add llm.test.ts ( #11375 )
2026-01-30 17:04:32 -06:00
558590712d
fix: ensure parallel tool calls dont double load AGENTS.md
2026-01-28 01:38:10 -05:00
f12f7e7718
tweak: adjust retry check to be more defensive
2026-01-27 11:31:02 -05:00
8b5dde5536
tweak: retry logic to catch certain provider problems
2026-01-26 16:12:41 -05:00
18bfc740c8
chore: generate
2026-01-26 18:50:58 +00:00
837037cd04
fix: ensure openai 404 errors are retried ( #10590 )
2026-01-26 13:50:04 -05:00
39a73d4894
feat: dynamically resolve AGENTS.md files from subdirectories as agent explores them ( #10678 )
2026-01-26 10:49:41 -05:00
c2844697f3
fix: ensure images are properly returned as tool results
2026-01-21 23:54:44 -06:00
Kit Langton