fix(core): stream shell progress tail

Co-authored-by: Kit Langton <kit.langton@gmail.com>

packages/ai/README.md

@@ -127,23 +127,37 @@ OpenAI Chat and OpenAI Responses are separate semantic entrypoints:
 - `@opencode-ai/ai/providers/openai/responses`
 - `@opencode-ai/ai/providers/openai-compatible/responses`
 - `@opencode-ai/ai/providers/anthropic-compatible`
-- `@opencode-ai/ai/providers/google-vertex`
-- `@opencode-ai/ai/providers/google-vertex/anthropic`
+- `@opencode-ai/ai/providers/google-vertex/gemini`
+- `@opencode-ai/ai/providers/google-vertex/chat`
+- `@opencode-ai/ai/providers/google-vertex/responses`
+- `@opencode-ai/ai/providers/google-vertex/messages`
 
 Responses HTTP versus WebSocket is a scoped `transport` setting on the OpenAI Responses entrypoint, not another entrypoint. Azure follows the same Chat/Responses split at `providers/azure/chat` and `providers/azure/responses`. Generic OpenAI-compatible Chat remains at `providers/openai-compatible`; compatible Responses is separate at `providers/openai-compatible/responses`. Generic Anthropic Messages-compatible providers use `providers/anthropic-compatible`, which the named Anthropic provider composes. Google Gemini and Amazon Bedrock expose their single native API through their existing provider paths.
 
-Vertex Gemini and Vertex Anthropic are separate products with separate entrypoints. Both accept `project`, `location`, and an optional `accessToken`; when no explicit token or auth override is supplied they lazily use Google Application Default Credentials. Vertex Gemini instead selects express mode when `apiKey` or `GOOGLE_VERTEX_API_KEY` is present.
+Vertex Gemini, Vertex Chat, Vertex Responses, and Vertex Messages are separate API entrypoints. All accept `project`, `location`, and an optional `accessToken`; when no explicit token or auth override is supplied they lazily use Google Application Default Credentials. Vertex Gemini instead selects express mode when `apiKey` or `GOOGLE_VERTEX_API_KEY` is present. Vertex Chat targets MaaS models through the OpenAI-compatible Chat Completions endpoint, while Vertex Responses targets Grok models and defaults `store` to `false` as required by Vertex. `providers/google-vertex` remains the default alias for `providers/google-vertex/gemini`.
 
 Tuned Vertex Gemini deployments use model ids shaped like `endpoints/1234567890` and require OAuth or ADC; Vertex express-mode API keys support publisher models only.
 
 ```ts
-import { model } from "@opencode-ai/ai/providers/google-vertex"
+import { model } from "@opencode-ai/ai/providers/google-vertex/gemini"
 
 model("gemini-3.5-flash", { project: "my-project", location: "global" })
 ```
 
 ```ts
-import { model } from "@opencode-ai/ai/providers/google-vertex/anthropic"
+import { model } from "@opencode-ai/ai/providers/google-vertex/chat"
+
+model("deepseek-ai/deepseek-v3.2-maas", { project: "my-project", location: "global" })
+```
+
+```ts
+import { model } from "@opencode-ai/ai/providers/google-vertex/responses"
+
+model("xai/grok-4.20-reasoning", { project: "my-project", location: "global" })
+```
+
+```ts
+import { model } from "@opencode-ai/ai/providers/google-vertex/messages"
 
 model("claude-sonnet-4-6", { project: "my-project", location: "global" })
 ```

packages/ai/STATUS.md

@@ -1,15 +1,15 @@
 # LLM Provider Parity Status
 
-Last reviewed: 2026-07-15
+Last reviewed: 2026-07-16
 
 This file tracks the gap between the native `@opencode-ai/ai` package and the AI SDK provider packages that opencode still depends on for many catalog/runtime paths.
 
 ## Existing Status Sources
 
-| File                                 | What it tracks                                                                   | Limitation                                              |
-| ------------------------------------ | -------------------------------------------------------------------------------- | ------------------------------------------------------- |
-| `packages/ai/DESIGN.md`             | Future clean-break API proposal for `@opencode-ai/ai`.                       | Not a provider parity tracker.                          |
-| `packages/ai/example/call-sites.md` | Route/value/provider-facade migration checklist and call-site sketches.          | Architecture migration only; not AI SDK package parity. |
+| File                                | What it tracks                                                          | Limitation                                              |
+| ----------------------------------- | ----------------------------------------------------------------------- | ------------------------------------------------------- |
+| `packages/ai/DESIGN.md`             | Future clean-break API proposal for `@opencode-ai/ai`.                  | Not a provider parity tracker.                          |
+| `packages/ai/example/call-sites.md` | Route/value/provider-facade migration checklist and call-site sketches. | Architecture migration only; not AI SDK package parity. |
 
 ## Current Implementation Snapshot
 
@@ -23,8 +23,10 @@ This file tracks the gap between the native `@opencode-ai/ai` package and the AI
 | Anthropic-compatible Messages      | `src/protocols/anthropic-messages.ts`, `src/providers/anthropic-compatible.ts`                 | Usable for deployments that implement the Anthropic Messages wire protocol. Named Anthropic composes this base.                                                          | No named compatible family profiles or recorded deployment coverage yet.                                                                                                                                                                                  |
 | Anthropic Messages                 | `src/protocols/anthropic-messages.ts`, `src/providers/anthropic.ts`                            | Usable. Supports tools, thinking, cache control, images, server-hosted tool events, and usage.                                                                           | Provider option surface is small. Beta/header handling, metadata, and newer Messages fields need a typed parity pass.                                                                                                                                     |
 | Gemini Developer API               | `src/protocols/gemini.ts`, `src/providers/google.ts`                                           | Usable for Google API key flow. Supports text, images, tools, thinking signatures, and cache usage.                                                                      | This is not Vertex. Typed provider options are narrow; many Gemini request fields currently require raw `http.body` overlays.                                                                                                                             |
-| Vertex Gemini                      | `src/protocols/google-vertex-gemini.ts`, `src/providers/google-vertex.ts`                      | Usable through API-key express mode, explicit OAuth tokens, or ADC with project/location endpoint derivation, including tuned `endpoints/...` deployments.               | Core runner/catalog mapping and recorded provider coverage are missing.                                                                                                                                                                                   |
-| Vertex Anthropic Messages          | `src/protocols/google-vertex-anthropic.ts`, `src/providers/google-vertex-anthropic.ts`         | Usable through explicit OAuth tokens or ADC, including global, regional, and `eu`/`us` multi-region endpoints.                                                           | Core runner/catalog mapping and recorded provider coverage are missing; Vertex-specific hosted-tool parity needs review.                                                                                                                                  |
+| Vertex Gemini                      | `src/protocols/gemini.ts`, `src/providers/google-vertex.ts`                                    | Usable through API-key express mode, explicit OAuth tokens, or ADC with project/location endpoint derivation, including tuned `endpoints/...` deployments.               | Core runner/catalog mapping and recorded provider coverage are missing.                                                                                                                                                                                   |
+| Vertex Chat                        | `src/protocols/openai-chat.ts`, `src/providers/google-vertex-chat.ts`                          | Usable for MaaS models through OpenAI-compatible Chat Completions with explicit OAuth tokens or ADC and project/location endpoint derivation.                            | Core runner/catalog mapping and recorded provider coverage are missing; MaaS family-specific request parity needs review.                                                                                                                                 |
+| Vertex Responses                   | `src/protocols/openai-responses.ts`, `src/providers/google-vertex-responses.ts`                | Usable for Grok models through OpenAI-compatible Responses with explicit OAuth tokens or ADC, project/location endpoint derivation, and storage disabled by default.     | Core runner/catalog mapping and recorded provider coverage are missing; stateful continuation is not supported by Vertex.                                                                                                                                 |
+| Vertex Messages                    | `src/protocols/anthropic-messages.ts`, `src/providers/google-vertex-messages.ts`               | Usable through explicit OAuth tokens or ADC, including global, regional, and `eu`/`us` multi-region endpoints.                                                           | Core runner/catalog mapping and recorded provider coverage are missing; Vertex-specific hosted-tool parity needs review.                                                                                                                                  |
 | Bedrock Converse                   | `src/protocols/bedrock-converse.ts`, `src/providers/amazon-bedrock.ts`                         | Partial but real. Supports AWS event-stream framing, SigV4 with supplied credentials, bearer auth, tools, reasoning signatures, media, cache points, and recorded tests. | Native facade does not mirror the AI SDK plugin's default AWS credential chain/profile behavior. Runner/catalog mapping is missing. Guardrails, inference profiles, region-specific model ID fixes, and model-specific request fields need a parity pass. |
 | Azure OpenAI                       | `src/providers/azure.ts` using OpenAI Chat/Responses protocols                                 | Partial. Supports resource/base URL setup, API key auth, API version query, Chat, and Responses selectors.                                                               | Core runner does not map `@ai-sdk/azure` to this native facade. AAD/token auth and Azure-specific endpoint variants need review.                                                                                                                          |
 | Cloudflare AI Gateway / Workers AI | `src/providers/cloudflare.ts`                                                                  | Present via OpenAI-compatible Chat routes.                                                                                                                               | Useful but not part of the critical AI SDK replacement set yet. Needs per-product recorded coverage before relying on it broadly.                                                                                                                         |
@@ -54,8 +56,8 @@ Other `aisdk:` packages, including Google Vertex, Azure, and Bedrock, currently
 | `@ai-sdk/google`                  | Gemini Developer API                                           | Partial / usable | Add typed options for safety, response schema/modalities, cached content, grounding/search/code execution, and non-text output modes where supported.                                                               |
 | `@ai-sdk/google-vertex`           | Vertex Gemini namespace/facade                                 | Partial / usable | Add runner/catalog mapping, recorded coverage, and broader provider-option parity.                                                                                                                                  |
 | `@ai-sdk/google-vertex/anthropic` | Anthropic Messages over Vertex namespace/facade                | Partial / usable | Add runner/catalog mapping, recorded coverage, and Vertex-specific hosted-tool parity.                                                                                                                              |
-| `@ai-sdk/google-vertex/maas`      | Vertex MaaS OpenAI-compatible namespace/facade                 | Missing          | Decide native Chat/Responses selection, endpoint derivation, auth, and catalog mapping.                                                                                                                             |
-| `@ai-sdk/google-vertex/xai`       | Vertex xAI OpenAI-compatible namespace/facade                  | Missing          | Decide whether this composes the generic compatible bases or the xAI facade, then add endpoint/auth mapping and tests.                                                                                              |
+| `@ai-sdk/google-vertex/maas`      | Vertex Chat                                                    | Partial / usable | Add runner/catalog mapping, recorded coverage, and MaaS family-specific request parity.                                                                                                                             |
+| `@ai-sdk/google-vertex/xai`       | Vertex Chat / Responses                                        | Partial / usable | Decide Chat/Responses selection for catalog models, add runner mapping and recorded coverage, and review xAI-specific request options.                                                                              |
 | `@ai-sdk/azure`                   | Azure OpenAI Chat/Responses facade                             | Partial          | Map runner/catalog metadata to native Azure, handle resourceName/baseURL/apiVersion variants, add AAD/token auth story, and verify Chat vs Responses deployment selection.                                          |
 | `@ai-sdk/amazon-bedrock`          | Bedrock Converse                                               | Partial          | Add default AWS credential chain/profile support, region/inference-profile model ID handling, provider option parity via `additionalModelRequestFields`, guardrails/performance config, and runner/catalog mapping. |
 | `@ai-sdk/amazon-bedrock/mantle`   | Bedrock Mantle OpenAI-compatible Chat/Responses namespace      | Missing          | Decide native Mantle shape, likely separate from Converse because it uses OpenAI-compatible Chat/Responses semantics over Bedrock. Add package mapping and tests.                                                   |
@@ -65,19 +67,19 @@ Other `aisdk:` packages, including Google Vertex, Azure, and Bedrock, currently
 1. Runner support is narrower than the LLM package. The package has native provider facades for Google, Azure, and Bedrock, but the V2 Session runner only maps OpenAI, Anthropic, and explicit OpenAI-compatible Chat from `aisdk` catalog metadata.
 2. OpenAI-compatible Responses is available as a separate package entrypoint, but the V2 runner still maps `@ai-sdk/openai-compatible` to Chat only. Catalog selection must become API-aware before Responses deployments can use it.
 3. Bedrock native auth is not AI SDK parity. The AI SDK plugin uses the default AWS provider chain, profile, container credentials, and Bedrock bearer token env behavior. Native Bedrock currently expects explicit credentials or bearer auth on the facade.
-4. Vertex Gemini and Vertex Anthropic now have native package entrypoints, but the core runner does not map catalog metadata to them yet and recorded provider coverage is still missing.
+4. Vertex Gemini, Vertex Chat, Vertex Responses, and Vertex Messages now have native package entrypoints, but the core runner does not map catalog metadata to them yet and recorded provider coverage is still missing.
 5. Azure is only a provider facade, not a full runtime replacement. Native Azure exists, but the catalog runner does not select it, and token auth/resource variants need review.
 6. Provider option typing is uneven. OpenAI, Anthropic, Gemini, Bedrock, and OpenRouter each expose a small typed subset plus raw HTTP overlays; this is useful but not equivalent to AI SDK provider option coverage.
 7. Structured output is not provider-native yet. `LLM.generateObject` still uses a synthetic tool strategy, while the future design expects native structured output where reliable and tool fallback where needed.
-8. Package/namespace boundaries for the current native loading set are explicit in docs and exports. Other exported provider facades are not catalog package entrypoints until they implement the contract. Missing native boundaries remain for Vertex MaaS, Vertex xAI, and Bedrock Mantle.
+8. Package/namespace boundaries for the current native loading set are explicit in docs and exports. Other exported provider facades are not catalog package entrypoints until they implement the contract. Vertex xAI still needs catalog API selection; the missing native boundary is Bedrock Mantle.
 9. Recorded coverage is uneven. OpenAI, Anthropic, Gemini, Bedrock Converse, Cloudflare, OpenRouter, and several OpenAI-compatible Chat providers have cassettes. Azure, Vertex, and Mantle need first-class recorded scenarios before switching defaults.
 
 ## Native Namespace Shape
 
 These are implementation/API slices, not separate npm packages.
 
-| API slice                     | Package-like entrypoint                                  | Purpose                                                                      |
-| ----------------------------- | -------------------------------------------------------- | ---------------------------------------------------------------------------- |
+| API slice                     | Package-like entrypoint                                 | Purpose                                                                      |
+| ----------------------------- | ------------------------------------------------------- | ---------------------------------------------------------------------------- |
 | OpenAI Chat                   | `@opencode-ai/ai/providers/openai/chat`                 | OpenAI `/chat/completions` semantics.                                        |
 | OpenAI Responses              | `@opencode-ai/ai/providers/openai/responses`            | OpenAI `/responses` semantics with HTTP/WebSocket selected through settings. |
 | OpenAI-compatible Chat        | `@opencode-ai/ai/providers/openai-compatible`           | Generic OpenAI-compatible `/chat/completions`.                               |
@@ -85,12 +87,12 @@ These are implementation/API slices, not separate npm packages.
 | Anthropic-compatible Messages | `@opencode-ai/ai/providers/anthropic-compatible`        | Generic Anthropic-compatible `/messages`.                                    |
 | Anthropic Messages            | `@opencode-ai/ai/providers/anthropic`                   | Anthropic Messages API.                                                      |
 | Gemini Developer API          | `@opencode-ai/ai/providers/google`                      | Google AI Studio Gemini API.                                                 |
-| Vertex Gemini                 | `@opencode-ai/ai/providers/google-vertex`               | Vertex Gemini API.                                                           |
-| Vertex Anthropic Messages     | `@opencode-ai/ai/providers/google-vertex/anthropic`     | Vertex-hosted Anthropic Messages API.                                        |
-| Vertex MaaS                   | Missing                                                  | Vertex OpenAI-compatible MaaS APIs.                                          |
-| Vertex xAI                    | Missing                                                  | Vertex-hosted xAI APIs.                                                      |
+| Vertex Gemini                 | `@opencode-ai/ai/providers/google-vertex/gemini`        | Vertex Gemini API; `providers/google-vertex` is the default alias.           |
+| Vertex Chat                   | `@opencode-ai/ai/providers/google-vertex/chat`          | Vertex OpenAI-compatible Chat Completions for MaaS models.                   |
+| Vertex Responses              | `@opencode-ai/ai/providers/google-vertex/responses`     | Vertex OpenAI-compatible Responses for Grok models.                          |
+| Vertex Messages               | `@opencode-ai/ai/providers/google-vertex/messages`      | Vertex-hosted Anthropic Messages API.                                        |
 | Bedrock Converse              | `@opencode-ai/ai/providers/amazon-bedrock`              | AWS Bedrock Converse API.                                                    |
-| Bedrock Mantle                | Missing                                                  | AWS Bedrock Mantle OpenAI-compatible APIs.                                   |
+| Bedrock Mantle                | Missing                                                 | AWS Bedrock Mantle OpenAI-compatible APIs.                                   |
 | Azure OpenAI Chat             | `@opencode-ai/ai/providers/azure/chat`                  | Azure specialization of OpenAI Chat.                                         |
 | Azure OpenAI Responses        | `@opencode-ai/ai/providers/azure/responses`             | Azure specialization of OpenAI Responses.                                    |
 
@@ -99,8 +101,8 @@ These are implementation/API slices, not separate npm packages.
 1. Add native runner/catalog mappings for `@ai-sdk/azure`, `@ai-sdk/google`, and `@ai-sdk/amazon-bedrock` where the existing native facades are already close.
 2. Add API-aware runner/catalog selection between OpenAI-compatible Chat and Responses.
 3. Bring Bedrock native auth/config to AI SDK parity: region, profile, default AWS credential chain, bearer token env, endpoint override, and cross-region inference profile handling.
-4. Add runner/catalog mappings and recorded scenarios for the native Vertex Gemini and Vertex Anthropic entrypoints.
-5. Add native Vertex MaaS and Vertex xAI entrypoints by composing the compatible bases and shared Vertex auth/endpoint setup.
+4. Add runner/catalog mappings and recorded scenarios for the native Vertex Gemini, Chat, Responses, and Messages entrypoints.
+5. Decide Chat/Responses selection for `@ai-sdk/google-vertex/xai` catalog models.
 6. Add Bedrock Mantle as a separate OpenAI-compatible Bedrock namespace after deciding whether it uses Chat, Responses, or both by model.
 7. Expand typed provider options from the existing V1 lowerer knowledge in `packages/core/src/v1/config/provider-options.ts` before adding more raw overlay examples.
-8. Add recorded provider tests for Azure, Vertex Gemini, Vertex Anthropic, Bedrock credential-chain behavior, and Mantle before making native runtime the default for those packages.
+8. Add recorded provider tests for Azure, Vertex Gemini, Vertex Chat, Vertex Responses, Vertex Messages, Bedrock credential-chain behavior, and Mantle before making native runtime the default for those packages.

packages/ai/example/call-sites.md

@@ -360,17 +360,29 @@ import { model } from "@opencode-ai/ai/providers/openai/responses"
 model("gpt-4o", { apiKey, transport: "websocket" })
 ```
 
-Vertex keeps Gemini and Anthropic Messages as separate package-like entrypoints,
+Vertex keeps Gemini, Chat, Responses, and Messages as separate package-like entrypoints,
 while sharing project/location resolution and ADC authentication internally:
 
 ```ts
-import { model } from "@opencode-ai/ai/providers/google-vertex"
+import { model } from "@opencode-ai/ai/providers/google-vertex/gemini"
 
 model("gemini-3.5-flash", { project, location: "global" })
 ```
 
 ```ts
-import { model } from "@opencode-ai/ai/providers/google-vertex/anthropic"
+import { model } from "@opencode-ai/ai/providers/google-vertex/chat"
+
+model("deepseek-ai/deepseek-v3.2-maas", { project, location: "global" })
+```
+
+```ts
+import { model } from "@opencode-ai/ai/providers/google-vertex/responses"
+
+model("xai/grok-4.20-reasoning", { project, location: "global" })
+```
+
+```ts
+import { model } from "@opencode-ai/ai/providers/google-vertex/messages"
 
 model("claude-sonnet-4-6", { project, location: "global" })
 ```

packages/ai/src/protocols/google-vertex-anthropic.ts

@@ -1,42 +0,0 @@
-import { Effect, Schema, Struct } from "effect"
-import { AnthropicMessages } from "./anthropic-messages"
-import { Auth } from "../route/auth"
-import { Route } from "../route/client"
-import { Endpoint } from "../route/endpoint"
-import { Framing } from "../route/framing"
-import { Protocol } from "../route/protocol"
-
-const VERSION = "vertex-2023-10-16" as const
-
-export const GoogleVertexAnthropicBody = Schema.Struct({
-  ...Struct.omit(AnthropicMessages.AnthropicMessagesBody.fields, ["model"]),
-  anthropic_version: Schema.Literal(VERSION),
-})
-export type GoogleVertexAnthropicBody = Schema.Schema.Type<typeof GoogleVertexAnthropicBody>
-
-export const protocol = Protocol.make({
-  id: "google-vertex-anthropic",
-  body: {
-    schema: GoogleVertexAnthropicBody,
-    from: (request) =>
-      AnthropicMessages.protocol.body.from(request).pipe(
-        Effect.map((body) => ({
-          ...Struct.omit(body, ["model"]),
-          anthropic_version: VERSION,
-        })),
-      ),
-  },
-  stream: AnthropicMessages.protocol.stream,
-})
-
-export const route = Route.make({
-  id: "google-vertex-anthropic",
-  provider: "google-vertex-anthropic",
-  providerMetadataKey: "anthropic",
-  protocol,
-  endpoint: Endpoint.path(({ request }) => `/${request.model.id}:streamRawPredict`),
-  auth: Auth.none,
-  framing: Framing.sse,
-})
-
-export * as GoogleVertexAnthropic from "./google-vertex-anthropic"

packages/ai/src/protocols/google-vertex-gemini.ts

@@ -1,20 +0,0 @@
-import { Gemini } from "./gemini"
-import { Auth } from "../route/auth"
-import { Route } from "../route/client"
-import { Endpoint } from "../route/endpoint"
-import { Framing } from "../route/framing"
-
-export const route = Route.make({
-  id: "google-vertex-gemini",
-  provider: "google-vertex",
-  providerMetadataKey: "google",
-  protocol: Gemini.protocol,
-  endpoint: Endpoint.path(({ request }) => {
-    const model = String(request.model.id)
-    return `/${model.startsWith("endpoints/") ? model : `models/${model}`}:streamGenerateContent?alt=sse`
-  }),
-  auth: Auth.none,
-  framing: Framing.sse,
-})
-
-export * as GoogleVertexGemini from "./google-vertex-gemini"

packages/ai/src/protocols/index.ts

@@ -1,6 +1,4 @@
 export * as AnthropicMessages from "./anthropic-messages"
-export * as GoogleVertexAnthropic from "./google-vertex-anthropic"
-export * as GoogleVertexGemini from "./google-vertex-gemini"
 export * as BedrockConverse from "./bedrock-converse"
 export * as Gemini from "./gemini"
 export * as OpenAIChat from "./openai-chat"

packages/ai/src/providers/google-vertex-chat.ts

@@ -0,0 +1,81 @@
+import type { ProviderPackage } from "../provider-package"
+import { OpenAICompatibleChat } from "../protocols/openai-compatible-chat"
+import type { RouteDefaultsInput } from "../route/client"
+import { ProviderID, type ModelID, type ProviderOptions } from "../schema"
+import { GoogleVertexShared } from "./google-vertex-shared"
+
+export const id = ProviderID.make("google-vertex")
+
+export type Config = RouteDefaultsInput &
+  GoogleVertexShared.OAuthOptions & {
+    readonly baseURL?: string
+    readonly location?: string
+    readonly project?: string
+  }
+
+export interface Settings extends ProviderPackage.Settings {
+  readonly accessToken?: string
+  readonly apiKey?: never
+  readonly baseURL?: string
+  readonly location?: string
+  readonly project?: string
+  readonly providerOptions?: ProviderOptions
+}
+
+const route = OpenAICompatibleChat.route.with({
+  id: "google-vertex-chat",
+  provider: id,
+})
+
+export const routes = [route]
+
+const configuredRoute = (input: Config) => {
+  if ("apiKey" in input && input.apiKey !== undefined) throw new Error("Google Vertex Chat does not support API keys")
+  const {
+    accessToken: _accessToken,
+    auth: _auth,
+    baseURL,
+    location: inputLocation,
+    project: inputProject,
+    ...rest
+  } = input
+  const location = GoogleVertexShared.location(inputLocation, "global")
+  const project = GoogleVertexShared.project(inputProject)
+  return route.with({
+    ...rest,
+    endpoint: {
+      baseURL:
+        baseURL ??
+        `https://aiplatform.googleapis.com/v1/projects/${GoogleVertexShared.requireProject(project)}/locations/${location}/endpoints/openapi`,
+    },
+    auth: GoogleVertexShared.oauth(input, project),
+  })
+}
+
+export const configure = (input: Config = {}) => {
+  const route = configuredRoute(input)
+  return {
+    id,
+    model: (modelID: string | ModelID) => route.model({ id: modelID }),
+    configure,
+  }
+}
+
+export const provider = {
+  id,
+  configure,
+}
+
+export const model: ProviderPackage.Definition<Settings>["model"] = (modelID, settings) => {
+  if (settings.apiKey !== undefined) throw new Error("Google Vertex Chat does not support API keys")
+  return configure({
+    accessToken: settings.accessToken,
+    baseURL: settings.baseURL,
+    headers: settings.headers === undefined ? undefined : { ...settings.headers },
+    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
+    limits: settings.limits,
+    location: settings.location,
+    project: settings.project,
+    providerOptions: settings.providerOptions,
+  }).model(modelID)
+}

packages/ai/src/providers/google-vertex-messages.ts

@@ -0,0 +1,111 @@
+import { Effect, Schema, Struct } from "effect"
+import type { ProviderPackage } from "../provider-package"
+import { AnthropicMessages } from "../protocols/anthropic-messages"
+import { Auth } from "../route/auth"
+import { Route, type RouteDefaultsInput } from "../route/client"
+import { Endpoint } from "../route/endpoint"
+import { Framing } from "../route/framing"
+import { Protocol } from "../route/protocol"
+import { ProviderID, type ModelID, type ProviderOptions } from "../schema"
+import { GoogleVertexShared } from "./google-vertex-shared"
+
+const VERSION = "vertex-2023-10-16" as const
+
+// models.dev uses this provider id even though the API contract is Anthropic Messages.
+export const id = ProviderID.make("google-vertex-anthropic")
+
+export type Config = RouteDefaultsInput &
+  GoogleVertexShared.OAuthOptions & {
+    readonly baseURL?: string
+    readonly location?: string
+    readonly project?: string
+  }
+
+export interface Settings extends ProviderPackage.Settings {
+  readonly accessToken?: string
+  readonly apiKey?: never
+  readonly baseURL?: string
+  readonly location?: string
+  readonly project?: string
+  readonly providerOptions?: ProviderOptions
+}
+
+const route = Route.make({
+  id: "google-vertex-messages",
+  provider: id,
+  providerMetadataKey: "anthropic",
+  protocol: Protocol.make({
+    id: AnthropicMessages.protocol.id,
+    body: {
+      schema: Schema.Struct({
+        ...Struct.omit(AnthropicMessages.AnthropicMessagesBody.fields, ["model"]),
+        anthropic_version: Schema.Literal(VERSION),
+      }),
+      from: (request) =>
+        AnthropicMessages.protocol.body.from(request).pipe(
+          Effect.map((body) => ({
+            ...Struct.omit(body, ["model"]),
+            anthropic_version: VERSION,
+          })),
+        ),
+    },
+    stream: AnthropicMessages.protocol.stream,
+  }),
+  endpoint: Endpoint.path(({ request }) => `/${request.model.id}:streamRawPredict`),
+  auth: Auth.none,
+  framing: Framing.sse,
+})
+
+export const routes = [route]
+
+const configuredRoute = (input: Config) => {
+  if ("apiKey" in input && input.apiKey !== undefined)
+    throw new Error("Google Vertex Messages does not support API keys")
+  const {
+    accessToken: _accessToken,
+    auth: _auth,
+    baseURL,
+    location: inputLocation,
+    project: inputProject,
+    ...rest
+  } = input
+  const location = GoogleVertexShared.location(inputLocation, "global")
+  const project = GoogleVertexShared.project(inputProject)
+  return route.with({
+    ...rest,
+    endpoint: {
+      baseURL:
+        baseURL ??
+        `https://${GoogleVertexShared.host(location)}/v1/projects/${GoogleVertexShared.requireProject(project)}/locations/${location}/publishers/anthropic/models`,
+    },
+    auth: GoogleVertexShared.oauth(input, project),
+  })
+}
+
+export const configure = (input: Config = {}) => {
+  const route = configuredRoute(input)
+  return {
+    id,
+    model: (modelID: string | ModelID) => route.model({ id: modelID }),
+    configure,
+  }
+}
+
+export const provider = {
+  id,
+  configure,
+}
+
+export const model: ProviderPackage.Definition<Settings>["model"] = (modelID, settings) => {
+  if (settings.apiKey !== undefined) throw new Error("Google Vertex Messages does not support API keys")
+  return configure({
+    accessToken: settings.accessToken,
+    baseURL: settings.baseURL,
+    headers: settings.headers === undefined ? undefined : { ...settings.headers },
+    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
+    limits: settings.limits,
+    location: settings.location,
+    project: settings.project,
+    providerOptions: settings.providerOptions,
+  }).model(modelID)
+}

packages/ai/src/providers/google-vertex-responses.ts

@@ -1,10 +1,10 @@
 import type { ProviderPackage } from "../provider-package"
-import { GoogleVertexAnthropic } from "../protocols/google-vertex-anthropic"
+import { OpenAICompatibleResponses } from "../protocols/openai-compatible-responses"
 import type { RouteDefaultsInput } from "../route/client"
 import { ProviderID, type ModelID, type ProviderOptions } from "../schema"
 import { GoogleVertexShared } from "./google-vertex-shared"
 
-export const id = ProviderID.make("google-vertex-anthropic")
+export const id = ProviderID.make("google-vertex")
 
 export type Config = RouteDefaultsInput &
   GoogleVertexShared.OAuthOptions & {
@@ -22,11 +22,16 @@ export interface Settings extends ProviderPackage.Settings {
   readonly providerOptions?: ProviderOptions
 }
 
-export const routes = [GoogleVertexAnthropic.route]
+const route = OpenAICompatibleResponses.route.with({
+  id: "google-vertex-responses",
+  provider: id,
+})
+
+export const routes = [route]
 
 const configuredRoute = (input: Config) => {
   if ("apiKey" in input && input.apiKey !== undefined)
-    throw new Error("Google Vertex Anthropic does not support API keys")
+    throw new Error("Google Vertex Responses does not support API keys")
   const {
     accessToken: _accessToken,
     auth: _auth,
@@ -37,12 +42,12 @@ const configuredRoute = (input: Config) => {
   } = input
   const location = GoogleVertexShared.location(inputLocation, "global")
   const project = GoogleVertexShared.project(inputProject)
-  return GoogleVertexAnthropic.route.with({
+  return route.with({
     ...rest,
     endpoint: {
       baseURL:
         baseURL ??
-        `https://${GoogleVertexShared.host(location)}/v1/projects/${GoogleVertexShared.requireProject(project)}/locations/${location}/publishers/anthropic/models`,
+        `https://aiplatform.googleapis.com/v1/projects/${GoogleVertexShared.requireProject(project)}/locations/${location}/endpoints/openapi`,
     },
     auth: GoogleVertexShared.oauth(input, project),
   })
@@ -63,7 +68,7 @@ export const provider = {
 }
 
 export const model: ProviderPackage.Definition<Settings>["model"] = (modelID, settings) => {
-  if (settings.apiKey !== undefined) throw new Error("Google Vertex Anthropic does not support API keys")
+  if (settings.apiKey !== undefined) throw new Error("Google Vertex Responses does not support API keys")
   return configure({
     accessToken: settings.accessToken,
     baseURL: settings.baseURL,

packages/ai/src/providers/google-vertex.ts

@@ -1,7 +1,9 @@
 import type { ProviderPackage } from "../provider-package"
-import { GoogleVertexGemini } from "../protocols/google-vertex-gemini"
+import { Gemini } from "../protocols/gemini"
 import { Auth } from "../route/auth"
-import type { RouteDefaultsInput } from "../route/client"
+import { Route, type RouteDefaultsInput } from "../route/client"
+import { Endpoint } from "../route/endpoint"
+import { Framing } from "../route/framing"
 import { ProviderID, type ModelID, type ProviderOptions } from "../schema"
 import { GoogleVertexShared } from "./google-vertex-shared"
 
@@ -25,7 +27,20 @@ export type Settings = ProviderPackage.Settings &
     readonly providerOptions?: ProviderOptions
   }
 
-export const routes = [GoogleVertexGemini.route]
+const route = Route.make({
+  id: "google-vertex-gemini",
+  provider: id,
+  providerMetadataKey: "google",
+  protocol: Gemini.protocol,
+  endpoint: Endpoint.path(({ request }) => {
+    const model = String(request.model.id)
+    return `/${model.startsWith("endpoints/") ? model : `models/${model}`}:streamGenerateContent?alt=sse`
+  }),
+  auth: Auth.none,
+  framing: Framing.sse,
+})
+
+export const routes = [route]
 
 const configuredRoute = (input: Config, modelID: string | ModelID) => {
   const {
@@ -48,7 +63,7 @@ const configuredRoute = (input: Config, modelID: string | ModelID) => {
     (apiKey
       ? "https://aiplatform.googleapis.com/v1/publishers/google"
       : `https://${GoogleVertexShared.host(location)}/v1beta1/projects/${GoogleVertexShared.requireProject(project)}/locations/${location}${endpointModel ? "" : "/publishers/google"}`)
-  return GoogleVertexGemini.route.with({
+  return route.with({
     ...rest,
     endpoint: { baseURL: endpoint },
     auth: apiKey === undefined ? GoogleVertexShared.oauth(input, project) : Auth.header("x-goog-api-key", apiKey),

packages/ai/src/providers/google-vertex/anthropic.ts

@@ -1,2 +0,0 @@
-export { model } from "../google-vertex-anthropic"
-export type { Settings } from "../google-vertex-anthropic"

packages/ai/src/providers/google-vertex/chat.ts

@@ -0,0 +1,2 @@
+export { model } from "../google-vertex-chat"
+export type { Settings } from "../google-vertex-chat"

packages/ai/src/providers/google-vertex/gemini.ts

@@ -0,0 +1,2 @@
+export { model } from "../google-vertex"
+export type { Settings } from "../google-vertex"

packages/ai/src/providers/google-vertex/messages.ts

@@ -0,0 +1,2 @@
+export { model } from "../google-vertex-messages"
+export type { Settings } from "../google-vertex-messages"

packages/ai/src/providers/google-vertex/responses.ts

@@ -0,0 +1,2 @@
+export { model } from "../google-vertex-responses"
+export type { Settings } from "../google-vertex-responses"

packages/ai/src/providers/index.ts

@@ -7,7 +7,9 @@ export { CloudflareAIGateway, CloudflareWorkersAI } from "./cloudflare"
 export * as GitHubCopilot from "./github-copilot"
 export * as Google from "./google"
 export * as GoogleVertex from "./google-vertex"
-export * as GoogleVertexAnthropic from "./google-vertex-anthropic"
+export * as GoogleVertexChat from "./google-vertex-chat"
+export * as GoogleVertexMessages from "./google-vertex-messages"
+export * as GoogleVertexResponses from "./google-vertex-responses"
 export * as OpenAI from "./openai"
 export * as OpenAICompatible from "./openai-compatible"
 export * as OpenAICompatibleResponses from "./openai-compatible-responses"

packages/ai/test/auth-options.types.ts

@@ -11,7 +11,9 @@ import * as Cloudflare from "../src/providers/cloudflare"
 import * as GitHubCopilot from "../src/providers/github-copilot"
 import * as Google from "../src/providers/google"
 import * as GoogleVertex from "../src/providers/google-vertex"
-import * as GoogleVertexAnthropic from "../src/providers/google-vertex-anthropic"
+import * as GoogleVertexChat from "../src/providers/google-vertex-chat"
+import * as GoogleVertexMessages from "../src/providers/google-vertex-messages"
+import * as GoogleVertexResponses from "../src/providers/google-vertex-responses"
 import * as OpenAI from "../src/providers/openai"
 import * as OpenAICompatible from "../src/providers/openai-compatible"
 import * as OpenRouter from "../src/providers/openrouter"
@@ -171,20 +173,56 @@ GoogleVertex.configure({ accessToken: "vertex-token", apiKey: "vertex-key", proj
 // @ts-expect-error Vertex Gemini package settings accept only one auth source.
 GoogleVertex.model("gemini-3.5-flash", { accessToken: "vertex-token", apiKey: "vertex-key", project: "project" })
 
-GoogleVertexAnthropic.configure({ accessToken: "vertex-token", project: "project" }).model("claude-sonnet-4-6")
-// @ts-expect-error Vertex Anthropic package settings do not accept API keys.
-GoogleVertexAnthropic.model("claude-sonnet-4-6", { apiKey: "vertex-key", project: "project" })
-GoogleVertexAnthropic.configure({ auth: RuntimeAuth.bearer("vertex-token"), project: "project" }).model(
-  "claude-sonnet-4-6",
+GoogleVertexChat.configure({ accessToken: "vertex-token", project: "project" }).model("deepseek-ai/deepseek-v3.2-maas")
+GoogleVertexChat.configure({ auth: RuntimeAuth.bearer("vertex-token"), project: "project" }).model(
+  "deepseek-ai/deepseek-v3.2-maas",
 )
-GoogleVertexAnthropic.configure({ accessToken: "vertex-token", project: "project" }).model(
-  "claude-sonnet-4-6",
-  // @ts-expect-error Vertex Anthropic model selectors only accept model ids.
+// @ts-expect-error Vertex Chat package settings do not accept API keys.
+GoogleVertexChat.model("deepseek-ai/deepseek-v3.2-maas", { apiKey: "vertex-key", project: "project" })
+GoogleVertexChat.configure({ accessToken: "vertex-token", project: "project" }).model(
+  "deepseek-ai/deepseek-v3.2-maas",
+  // @ts-expect-error Vertex Chat model selectors only accept model ids.
   {},
 )
-GoogleVertexAnthropic.configure({
+GoogleVertexChat.configure({
   accessToken: "vertex-token",
-  // @ts-expect-error Vertex Anthropic config accepts only one auth source.
+  // @ts-expect-error Vertex Chat config accepts only one auth source.
+  auth: RuntimeAuth.bearer("vertex-token"),
+  project: "project",
+})
+
+GoogleVertexResponses.configure({ accessToken: "vertex-token", project: "project" }).model("xai/grok-4.20-reasoning")
+GoogleVertexResponses.configure({ auth: RuntimeAuth.bearer("vertex-token"), project: "project" }).model(
+  "xai/grok-4.20-reasoning",
+)
+// @ts-expect-error Vertex Responses package settings do not accept API keys.
+GoogleVertexResponses.model("xai/grok-4.20-reasoning", { apiKey: "vertex-key", project: "project" })
+GoogleVertexResponses.configure({ accessToken: "vertex-token", project: "project" }).model(
+  "xai/grok-4.20-reasoning",
+  // @ts-expect-error Vertex Responses model selectors only accept model ids.
+  {},
+)
+GoogleVertexResponses.configure({
+  accessToken: "vertex-token",
+  // @ts-expect-error Vertex Responses config accepts only one auth source.
+  auth: RuntimeAuth.bearer("vertex-token"),
+  project: "project",
+})
+
+GoogleVertexMessages.configure({ accessToken: "vertex-token", project: "project" }).model("claude-sonnet-4-6")
+// @ts-expect-error Vertex Messages package settings do not accept API keys.
+GoogleVertexMessages.model("claude-sonnet-4-6", { apiKey: "vertex-key", project: "project" })
+GoogleVertexMessages.configure({ auth: RuntimeAuth.bearer("vertex-token"), project: "project" }).model(
+  "claude-sonnet-4-6",
+)
+GoogleVertexMessages.configure({ accessToken: "vertex-token", project: "project" }).model(
+  "claude-sonnet-4-6",
+  // @ts-expect-error Vertex Messages model selectors only accept model ids.
+  {},
+)
+GoogleVertexMessages.configure({
+  accessToken: "vertex-token",
+  // @ts-expect-error Vertex Messages config accepts only one auth source.
   auth: RuntimeAuth.bearer("vertex-token"),
   project: "project",
 })

packages/ai/test/provider-package.test.ts

@@ -17,12 +17,16 @@ describe("provider package entrypoints", () => {
       import("@opencode-ai/ai/providers/azure/chat"),
       import("@opencode-ai/ai/providers/google"),
       import("@opencode-ai/ai/providers/google-vertex"),
-      import("@opencode-ai/ai/providers/google-vertex/anthropic"),
+      import("@opencode-ai/ai/providers/google-vertex/gemini"),
+      import("@opencode-ai/ai/providers/google-vertex/chat"),
+      import("@opencode-ai/ai/providers/google-vertex/responses"),
+      import("@opencode-ai/ai/providers/google-vertex/messages"),
     ])
 
     for (const module of modules) expect(module.model).toBeFunction()
     expect(modules[0].model).toBe(modules[1].model)
     expect(modules[8].model).toBe(modules[9].model)
+    expect(modules[12].model).toBe(modules[13].model)
   })
 
   test("maps package settings onto the executable model", () => {
@@ -179,20 +183,35 @@ describe("provider package entrypoints", () => {
 
   test("selects Vertex entrypoints with the same model contract", async () => {
     const GoogleVertex = await import("@opencode-ai/ai/providers/google-vertex")
-    const GoogleVertexAnthropic = await import("@opencode-ai/ai/providers/google-vertex/anthropic")
+    const GoogleVertexGemini = await import("@opencode-ai/ai/providers/google-vertex/gemini")
+    const GoogleVertexChat = await import("@opencode-ai/ai/providers/google-vertex/chat")
+    const GoogleVertexResponses = await import("@opencode-ai/ai/providers/google-vertex/responses")
+    const GoogleVertexMessages = await import("@opencode-ai/ai/providers/google-vertex/messages")
     const gemini = GoogleVertex.model("gemini-3.5-flash", {
       apiKey: "fixture",
       headers: { "x-application": "opencode" },
       body: { safetySettings: [] },
       limits: { context: 1_000_000, output: 65_536 },
     })
-    const anthropic = GoogleVertexAnthropic.model("claude-sonnet-4-6", {
+    const messages = GoogleVertexMessages.model("claude-sonnet-4-6", {
+      accessToken: "fixture",
+      location: "global",
+      project: "vertex-project",
+    })
+    const chat = GoogleVertexChat.model("deepseek-ai/deepseek-v3.2-maas", {
+      accessToken: "fixture",
+      location: "global",
+      project: "vertex-project",
+    })
+    const responses = GoogleVertexResponses.model("xai/grok-4.20-reasoning", {
       accessToken: "fixture",
       location: "global",
       project: "vertex-project",
     })
 
+    expect(GoogleVertexGemini.model).toBe(GoogleVertex.model)
     expect(gemini.route.id).toBe("google-vertex-gemini")
+    expect(gemini.route.protocol).toBe("gemini")
     expect(gemini.route.endpoint.baseURL).toBe("https://aiplatform.googleapis.com/v1/publishers/google")
     expect(gemini.route.defaults.headers).toEqual({ "x-application": "opencode" })
     expect(gemini.route.defaults.http?.body).toEqual({ safetySettings: [] })
@@ -204,15 +223,31 @@ describe("provider package entrypoints", () => {
         project: "vertex-project",
       }).route.endpoint.baseURL,
     ).toBe("https://aiplatform.eu.rep.googleapis.com/v1beta1/projects/vertex-project/locations/eu/publishers/google")
-    expect(anthropic.route.id).toBe("google-vertex-anthropic")
-    expect(anthropic.route.endpoint.baseURL).toBe(
+    expect(messages.route.id).toBe("google-vertex-messages")
+    expect(messages.route.protocol).toBe("anthropic-messages")
+    expect(messages.route.endpoint.baseURL).toBe(
       "https://aiplatform.googleapis.com/v1/projects/vertex-project/locations/global/publishers/anthropic/models",
     )
+    expect(chat.route.id).toBe("google-vertex-chat")
+    expect(chat.route.protocol).toBe("openai-chat")
+    expect(chat.route.endpoint).toMatchObject({
+      baseURL: "https://aiplatform.googleapis.com/v1/projects/vertex-project/locations/global/endpoints/openapi",
+      path: "/chat/completions",
+    })
+    expect(responses.route.id).toBe("google-vertex-responses")
+    expect(responses.route.protocol).toBe("openai-responses")
+    expect(responses.route.endpoint).toMatchObject({
+      baseURL: "https://aiplatform.googleapis.com/v1/projects/vertex-project/locations/global/endpoints/openapi",
+      path: "/responses",
+    })
+    expect(responses.route.defaults.providerOptions).toEqual({ openai: { store: false } })
   })
 
   test("rejects conflicting Vertex auth settings at runtime", async () => {
     const GoogleVertex = await import("@opencode-ai/ai/providers/google-vertex")
-    const GoogleVertexAnthropic = await import("@opencode-ai/ai/providers/google-vertex/anthropic")
+    const GoogleVertexChat = await import("@opencode-ai/ai/providers/google-vertex/chat")
+    const GoogleVertexMessages = await import("@opencode-ai/ai/providers/google-vertex/messages")
+    const GoogleVertexResponses = await import("@opencode-ai/ai/providers/google-vertex/responses")
     const Providers = await import("@opencode-ai/ai/providers")
     expect(() =>
       Reflect.apply(GoogleVertex.model, undefined, [
@@ -225,15 +260,37 @@ describe("provider package entrypoints", () => {
     ])
     expect(() => configured.model("gemini-3.5-flash")).toThrow("Google Vertex accessToken cannot be combined with auth")
     expect(() =>
-      Reflect.apply(GoogleVertexAnthropic.model, undefined, [
+      Reflect.apply(GoogleVertexMessages.model, undefined, [
         "claude-sonnet-4-6",
         { apiKey: "fixture", project: "vertex-project" },
       ]),
-    ).toThrow("Google Vertex Anthropic does not support API keys")
+    ).toThrow("Google Vertex Messages does not support API keys")
     expect(() =>
-      Reflect.apply(Providers.GoogleVertexAnthropic.configure, undefined, [
+      Reflect.apply(Providers.GoogleVertexMessages.configure, undefined, [
         { apiKey: "fixture", project: "vertex-project" },
       ]),
-    ).toThrow("Google Vertex Anthropic does not support API keys")
+    ).toThrow("Google Vertex Messages does not support API keys")
+    expect(() =>
+      Reflect.apply(GoogleVertexChat.model, undefined, [
+        "deepseek-ai/deepseek-v3.2-maas",
+        { apiKey: "fixture", project: "vertex-project" },
+      ]),
+    ).toThrow("Google Vertex Chat does not support API keys")
+    expect(() =>
+      Reflect.apply(Providers.GoogleVertexChat.configure, undefined, [
+        { apiKey: "fixture", project: "vertex-project" },
+      ]),
+    ).toThrow("Google Vertex Chat does not support API keys")
+    expect(() =>
+      Reflect.apply(GoogleVertexResponses.model, undefined, [
+        "xai/grok-4.20-reasoning",
+        { apiKey: "fixture", project: "vertex-project" },
+      ]),
+    ).toThrow("Google Vertex Responses does not support API keys")
+    expect(() =>
+      Reflect.apply(Providers.GoogleVertexResponses.configure, undefined, [
+        { apiKey: "fixture", project: "vertex-project" },
+      ]),
+    ).toThrow("Google Vertex Responses does not support API keys")
   })
 })

packages/ai/test/provider/google-vertex.test.ts

@@ -2,10 +2,11 @@ import { describe, expect } from "bun:test"
 import { Effect } from "effect"
 import { HttpClientRequest } from "effect/unstable/http"
 import { LLM } from "../../src"
-import { GoogleVertex, GoogleVertexAnthropic } from "../../src/providers"
+import { GoogleVertex, GoogleVertexChat, GoogleVertexMessages, GoogleVertexResponses } from "../../src/providers"
 import { LLMClient } from "../../src/route"
 import { it } from "../lib/effect"
 import { dynamicResponse } from "../lib/http"
+import { deltaChunk, finishChunk } from "../lib/openai-chunks"
 import { sseEvents } from "../lib/sse"
 
 describe("Google Vertex providers", () => {
@@ -56,7 +57,7 @@ describe("Google Vertex providers", () => {
     Effect.gen(function* () {
       const response = yield* LLMClient.generate(
         LLM.request({
-          model: GoogleVertexAnthropic.configure({
+          model: GoogleVertexMessages.configure({
             accessToken: "vertex-token",
             location: "eu",
             project: "vertex-project",
@@ -99,11 +100,91 @@ describe("Google Vertex providers", () => {
     }),
   )
 
-  it.effect("protects the Vertex Anthropic API version from body overlays", () =>
+  it.effect("sends MaaS requests through Vertex Chat Completions", () =>
+    Effect.gen(function* () {
+      const response = yield* LLMClient.generate(
+        LLM.request({
+          model: GoogleVertexChat.configure({
+            accessToken: "vertex-token",
+            location: "global",
+            project: "vertex-project",
+          }).model("deepseek-ai/deepseek-v3.2-maas"),
+          prompt: "Say hello.",
+        }),
+      ).pipe(
+        Effect.provide(
+          dynamicResponse((input) =>
+            Effect.gen(function* () {
+              const request = yield* HttpClientRequest.toWeb(input.request).pipe(Effect.orDie)
+              expect(request.url).toBe(
+                "https://aiplatform.googleapis.com/v1/projects/vertex-project/locations/global/endpoints/openapi/chat/completions",
+              )
+              expect(request.headers.get("authorization")).toBe("Bearer vertex-token")
+              expect(yield* Effect.promise(() => request.json())).toMatchObject({
+                model: "deepseek-ai/deepseek-v3.2-maas",
+                messages: [{ role: "user", content: "Say hello." }],
+                stream: true,
+                stream_options: { include_usage: true },
+              })
+              return input.respond(sseEvents(deltaChunk({ content: "Hello." }), finishChunk("stop")), {
+                headers: { "content-type": "text/event-stream" },
+              })
+            }),
+          ),
+        ),
+      )
+
+      expect(response.text).toBe("Hello.")
+    }),
+  )
+
+  it.effect("sends Grok requests through Vertex Responses", () =>
+    Effect.gen(function* () {
+      const response = yield* LLMClient.generate(
+        LLM.request({
+          model: GoogleVertexResponses.configure({
+            accessToken: "vertex-token",
+            location: "global",
+            project: "vertex-project",
+          }).model("xai/grok-4.20-reasoning"),
+          prompt: "Say hello.",
+        }),
+      ).pipe(
+        Effect.provide(
+          dynamicResponse((input) =>
+            Effect.gen(function* () {
+              const request = yield* HttpClientRequest.toWeb(input.request).pipe(Effect.orDie)
+              expect(request.url).toBe(
+                "https://aiplatform.googleapis.com/v1/projects/vertex-project/locations/global/endpoints/openapi/responses",
+              )
+              expect(request.headers.get("authorization")).toBe("Bearer vertex-token")
+              expect(yield* Effect.promise(() => request.json())).toMatchObject({
+                model: "xai/grok-4.20-reasoning",
+                input: [{ role: "user", content: [{ type: "input_text", text: "Say hello." }] }],
+                store: false,
+                stream: true,
+              })
+              return input.respond(
+                sseEvents(
+                  { type: "response.output_text.delta", item_id: "msg_1", delta: "Hello." },
+                  { type: "response.completed", response: { id: "resp_1" } },
+                ),
+                { headers: { "content-type": "text/event-stream" } },
+              )
+            }),
+          ),
+        ),
+      )
+
+      expect(response.text).toBe("Hello.")
+    }),
+  )
+
+  it.effect("protects the Vertex Messages API version from body overlays", () =>
     Effect.gen(function* () {
       const error = yield* LLMClient.prepare(
         LLM.request({
-          model: GoogleVertexAnthropic.configure({
+          model: GoogleVertexMessages.configure({
             accessToken: "vertex-token",
             http: { body: { anthropic_version: "wrong" } },
             project: "vertex-project",

packages/cli/src/commands/commands.ts

@@ -66,6 +66,17 @@ export const Commands = Spec.make(typeof OPENCODE_CLI_NAME === "string" ? OPENCO
         }),
       ],
     }),
+    Spec.make("auth", {
+      description: "Manage authentication",
+      commands: [
+        Spec.make("connect", {
+          description: "Connect to a wellknown authentication provider",
+          params: {
+            url: Argument.string("url").pipe(Argument.withDescription("Wellknown provider URL")),
+          },
+        }),
+      ],
+    }),
     Spec.make("mcp", {
       description: "Manage MCP (Model Context Protocol) servers",
       commands: [
@@ -178,10 +189,7 @@ export const Commands = Spec.make(typeof OPENCODE_CLI_NAME === "string" ? OPENCO
           Flag.atMost(100),
         ),
         title: Flag.string("title").pipe(Flag.withDescription("Session title"), Flag.optional),
-        thinking: Flag.boolean("thinking").pipe(
-          Flag.withDescription("Show thinking blocks"),
-          Flag.withDefault(false),
-        ),
+        thinking: Flag.boolean("thinking").pipe(Flag.withDescription("Show thinking blocks"), Flag.withDefault(false)),
         auto: Flag.boolean("auto").pipe(
           Flag.withDescription("Auto-approve permissions that are not explicitly denied"),
           Flag.withDefault(false),

packages/cli/src/commands/handlers/auth/connect.ts

@@ -0,0 +1,52 @@
+import { EOL } from "node:os"
+import { Effect } from "effect"
+import { Service } from "@opencode-ai/client/effect/service"
+import { OpenCode, type IntegrationCommandStatusOutput, type OpenCodeClient } from "@opencode-ai/client/promise"
+import { Commands } from "../../commands"
+import { Runtime } from "../../../framework/runtime"
+import { ServiceConfig } from "../../../services/service-config"
+
+const location = { directory: process.cwd() }
+
+export default Runtime.handler(
+  Commands.commands.auth.commands.connect,
+  Effect.fn("cli.auth.connect")(function* (input) {
+    process.stdout.write("Connecting..." + EOL + EOL)
+    const endpoint = yield* Service.ensure(yield* ServiceConfig.options())
+    const client = OpenCode.make({ baseUrl: endpoint.url, headers: Service.headers(endpoint) })
+    yield* request(() => client.integration.wellknown.add({ url: input.url, location }))
+    const integrationID = input.url.replace(/\/+$/, "")
+    const started = yield* request(() =>
+      client.integration.command.connect({ integrationID, methodID: "login", location }),
+    )
+    yield* Effect.addFinalizer(() =>
+      request(() =>
+        client.integration.command.cancel({ integrationID, attemptID: started.data.attemptID, location }),
+      ).pipe(Effect.ignore),
+    )
+
+    const status = yield* wait(client, integrationID, started.data.attemptID)
+    if (status.status === "failed") return yield* Effect.fail(new Error(status.message))
+    if (status.status === "expired") return yield* Effect.fail(new Error("Authentication expired"))
+    process.stdout.write("Connected" + EOL)
+  }),
+)
+
+const wait = (
+  client: OpenCodeClient,
+  integrationID: string,
+  attemptID: string,
+  shown = false,
+): Effect.Effect<Exclude<IntegrationCommandStatusOutput["data"], { status: "pending" }>, unknown> =>
+  Effect.gen(function* () {
+    const response = yield* request(() => client.integration.command.status({ integrationID, attemptID, location }))
+    if (response.data.status !== "pending") return response.data
+    const output = response.data.message?.trim()
+    if (!shown && output) process.stdout.write(output + EOL + EOL)
+    yield* Effect.sleep(500)
+    return yield* wait(client, integrationID, attemptID, shown || !!output)
+  })
+
+function request<A>(task: () => Promise<A>) {
+  return Effect.tryPromise({ try: task, catch: (cause) => cause })
+}

packages/cli/src/index.ts

@@ -17,6 +17,9 @@ import { Npm } from "@opencode-ai/core/npm"
 const Handlers = Runtime.handlers(Commands, {
   $: () => import("./commands/handlers/default"),
   api: () => import("./commands/handlers/api"),
+  auth: {
+    connect: () => import("./commands/handlers/auth/connect"),
+  },
   debug: {
     agents: () => import("./commands/handlers/debug/agents"),
   },

packages/client/src/effect/api/api.ts

@@ -407,102 +407,113 @@ export type Endpoint10_1Input = {
 export type Endpoint10_1Output = EffectValue<ReturnType<RawClient["server.integration"]["integration.get"]>>
 export type IntegrationGetOperation<E = never> = (input: Endpoint10_1Input) => Effect.Effect<Endpoint10_1Output, E>
 
-type Endpoint10_2Request = Parameters<RawClient["server.integration"]["integration.connect.key"]>[0]
+type Endpoint10_2Request = Parameters<RawClient["server.integration"]["integration.wellknown.add"]>[0]
 export type Endpoint10_2Input = {
-  readonly integrationID: Endpoint10_2Request["params"]["integrationID"]
   readonly location?: Endpoint10_2Request["query"]["location"]
-  readonly key: Endpoint10_2Request["payload"]["key"]
-  readonly label?: Endpoint10_2Request["payload"]["label"]
+  readonly url: Endpoint10_2Request["payload"]["url"]
 }
-export type Endpoint10_2Output = EffectValue<ReturnType<RawClient["server.integration"]["integration.connect.key"]>>
-export type IntegrationConnectKeyOperation<E = never> = (
+export type Endpoint10_2Output = EffectValue<ReturnType<RawClient["server.integration"]["integration.wellknown.add"]>>
+export type IntegrationWellknownAddOperation<E = never> = (
   input: Endpoint10_2Input,
 ) => Effect.Effect<Endpoint10_2Output, E>
 
-type Endpoint10_3Request = Parameters<RawClient["server.integration"]["integration.oauth.connect"]>[0]
+type Endpoint10_3Request = Parameters<RawClient["server.integration"]["integration.connect.key"]>[0]
 export type Endpoint10_3Input = {
   readonly integrationID: Endpoint10_3Request["params"]["integrationID"]
   readonly location?: Endpoint10_3Request["query"]["location"]
-  readonly methodID: Endpoint10_3Request["payload"]["methodID"]
-  readonly inputs: Endpoint10_3Request["payload"]["inputs"]
+  readonly key: Endpoint10_3Request["payload"]["key"]
   readonly label?: Endpoint10_3Request["payload"]["label"]
 }
-export type Endpoint10_3Output = EffectValue<ReturnType<RawClient["server.integration"]["integration.oauth.connect"]>>
-export type IntegrationOauthConnectOperation<E = never> = (
+export type Endpoint10_3Output = EffectValue<ReturnType<RawClient["server.integration"]["integration.connect.key"]>>
+export type IntegrationConnectKeyOperation<E = never> = (
   input: Endpoint10_3Input,
 ) => Effect.Effect<Endpoint10_3Output, E>
 
-type Endpoint10_4Request = Parameters<RawClient["server.integration"]["integration.oauth.status"]>[0]
+type Endpoint10_4Request = Parameters<RawClient["server.integration"]["integration.oauth.connect"]>[0]
 export type Endpoint10_4Input = {
   readonly integrationID: Endpoint10_4Request["params"]["integrationID"]
-  readonly attemptID: Endpoint10_4Request["params"]["attemptID"]
   readonly location?: Endpoint10_4Request["query"]["location"]
+  readonly methodID: Endpoint10_4Request["payload"]["methodID"]
+  readonly inputs: Endpoint10_4Request["payload"]["inputs"]
+  readonly label?: Endpoint10_4Request["payload"]["label"]
 }
-export type Endpoint10_4Output = EffectValue<ReturnType<RawClient["server.integration"]["integration.oauth.status"]>>
-export type IntegrationOauthStatusOperation<E = never> = (
+export type Endpoint10_4Output = EffectValue<ReturnType<RawClient["server.integration"]["integration.oauth.connect"]>>
+export type IntegrationOauthConnectOperation<E = never> = (
   input: Endpoint10_4Input,
 ) => Effect.Effect<Endpoint10_4Output, E>
 
-type Endpoint10_5Request = Parameters<RawClient["server.integration"]["integration.oauth.complete"]>[0]
+type Endpoint10_5Request = Parameters<RawClient["server.integration"]["integration.oauth.status"]>[0]
 export type Endpoint10_5Input = {
   readonly integrationID: Endpoint10_5Request["params"]["integrationID"]
   readonly attemptID: Endpoint10_5Request["params"]["attemptID"]
   readonly location?: Endpoint10_5Request["query"]["location"]
-  readonly code?: Endpoint10_5Request["payload"]["code"]
 }
-export type Endpoint10_5Output = EffectValue<ReturnType<RawClient["server.integration"]["integration.oauth.complete"]>>
-export type IntegrationOauthCompleteOperation<E = never> = (
+export type Endpoint10_5Output = EffectValue<ReturnType<RawClient["server.integration"]["integration.oauth.status"]>>
+export type IntegrationOauthStatusOperation<E = never> = (
   input: Endpoint10_5Input,
 ) => Effect.Effect<Endpoint10_5Output, E>
 
-type Endpoint10_6Request = Parameters<RawClient["server.integration"]["integration.oauth.cancel"]>[0]
+type Endpoint10_6Request = Parameters<RawClient["server.integration"]["integration.oauth.complete"]>[0]
 export type Endpoint10_6Input = {
   readonly integrationID: Endpoint10_6Request["params"]["integrationID"]
   readonly attemptID: Endpoint10_6Request["params"]["attemptID"]
   readonly location?: Endpoint10_6Request["query"]["location"]
+  readonly code?: Endpoint10_6Request["payload"]["code"]
 }
-export type Endpoint10_6Output = EffectValue<ReturnType<RawClient["server.integration"]["integration.oauth.cancel"]>>
-export type IntegrationOauthCancelOperation<E = never> = (
+export type Endpoint10_6Output = EffectValue<ReturnType<RawClient["server.integration"]["integration.oauth.complete"]>>
+export type IntegrationOauthCompleteOperation<E = never> = (
   input: Endpoint10_6Input,
 ) => Effect.Effect<Endpoint10_6Output, E>
 
-type Endpoint10_7Request = Parameters<RawClient["server.integration"]["integration.command.connect"]>[0]
+type Endpoint10_7Request = Parameters<RawClient["server.integration"]["integration.oauth.cancel"]>[0]
 export type Endpoint10_7Input = {
   readonly integrationID: Endpoint10_7Request["params"]["integrationID"]
+  readonly attemptID: Endpoint10_7Request["params"]["attemptID"]
   readonly location?: Endpoint10_7Request["query"]["location"]
-  readonly methodID: Endpoint10_7Request["payload"]["methodID"]
-  readonly label?: Endpoint10_7Request["payload"]["label"]
 }
-export type Endpoint10_7Output = EffectValue<ReturnType<RawClient["server.integration"]["integration.command.connect"]>>
-export type IntegrationCommandConnectOperation<E = never> = (
+export type Endpoint10_7Output = EffectValue<ReturnType<RawClient["server.integration"]["integration.oauth.cancel"]>>
+export type IntegrationOauthCancelOperation<E = never> = (
   input: Endpoint10_7Input,
 ) => Effect.Effect<Endpoint10_7Output, E>
 
-type Endpoint10_8Request = Parameters<RawClient["server.integration"]["integration.command.status"]>[0]
+type Endpoint10_8Request = Parameters<RawClient["server.integration"]["integration.command.connect"]>[0]
 export type Endpoint10_8Input = {
   readonly integrationID: Endpoint10_8Request["params"]["integrationID"]
-  readonly attemptID: Endpoint10_8Request["params"]["attemptID"]
   readonly location?: Endpoint10_8Request["query"]["location"]
+  readonly methodID: Endpoint10_8Request["payload"]["methodID"]
+  readonly label?: Endpoint10_8Request["payload"]["label"]
 }
-export type Endpoint10_8Output = EffectValue<ReturnType<RawClient["server.integration"]["integration.command.status"]>>
-export type IntegrationCommandStatusOperation<E = never> = (
+export type Endpoint10_8Output = EffectValue<ReturnType<RawClient["server.integration"]["integration.command.connect"]>>
+export type IntegrationCommandConnectOperation<E = never> = (
   input: Endpoint10_8Input,
 ) => Effect.Effect<Endpoint10_8Output, E>
 
-type Endpoint10_9Request = Parameters<RawClient["server.integration"]["integration.command.cancel"]>[0]
+type Endpoint10_9Request = Parameters<RawClient["server.integration"]["integration.command.status"]>[0]
 export type Endpoint10_9Input = {
   readonly integrationID: Endpoint10_9Request["params"]["integrationID"]
   readonly attemptID: Endpoint10_9Request["params"]["attemptID"]
   readonly location?: Endpoint10_9Request["query"]["location"]
 }
-export type Endpoint10_9Output = EffectValue<ReturnType<RawClient["server.integration"]["integration.command.cancel"]>>
-export type IntegrationCommandCancelOperation<E = never> = (
+export type Endpoint10_9Output = EffectValue<ReturnType<RawClient["server.integration"]["integration.command.status"]>>
+export type IntegrationCommandStatusOperation<E = never> = (
   input: Endpoint10_9Input,
 ) => Effect.Effect<Endpoint10_9Output, E>
 
+type Endpoint10_10Request = Parameters<RawClient["server.integration"]["integration.command.cancel"]>[0]
+export type Endpoint10_10Input = {
+  readonly integrationID: Endpoint10_10Request["params"]["integrationID"]
+  readonly attemptID: Endpoint10_10Request["params"]["attemptID"]
+  readonly location?: Endpoint10_10Request["query"]["location"]
+}
+export type Endpoint10_10Output = EffectValue<ReturnType<RawClient["server.integration"]["integration.command.cancel"]>>
+export type IntegrationCommandCancelOperation<E = never> = (
+  input: Endpoint10_10Input,
+) => Effect.Effect<Endpoint10_10Output, E>
+
 export interface IntegrationApi<E = never> {
   readonly list: IntegrationListOperation<E>
   readonly get: IntegrationGetOperation<E>
+  readonly wellknown: { readonly add: IntegrationWellknownAddOperation<E> }
   readonly connect: { readonly key: IntegrationConnectKeyOperation<E> }
   readonly oauth: {
     readonly connect: IntegrationOauthConnectOperation<E>

packages/client/src/effect/generated/client.ts

@@ -504,106 +504,116 @@ const Endpoint10_1 = (raw: RawClient["server.integration"]) => (input: Endpoint1
     query: { location: input["location"] },
   }).pipe(Effect.mapError(mapClientError))
 
-type Endpoint10_2Request = Parameters<RawClient["server.integration"]["integration.connect.key"]>[0]
+type Endpoint10_2Request = Parameters<RawClient["server.integration"]["integration.wellknown.add"]>[0]
 type Endpoint10_2Input = {
-  readonly integrationID: Endpoint10_2Request["params"]["integrationID"]
   readonly location?: Endpoint10_2Request["query"]["location"]
-  readonly key: Endpoint10_2Request["payload"]["key"]
-  readonly label?: Endpoint10_2Request["payload"]["label"]
+  readonly url: Endpoint10_2Request["payload"]["url"]
 }
 const Endpoint10_2 = (raw: RawClient["server.integration"]) => (input: Endpoint10_2Input) =>
+  raw["integration.wellknown.add"]({ query: { location: input["location"] }, payload: { url: input["url"] } }).pipe(
+    Effect.mapError(mapClientError),
+  )
+
+type Endpoint10_3Request = Parameters<RawClient["server.integration"]["integration.connect.key"]>[0]
+type Endpoint10_3Input = {
+  readonly integrationID: Endpoint10_3Request["params"]["integrationID"]
+  readonly location?: Endpoint10_3Request["query"]["location"]
+  readonly key: Endpoint10_3Request["payload"]["key"]
+  readonly label?: Endpoint10_3Request["payload"]["label"]
+}
+const Endpoint10_3 = (raw: RawClient["server.integration"]) => (input: Endpoint10_3Input) =>
   raw["integration.connect.key"]({
     params: { integrationID: input["integrationID"] },
     query: { location: input["location"] },
     payload: { key: input["key"], label: input["label"] },
   }).pipe(Effect.mapError(mapClientError))
 
-type Endpoint10_3Request = Parameters<RawClient["server.integration"]["integration.oauth.connect"]>[0]
-type Endpoint10_3Input = {
-  readonly integrationID: Endpoint10_3Request["params"]["integrationID"]
-  readonly location?: Endpoint10_3Request["query"]["location"]
-  readonly methodID: Endpoint10_3Request["payload"]["methodID"]
-  readonly inputs: Endpoint10_3Request["payload"]["inputs"]
-  readonly label?: Endpoint10_3Request["payload"]["label"]
+type Endpoint10_4Request = Parameters<RawClient["server.integration"]["integration.oauth.connect"]>[0]
+type Endpoint10_4Input = {
+  readonly integrationID: Endpoint10_4Request["params"]["integrationID"]
+  readonly location?: Endpoint10_4Request["query"]["location"]
+  readonly methodID: Endpoint10_4Request["payload"]["methodID"]
+  readonly inputs: Endpoint10_4Request["payload"]["inputs"]
+  readonly label?: Endpoint10_4Request["payload"]["label"]
 }
-const Endpoint10_3 = (raw: RawClient["server.integration"]) => (input: Endpoint10_3Input) =>
+const Endpoint10_4 = (raw: RawClient["server.integration"]) => (input: Endpoint10_4Input) =>
   raw["integration.oauth.connect"]({
     params: { integrationID: input["integrationID"] },
     query: { location: input["location"] },
     payload: { methodID: input["methodID"], inputs: input["inputs"], label: input["label"] },
   }).pipe(Effect.mapError(mapClientError))
 
-type Endpoint10_4Request = Parameters<RawClient["server.integration"]["integration.oauth.status"]>[0]
-type Endpoint10_4Input = {
-  readonly integrationID: Endpoint10_4Request["params"]["integrationID"]
-  readonly attemptID: Endpoint10_4Request["params"]["attemptID"]
-  readonly location?: Endpoint10_4Request["query"]["location"]
+type Endpoint10_5Request = Parameters<RawClient["server.integration"]["integration.oauth.status"]>[0]
+type Endpoint10_5Input = {
+  readonly integrationID: Endpoint10_5Request["params"]["integrationID"]
+  readonly attemptID: Endpoint10_5Request["params"]["attemptID"]
+  readonly location?: Endpoint10_5Request["query"]["location"]
 }
-const Endpoint10_4 = (raw: RawClient["server.integration"]) => (input: Endpoint10_4Input) =>
+const Endpoint10_5 = (raw: RawClient["server.integration"]) => (input: Endpoint10_5Input) =>
   raw["integration.oauth.status"]({
     params: { integrationID: input["integrationID"], attemptID: input["attemptID"] },
     query: { location: input["location"] },
   }).pipe(Effect.mapError(mapClientError))
 
-type Endpoint10_5Request = Parameters<RawClient["server.integration"]["integration.oauth.complete"]>[0]
-type Endpoint10_5Input = {
-  readonly integrationID: Endpoint10_5Request["params"]["integrationID"]
-  readonly attemptID: Endpoint10_5Request["params"]["attemptID"]
-  readonly location?: Endpoint10_5Request["query"]["location"]
-  readonly code?: Endpoint10_5Request["payload"]["code"]
+type Endpoint10_6Request = Parameters<RawClient["server.integration"]["integration.oauth.complete"]>[0]
+type Endpoint10_6Input = {
+  readonly integrationID: Endpoint10_6Request["params"]["integrationID"]
+  readonly attemptID: Endpoint10_6Request["params"]["attemptID"]
+  readonly location?: Endpoint10_6Request["query"]["location"]
+  readonly code?: Endpoint10_6Request["payload"]["code"]
 }
-const Endpoint10_5 = (raw: RawClient["server.integration"]) => (input: Endpoint10_5Input) =>
+const Endpoint10_6 = (raw: RawClient["server.integration"]) => (input: Endpoint10_6Input) =>
   raw["integration.oauth.complete"]({
     params: { integrationID: input["integrationID"], attemptID: input["attemptID"] },
     query: { location: input["location"] },
     payload: { code: input["code"] },
   }).pipe(Effect.mapError(mapClientError))
 
-type Endpoint10_6Request = Parameters<RawClient["server.integration"]["integration.oauth.cancel"]>[0]
-type Endpoint10_6Input = {
-  readonly integrationID: Endpoint10_6Request["params"]["integrationID"]
-  readonly attemptID: Endpoint10_6Request["params"]["attemptID"]
-  readonly location?: Endpoint10_6Request["query"]["location"]
+type Endpoint10_7Request = Parameters<RawClient["server.integration"]["integration.oauth.cancel"]>[0]
+type Endpoint10_7Input = {
+  readonly integrationID: Endpoint10_7Request["params"]["integrationID"]
+  readonly attemptID: Endpoint10_7Request["params"]["attemptID"]
+  readonly location?: Endpoint10_7Request["query"]["location"]
 }
-const Endpoint10_6 = (raw: RawClient["server.integration"]) => (input: Endpoint10_6Input) =>
+const Endpoint10_7 = (raw: RawClient["server.integration"]) => (input: Endpoint10_7Input) =>
   raw["integration.oauth.cancel"]({
     params: { integrationID: input["integrationID"], attemptID: input["attemptID"] },
     query: { location: input["location"] },
   }).pipe(Effect.mapError(mapClientError))
 
-type Endpoint10_7Request = Parameters<RawClient["server.integration"]["integration.command.connect"]>[0]
-type Endpoint10_7Input = {
-  readonly integrationID: Endpoint10_7Request["params"]["integrationID"]
-  readonly location?: Endpoint10_7Request["query"]["location"]
-  readonly methodID: Endpoint10_7Request["payload"]["methodID"]
-  readonly label?: Endpoint10_7Request["payload"]["label"]
+type Endpoint10_8Request = Parameters<RawClient["server.integration"]["integration.command.connect"]>[0]
+type Endpoint10_8Input = {
+  readonly integrationID: Endpoint10_8Request["params"]["integrationID"]
+  readonly location?: Endpoint10_8Request["query"]["location"]
+  readonly methodID: Endpoint10_8Request["payload"]["methodID"]
+  readonly label?: Endpoint10_8Request["payload"]["label"]
 }
-const Endpoint10_7 = (raw: RawClient["server.integration"]) => (input: Endpoint10_7Input) =>
+const Endpoint10_8 = (raw: RawClient["server.integration"]) => (input: Endpoint10_8Input) =>
   raw["integration.command.connect"]({
     params: { integrationID: input["integrationID"] },
     query: { location: input["location"] },
     payload: { methodID: input["methodID"], label: input["label"] },
   }).pipe(Effect.mapError(mapClientError))
 
-type Endpoint10_8Request = Parameters<RawClient["server.integration"]["integration.command.status"]>[0]
-type Endpoint10_8Input = {
-  readonly integrationID: Endpoint10_8Request["params"]["integrationID"]
-  readonly attemptID: Endpoint10_8Request["params"]["attemptID"]
-  readonly location?: Endpoint10_8Request["query"]["location"]
-}
-const Endpoint10_8 = (raw: RawClient["server.integration"]) => (input: Endpoint10_8Input) =>
-  raw["integration.command.status"]({
-    params: { integrationID: input["integrationID"], attemptID: input["attemptID"] },
-    query: { location: input["location"] },
-  }).pipe(Effect.mapError(mapClientError))
-
-type Endpoint10_9Request = Parameters<RawClient["server.integration"]["integration.command.cancel"]>[0]
+type Endpoint10_9Request = Parameters<RawClient["server.integration"]["integration.command.status"]>[0]
 type Endpoint10_9Input = {
   readonly integrationID: Endpoint10_9Request["params"]["integrationID"]
   readonly attemptID: Endpoint10_9Request["params"]["attemptID"]
   readonly location?: Endpoint10_9Request["query"]["location"]
 }
 const Endpoint10_9 = (raw: RawClient["server.integration"]) => (input: Endpoint10_9Input) =>
+  raw["integration.command.status"]({
+    params: { integrationID: input["integrationID"], attemptID: input["attemptID"] },
+    query: { location: input["location"] },
+  }).pipe(Effect.mapError(mapClientError))
+
+type Endpoint10_10Request = Parameters<RawClient["server.integration"]["integration.command.cancel"]>[0]
+type Endpoint10_10Input = {
+  readonly integrationID: Endpoint10_10Request["params"]["integrationID"]
+  readonly attemptID: Endpoint10_10Request["params"]["attemptID"]
+  readonly location?: Endpoint10_10Request["query"]["location"]
+}
+const Endpoint10_10 = (raw: RawClient["server.integration"]) => (input: Endpoint10_10Input) =>
   raw["integration.command.cancel"]({
     params: { integrationID: input["integrationID"], attemptID: input["attemptID"] },
     query: { location: input["location"] },
@@ -612,14 +622,15 @@ const Endpoint10_9 = (raw: RawClient["server.integration"]) => (input: Endpoint1
 const adaptGroup10 = (raw: RawClient["server.integration"]) => ({
   list: Endpoint10_0(raw),
   get: Endpoint10_1(raw),
-  connect: { key: Endpoint10_2(raw) },
+  wellknown: { add: Endpoint10_2(raw) },
+  connect: { key: Endpoint10_3(raw) },
   oauth: {
-    connect: Endpoint10_3(raw),
-    status: Endpoint10_4(raw),
-    complete: Endpoint10_5(raw),
-    cancel: Endpoint10_6(raw),
+    connect: Endpoint10_4(raw),
+    status: Endpoint10_5(raw),
+    complete: Endpoint10_6(raw),
+    cancel: Endpoint10_7(raw),
   },
-  command: { connect: Endpoint10_7(raw), status: Endpoint10_8(raw), cancel: Endpoint10_9(raw) },
+  command: { connect: Endpoint10_8(raw), status: Endpoint10_9(raw), cancel: Endpoint10_10(raw) },
 })
 
 type Endpoint11_0Request = Parameters<RawClient["server.mcp"]["mcp.list"]>[0]

packages/client/src/promise/generated/client.ts

@@ -82,6 +82,8 @@ import type {
   IntegrationListOutput,
   IntegrationGetInput,
   IntegrationGetOutput,
+  IntegrationWellknownAddInput,
+  IntegrationWellknownAddOutput,
   IntegrationConnectKeyInput,
   IntegrationConnectKeyOutput,
   IntegrationOauthConnectInput,
@@ -893,6 +895,21 @@ export function make(options: ClientOptions) {
           },
           requestOptions,
         ),
+      wellknown: {
+        add: (input: IntegrationWellknownAddInput, requestOptions?: RequestOptions) =>
+          request<IntegrationWellknownAddOutput>(
+            {
+              method: "POST",
+              path: `/api/experimental/integration/wellknown`,
+              query: { location: input["location"] },
+              body: { url: input["url"] },
+              successStatus: 204,
+              declaredStatuses: [400, 401],
+              empty: true,
+            },
+            requestOptions,
+          ),
+      },
       connect: {
         key: (input: IntegrationConnectKeyInput, requestOptions?: RequestOptions) =>
           request<IntegrationConnectKeyOutput>(

packages/client/src/promise/generated/types.ts

@@ -3304,6 +3304,15 @@ export type IntegrationGetOutput = {
   data: IntegrationInfo | null
 }
 
+export type IntegrationWellknownAddInput = {
+  readonly location?: {
+    readonly location?: { readonly directory?: string | undefined; readonly workspace?: string | undefined } | undefined
+  }["location"]
+  readonly url: { readonly url: string }["url"]
+}
+
+export type IntegrationWellknownAddOutput = void
+
 export type IntegrationConnectKeyInput = {
   readonly integrationID: { readonly integrationID: string }["integrationID"]
   readonly location?: {

packages/client/test/promise.test.ts

@@ -36,7 +36,8 @@ test("exposes every standard HTTP API group", () => {
   expect(Object.keys(client.debug)).toEqual(["location"])
   expect(Object.keys(client.debug.location)).toEqual(["list", "evict"])
   expect(Object.keys(client.message)).toEqual(["list"])
-  expect(Object.keys(client.integration)).toEqual(["list", "get", "connect", "oauth", "command"])
+  expect(Object.keys(client.integration)).toEqual(["list", "get", "wellknown", "connect", "oauth", "command"])
+  expect(Object.keys(client.integration.wellknown)).toEqual(["add"])
   expect(Object.keys(client.integration.connect)).toEqual(["key"])
   expect(Object.keys(client.integration.oauth)).toEqual(["connect", "status", "complete", "cancel"])
   expect(Object.keys(client.integration.command)).toEqual(["connect", "status", "cancel"])
@@ -62,6 +63,28 @@ test("server.get uses the public HTTP contract", async () => {
   expect(request?.url).toBe("http://localhost:3000/api/server")
 })
 
+test("experimental wellknown integration add uses the public HTTP contract", async () => {
+  let request: Request | undefined
+  const client = OpenCode.make({
+    baseUrl: "http://localhost:3000",
+    fetch: async (input, init) => {
+      request = input instanceof Request ? input : new Request(input, init)
+      return new Response(null, { status: 204 })
+    },
+  })
+
+  await client.integration.wellknown.add({
+    url: "https://example.com",
+    location: { directory: "/tmp/project" },
+  })
+
+  expect(request?.method).toBe("POST")
+  expect(request?.url).toBe(
+    "http://localhost:3000/api/experimental/integration/wellknown?location%5Bdirectory%5D=%2Ftmp%2Fproject",
+  )
+  expect(await request?.json()).toEqual({ url: "https://example.com" })
+})
+
 test("health.stop sends exact replacement identity", async () => {
   let request: Request | undefined
   const client = OpenCode.make({

packages/codemode/interpreter-support.md

@@ -4,10 +4,9 @@ This is the checkable support matrix for CodeMode's confined JavaScript interpre
 standard-library surface that programs can use today, plus concrete gaps that may be implemented later.
 
 - `[x]` means the feature is implemented at the scope described here.
-- `[ ]` means the feature is unavailable, incomplete, or intentionally divergent as described.
-- A checked item does not promise complete ECMAScript edge-case parity. Known differences are listed next to the
-  supported surface or under [Known semantic gaps](#known-semantic-gaps).
-- [Intentional exclusions](#intentional-exclusions) are boundaries, not backlog.
+- `[ ]` means a concrete compatibility gap remains.
+- Checked items do not promise complete ECMAScript edge-case parity; known differences are stated explicitly.
+- Intentional boundaries are not listed as compatibility work.
 
 When behavior changes, update this file and the tests in the same change. The implementation and tests remain the
 ultimate source of truth.
@@ -19,42 +18,45 @@ ultimate source of truth.
       TypeScript is transpiled first; the emitted JavaScript must still use the supported subset.
 - [x] Top-level `await` and `return` through the program's implicit async-function scope.
 - [x] Explicit `return`, final top-level expression as a REPL-style result, and `null` when no value is produced.
-- [x] JSON-like host boundaries with `undefined` and non-finite numbers normalized to `null`.
+- [x] Program results use JSON-like boundaries, with `undefined` and non-finite numbers normalized to `null`. Tool
+      arguments remain subject to their schema and the outbound-handling gap listed below.
 - [x] Live Date, RegExp, Map, Set, URL, and URLSearchParams values inside CodeMode.
 - [x] Tool calls through the host-provided `tools` tree only.
 - [x] The global `search(...)` built-in: synchronous tool discovery that counts as an admitted tool call and is
       shadowable by program declarations like other globals.
 - [x] Cooperative timeout, an optional total tool-call limit, output bounding, and unrestricted tool-call concurrency.
-- [ ] Full JavaScript or TypeScript compatibility. CodeMode is a bounded orchestration language.
 
 ## Values and literals
 
 - [x] `null`, `undefined`, booleans, finite and non-finite numbers, and strings.
 - [x] Array literals, including holes and spread from arrays, strings, Maps, Sets, and URLSearchParams.
-- [x] Object literals with shorthand, computed string/number keys, and object spread.
+- [x] Object literals with shorthand, computed string/number keys, and spread from plain data objects; `null` and
+      `undefined` are no-ops, while arrays are rejected.
 - [x] Template literals with interpolation.
 - [x] Regular-expression literals.
 - [x] `NaN` and `Infinity` globals.
-- [ ] BigInt literals and values.
-- [ ] Symbols.
-- [ ] Tagged template literals.
-- [ ] Getters and setters in object literals.
+- [ ] BigInt literals and in-interpreter BigInt arithmetic; BigInt remains invalid at JSON-like host boundaries.
+- [ ] Symbol primitive values and symbol-keyed properties.
+- [ ] Tagged-template calls.
+- [ ] Getter and setter definitions in object literals.
 
 ## Bindings and destructuring
 
 - [x] `const`, `let`, and accepted `var` declarations.
 - [x] Object and array destructuring in declarations, parameters, assignment expressions, and `for...of` bindings.
 - [x] Nested patterns, defaults, elisions, and rest elements.
-- [x] Assignment to identifiers, object fields, array indexes, and writable URL fields.
-- [x] Function declarations are hoisted within their interpreted scope.
+- [x] Assignment to identifiers, unblocked plain-object fields, non-negative integer array indexes, and writable URL
+      fields.
+- [x] Direct function declarations are hoisted in program and block statement lists.
 - [x] Parameter defaults observe a temporal dead zone for later parameters.
-- [ ] JavaScript-correct `var` function scope, hoisting, and redeclaration. Accepted `var` currently behaves like a
-      lexical declaration; prefer `let` or `const`.
-- [ ] Complete `let`/`const` temporal-dead-zone and declaration-hoisting semantics.
-- [ ] Computed object destructuring keys such as `const { [field]: value } = record`.
-- [ ] Object destructuring from arrays, such as `const { length } = values`.
-- [ ] Iterable array destructuring from Map, Set, string, or URLSearchParams values.
-- [ ] Dynamic property deletion with `delete object[key]`.
+- [ ] JavaScript-correct function scoping, hoisting, and redeclaration for accepted `var` declarations.
+- [ ] Predeclare `let` and `const` bindings in every lexical scope, including program/block bodies, switch bodies, and
+      loop headers, so reads before initialization and self- or cross-referential initializers observe the JavaScript
+      temporal dead zone.
+- [ ] Hoist function declarations accepted directly in switch cases.
+- [x] Computed object destructuring keys such as `const { [field]: value } = record`.
+- [x] Object destructuring from arrays, such as `const { length } = values`.
+- [x] Array destructuring from supported non-array iterables: strings, Maps, Sets, and URLSearchParams.
 
 ## Statements and control flow
 
@@ -69,7 +71,6 @@ ultimate source of truth.
 - [x] `throw` with arbitrary values.
 - [ ] Labeled statements, labeled `break`, and labeled `continue`.
 - [ ] `for await...of` and async iteration.
-- [ ] `with` and `debugger` statements.
 
 ## Functions and callbacks
 
@@ -90,15 +91,15 @@ ultimate source of truth.
       like JS.
 - [x] Tool references and detached `Promise` statics are rejected as callbacks with a hint to wrap them in an
       arrow function.
-- [x] Async string replacement callbacks; replacements are evaluated sequentially.
+- [ ] Stop automatically awaiting promise-returning string replacers; match JavaScript's synchronous callback-result
+      coercion.
 - [x] The optional `thisArg` of iteration methods is accepted and ignored: CodeMode functions have no `this`, so
       ignoring it matches JS arrow-function semantics exactly.
-- [ ] `this`, `super`, user-defined constructor functions, or function prototype methods such as `call`, `apply`,
-      and `bind`.
+- [ ] `this` in non-arrow CodeMode functions and callbacks.
+- [ ] User-defined constructor calls.
+- [ ] `Function.prototype.call`, `apply`, and `bind` for CodeMode functions.
 - [ ] Classes and private fields.
 - [ ] Generator functions and `yield`.
-- [ ] Async predicates, reducers, and comparators with automatic awaiting. Async mapping can be joined explicitly with
-      `Promise.all`, but a promise is not a meaningful predicate or sort result.
 
 ## Expressions and operators
 
@@ -108,16 +109,16 @@ ultimate source of truth.
 - [x] Sequence expressions (the comma operator).
 - [x] `await` for CodeMode promises; a plain value passes through unchanged, though every `await` still defers its
       continuation one reaction turn.
-- [x] `new` for Error types, Date, RegExp, Map, Set, URL, URLSearchParams, and Promise.
+- [x] `new` for Array, Object, Error types, Date, RegExp, Map, Set, URL, URLSearchParams, and Promise.
 - [x] Arithmetic operators: `+`, `-`, `*`, `/`, `%`, and `**`.
 - [x] Equality and ordering: `==`, `!=`, `===`, `!==`, `<`, `<=`, `>`, and `>=`.
 - [x] Bitwise operators: `&`, `|`, `^`, `~`, `<<`, `>>`, and `>>>`.
 - [x] Logical operators: `&&`, `||`, `??`, and `!`, with short-circuiting.
-- [x] Unary `+`, unary `-`, `typeof`, `instanceof`, and own-property-only `in`.
+- [x] Unary `+`, unary `-`, `void`, `typeof`, `instanceof`, and own-property-only `in`.
 - [x] Prefix and postfix `++` and `--`.
 - [x] Plain, arithmetic, bitwise, and logical assignment operators.
-- [ ] Unary `void` and `delete`.
-- [ ] Arbitrary constructors.
+- [x] Property deletion on plain data objects and arrays, including computed and optional forms; deleting an array index
+      creates a hole without changing its length.
 
 ## Promises and tools
 
@@ -152,8 +153,13 @@ ultimate source of truth.
       promise itself rejects with a `TypeError`. Resolver callables work anywhere callbacks are accepted, including
       `.then`/`.catch` handlers and collection callbacks, but remain opaque references that cannot cross the data
       boundary.
-- [ ] Thenable assimilation (objects with a `then` method are plain data, not promises).
-- [ ] Async iterables, host streams, and stream consumption.
+- [ ] Thenable assimilation; objects with a callable `then` field remain plain data.
+- [x] Dotted tool names are canonicalized into namespace paths; a path can be both callable and a namespace, and the
+      last definition supplied for a canonical path wins.
+- [x] Tool path segments may be named `constructor`, `prototype`, or `__proto__` because paths use inert Map keys.
+- [ ] Reject `undefined` and non-finite numbers in outbound tool arguments before render-only and OpenAPI tools run;
+      retain null normalization for program results and JSON serialization.
+- [ ] Tokenize and case-fold non-ASCII tool paths, descriptions, and queries for tool search.
 
 ## Objects and properties
 
@@ -164,19 +170,17 @@ ultimate source of truth.
 - [x] `Object.keys`, `Object.values`, `Object.entries`, `Object.hasOwn`, `Object.assign`, and `Object.fromEntries`.
 - [x] `Object.keys` over arrays and tool references.
 - [x] Object identity is preserved by in-CodeMode Object helpers.
-- [x] Blocked access to `__proto__`, `constructor`, and `prototype`.
-- [ ] `Object.is`; runtime and tool-reference identity semantics need to be defined first.
+- [x] Prototype traversal and mutation through `__proto__`, `constructor`, and `prototype` are blocked.
+- [ ] Legal own data fields named `__proto__`, `constructor`, or `prototype` are rejected at JSON/tool boundaries and
+      cannot be created, read, or written in CodeMode; tool path segments with those names remain supported.
+- [x] `Object.is` for supported data values.
 - [ ] `Object.groupBy`.
-- [ ] Object creation, descriptors, freezing/sealing, prototype APIs, and reflection APIs.
-- [ ] A final policy for legal data keys named `__proto__`, `constructor`, or `prototype` (tool path segments
-      already allow them; see known semantic gaps).
 
 ## Arrays
 
-- [x] The `Array` constructor with or without `new`: `Array(a, b)` collects arguments, `Array(n)` creates a
-      sparse array of that length (invalid lengths throw a `RangeError`). Holes behave like JS in iteration,
-      spread, join, and JSON; `sort` densifies holes into trailing `undefined`, and results returned to the
-      host normalize holes to `null`.
+- [x] The `Array` constructor with or without `new`: `Array(a, b)` collects arguments and `Array(n)` creates a sparse
+      array of that length; invalid lengths throw `RangeError`. Iteration, spread, join, and JSON handle holes like
+      JavaScript, and host results normalize holes to `null`.
 - [x] Static methods: `Array.isArray`, `Array.of`, and `Array.from`, including the `Array.from` mapper form with
       `(value, index)` arguments.
 - [x] Iteration/transformation: `map`, `filter`, `flatMap`, and `forEach`.
@@ -190,8 +194,10 @@ ultimate source of truth.
 - [x] `length`, numeric indexing, index assignment, spread, and `for...of`.
 - [x] The `thisArg` argument of `Array.from` is accepted and ignored, like JS arrows.
 - [ ] `Array.prototype.toSpliced`.
-- [ ] Canonical index handling: a key such as `"01"` must not alias index `1`.
-- [ ] Complete sparse-array parity. Promise combinators do consume holes as `undefined` members, as in JS.
+- [ ] Canonical array/string index parsing: a key such as `"01"` must remain an ordinary property key rather than
+      aliasing index `1`.
+- [ ] `Array.prototype.sort` and `toSorted` must preserve trailing holes; they currently turn holes into own
+      `undefined` elements.
 
 ## Strings
 
@@ -204,8 +210,8 @@ ultimate source of truth.
 - [x] `localeCompare`; locale and options arguments are currently ignored.
 - [x] `toString`, `length`, numeric indexing, spread, and `for...of` by Unicode code point.
 - [x] Static `String.fromCharCode` and `String.fromCodePoint`.
-- [ ] Locale/options-aware `localeCompare` and locale formatting APIs.
-- [ ] Exact native coercion across every string method; CodeMode often requires explicit strings/numbers.
+- [ ] Native argument coercion for supported String methods; for example, `includes(1)` and `slice("1")` currently
+      reject instead of coercing.
 - [ ] Native no-argument parity for `match()` and `search()`.
 
 ## Numbers and Math
@@ -221,20 +227,21 @@ ultimate source of truth.
       `floor`, `ceil`, `round`, `trunc`, `sign`, `sqrt`, `cbrt`, `pow`, `hypot`, `cos`, `cosh`, `sin`, `sinh`,
       `tan`, `tanh`, `log`, `log2`, `log10`, `log1p`, `exp`, `expm1`, `f16round`, `fround`, `clz32`, and `imul`.
 - [ ] Native zero-argument behavior for `Number()` and `String()`; they currently do not produce `0` and `""`.
-- [ ] Safe interpreter coercion for `++` and `--` rather than host `Number(...)` coercion.
-- [ ] Reliable feature detection for unknown static members.
+- [ ] `++` and `--` must use CodeMode numeric coercion and reject opaque runtime references; they currently call host
+      `Number(...)` directly.
+- [ ] Unknown static members must read as `undefined` for feature detection; some currently appear callable or throw
+      during property access.
 - [ ] `Math.sumPrecise`.
 - [ ] Global coercing `isFinite` and `isNaN`.
 
 ## JSON and console
 
-- [x] `JSON.parse` and `JSON.stringify`.
+- [x] `JSON.parse` and `JSON.stringify` for supported data objects; the blocked data-key gap listed above still applies.
 - [x] Numeric/string indentation for `JSON.stringify`.
 - [x] Captured `console.log`, `console.info`, `console.debug`, `console.warn`, and `console.error`.
 - [x] Captured `console.dir` and `console.table`.
 - [ ] `JSON.parse` reviver callbacks.
 - [ ] `JSON.stringify` function/array replacers.
-- [ ] Other console methods, timers, counters, groups, and host console access.
 
 ## Date
 
@@ -250,22 +257,22 @@ ultimate source of truth.
 - [x] `getTimezoneOffset`, arithmetic, relational comparison, and `instanceof Date`.
 - [x] Date values serialize to ISO strings; invalid dates serialize to `null`.
 - [ ] Date setters.
-- [ ] `toUTCString`, locale methods, and other Date formatting methods.
-- [ ] Exact native constructor coercion, local-time, and loose-equality semantics.
+- [ ] `Date.prototype.toUTCString` and its `toGMTString` alias.
+- [ ] Native one-argument Date coercion; unsupported boolean/object inputs currently become invalid dates instead of
+      being coerced.
+- [ ] Native Date loose-equality and default primitive-coercion semantics.
 - [ ] Native `RangeError` branding for invalid `toISOString()` calls.
-- [ ] Temporal and Intl date/time APIs.
 
 ## Regular expressions
 
 - [x] Literal and `RegExp(pattern, flags)` construction, with or without `new`.
 - [x] `test`, `exec`, and `toString`.
 - [x] Readable `source`, `flags`, `lastIndex`, `global`, `ignoreCase`, `multiline`, `sticky`, `unicode`, and `dotAll`.
-- [x] Captures, named groups, match indexes, and stateful global matching.
-- [x] Integration with supported String methods, including async function replacers.
+- [x] Captures, safe named groups (blocked member names are omitted), match `.index`, and stateful global matching.
+- [x] Integration with supported String methods, including function replacers.
 - [ ] Writable `lastIndex`.
-- [ ] Exposed metadata for the `d` and `v` flags.
+- [ ] `hasIndices`, match `indices`, and `unicodeSets` metadata for the `d` and `v` flags.
 - [ ] `RegExp.escape`.
-- [ ] Protection from pathological host-regex backtracking beyond the cooperative execution timeout.
 
 ## Map and Set
 
@@ -276,9 +283,8 @@ ultimate source of truth.
 - [x] Materialized `keys`, `values`, and `entries` arrays for Map and Set.
 - [x] Spread, `for...of`, `Array.from`, and `Object.fromEntries` integration.
 - [x] Map and Set values serialize to `{}` at host/JSON boundaries.
-- [ ] Set composition methods such as `union`, `intersection`, `difference`, and relation predicates.
-- [ ] WeakMap and WeakSet.
-- [ ] Native iterator objects and custom iterators.
+- [ ] Set composition and relation methods: `union`, `intersection`, `difference`, `symmetricDifference`, `isSubsetOf`,
+      `isSupersetOf`, and `isDisjointFrom`.
 
 ## URL and URI helpers
 
@@ -301,48 +307,12 @@ ultimate source of truth.
       an all-rejected `Promise.any`.
 - [x] Error `name`/`message`, error inheritance through `instanceof`, and plain-data serialization.
 - [x] `instanceof` for Date, RegExp, Map, Set, URL, URLSearchParams, Array, Object, Promise, and Error types.
-- [x] Catchable interpreter failures and awaited tool failures.
-- [x] Source locations on unsupported-syntax diagnostics when available.
+- [x] Catchable user throws, runtime failures raised during interpreted evaluation, awaited tool failures, and awaited
+      tool-call-limit failures; parse/compile failures, cooperative timeout, and output bounding remain outside program
+      `catch`.
+- [x] Source locations on unsupported-syntax diagnostics for JavaScript-shaped input; TypeScript transpilation may
+      shift them.
 - [x] Sanitized model-visible diagnostics and explicit safe `ToolError` messages.
-- [ ] Distinct public categories for user throws, tool refusal, tool internal failure, invalid returned data, compile
-      failures, and genuine interpreter defects.
-- [ ] Preservation of detailed recoverable failure categories inside `catch` and `Promise.allSettled`.
-
-## Known semantic gaps
-
-These are actionable implementation items. Check them off only when behavior and direct tests land.
-
-- [x] Return real promises from `Promise.all`, `Promise.allSettled`, and `Promise.race`.
-- [x] Canonicalize dotted tool names into namespace paths so every advertised dotted path is executable, one
-      canonical path can be both a callable tool and a namespace, and the last definition supplied for a canonical
-      path wins.
-- [x] Allow blocked member names (`constructor`, `prototype`, `__proto__`) as tool path segments: segments are Map
-      keys and inert strings, never plain-object property accesses, so every advertised path is executable. Blocked
-      member access on data values stays rejected. Tool names with empty segments are rejected at construction.
-- [ ] Define safe outbound handling for non-finite numbers and `undefined` so invalid values cannot silently become
-      `null` in render-only or OpenAPI tool calls.
-- [ ] Make regular-expression execution genuinely timeout-safe, or narrow the timeout guarantee explicitly.
-- [ ] Complete lexical declaration and destructuring semantics listed above.
-- [x] Make callback acceptance consistent across built-ins: collections, sort, string replacers, `Array.from`
-      mappers, and promise reactions share one acceptance rule.
-- [x] Reject every unsupported callable callback argument explicitly rather than silently ignoring it
-      (`JSON.stringify` replacers and `JSON.parse` revivers fail loudly). The iteration-method `thisArg` is
-      accepted and ignored: CodeMode functions have no `this`, so ignoring it matches JS arrow semantics.
-- [ ] Make async callback behavior consistent across built-ins; only string replacers settle async callback results
-      today.
-- [ ] Resolve the built-in correctness gaps listed in the Array, String, Number, Date, and RegExp sections.
-- [ ] Make tool search tokenization Unicode-aware.
-- [ ] Design explicit tagged representations and size limits before adding binary values or streams.
-
-## Intentional exclusions
-
-These constraints preserve CodeMode's confinement and host-neutral scope. They are not TODO items.
-
-- Ambient filesystem, process, environment, credential, network, or application access.
-- `fetch`, timers, crypto, or other host globals unless a future host explicitly supplies a bounded capability.
-- Static imports, dynamic imports, modules, npm packages, and module loading.
-- `eval`, `Function(...)`, arbitrary host execution, and prototype mutation.
-- Generic permission prompts, authorization policy, persistence, replay, or exactly-once side effects.
-- Arbitrary method dispatch outside the documented allowlists.
-- Automatic parsing of text tool results as JSON.
-- Full browser, Node.js, Bun, or ECMAScript runtime compatibility.
+- [ ] Distinguish user-thrown failures from interpreter defects and explicit tool refusals from sanitized internal tool
+      failures; preserve those categories in caught errors, promise rejection handlers, and `Promise.allSettled`
+      reasons.

packages/codemode/src/interpreter/runtime.ts

@@ -777,9 +777,9 @@ export class Interpreter<R> {
       }
 
       if (pattern.type === "ObjectPattern") {
-        if (value === null || typeof value !== "object" || Array.isArray(value) || isRuntimeReference(value)) {
+        if (value === null || typeof value !== "object" || isRuntimeReference(value)) {
           throw new InterpreterRuntimeError(
-            "Object destructuring requires a data object value.",
+            "Object destructuring requires a data object or array value.",
             pattern,
             "InvalidDataValue",
           )
@@ -798,38 +798,35 @@ export class Interpreter<R> {
             continue
           }
 
-          if (
-            property.type !== "Property" ||
-            getBoolean(property, "computed") ||
-            getString(property, "kind") !== "init"
-          ) {
-            throw new InterpreterRuntimeError("Only named object destructuring properties are supported.", property)
+          const key = yield* self.destructuringPropertyKey(property)
+          if (isBlockedMember(String(key))) {
+            throw new InterpreterRuntimeError(`Property '${String(key)}' is not available in CodeMode.`, property)
           }
-
-          const keyNode = getNode(property, "key")
-          const key = keyNode.type === "Identifier" ? getString(keyNode, "name") : String(keyNode.value)
-          if (isBlockedMember(key)) {
-            throw new InterpreterRuntimeError(`Property '${key}' is not available in CodeMode.`, keyNode)
-          }
-          consumed.add(key)
-          yield* self.declarePattern(getNode(property, "value"), (value as SafeObject)[key], mutable, property)
+          consumed.add(String(key))
+          yield* self.declarePattern(
+            getNode(property, "value"),
+            self.destructuringPropertyValue(value as SafeObject | Array<unknown>, key),
+            mutable,
+            property,
+          )
         }
         return
       }
 
       if (pattern.type === "ArrayPattern") {
-        if (!Array.isArray(value)) {
-          throw new InterpreterRuntimeError("Array destructuring requires an array value.", pattern)
+        const items = spreadItems(value)
+        if (items === undefined) {
+          throw new InterpreterRuntimeError("Array destructuring requires a supported iterable value.", pattern)
         }
 
         for (const [index, item] of getArray(pattern, "elements").entries()) {
           if (item === null) continue
           const element = asNode(item, `elements[${index}]`)
           if (element.type === "RestElement") {
-            yield* self.declarePattern(getNode(element, "argument"), value.slice(index), mutable, element)
+            yield* self.declarePattern(getNode(element, "argument"), items.slice(index), mutable, element)
             break
           }
-          yield* self.declarePattern(element, value[index], mutable, pattern)
+          yield* self.declarePattern(element, items[index], mutable, pattern)
         }
         return
       }
@@ -858,15 +855,15 @@ export class Interpreter<R> {
       }
 
       if (pattern.type === "ObjectPattern") {
-        if (value === null || typeof value !== "object" || Array.isArray(value) || isRuntimeReference(value)) {
+        if (value === null || typeof value !== "object" || isRuntimeReference(value)) {
           throw new InterpreterRuntimeError(
-            "Object destructuring requires a data object value.",
+            "Object destructuring requires a data object or array value.",
             pattern,
             "InvalidDataValue",
           )
         }
 
-        const source = value as SafeObject
+        const source = value as SafeObject | Array<unknown>
         const consumed = new Set<string>()
         for (const propertyValue of getArray(pattern, "properties")) {
           const property = asNode(propertyValue, "properties")
@@ -878,36 +875,29 @@ export class Interpreter<R> {
             yield* self.assignPattern(getNode(property, "argument"), rest, property)
             continue
           }
-          if (
-            property.type !== "Property" ||
-            getBoolean(property, "computed") ||
-            getString(property, "kind") !== "init"
-          ) {
-            throw new InterpreterRuntimeError("Only named object destructuring properties are supported.", property)
+          const key = yield* self.destructuringPropertyKey(property)
+          if (isBlockedMember(String(key))) {
+            throw new InterpreterRuntimeError(`Property '${String(key)}' is not available in CodeMode.`, property)
           }
-          const keyNode = getNode(property, "key")
-          const key = keyNode.type === "Identifier" ? getString(keyNode, "name") : String(keyNode.value)
-          if (isBlockedMember(key)) {
-            throw new InterpreterRuntimeError(`Property '${key}' is not available in CodeMode.`, keyNode)
-          }
-          consumed.add(key)
-          yield* self.assignPattern(getNode(property, "value"), source[key], property)
+          consumed.add(String(key))
+          yield* self.assignPattern(getNode(property, "value"), self.destructuringPropertyValue(source, key), property)
         }
         return
       }
 
       if (pattern.type === "ArrayPattern") {
-        if (!Array.isArray(value)) {
-          throw new InterpreterRuntimeError("Array destructuring requires an array value.", pattern)
+        const items = spreadItems(value)
+        if (items === undefined) {
+          throw new InterpreterRuntimeError("Array destructuring requires a supported iterable value.", pattern)
         }
         for (const [index, item] of getArray(pattern, "elements").entries()) {
           if (item === null) continue
           const element = asNode(item, `elements[${index}]`)
           if (element.type === "RestElement") {
-            yield* self.assignPattern(getNode(element, "argument"), value.slice(index), element)
+            yield* self.assignPattern(getNode(element, "argument"), items.slice(index), element)
             break
           }
-          yield* self.assignPattern(element, value[index], pattern)
+          yield* self.assignPattern(element, items[index], pattern)
         }
         return
       }
@@ -916,6 +906,26 @@ export class Interpreter<R> {
     })
   }
 
+  private destructuringPropertyKey(property: AstNode): Effect.Effect<string | number, unknown, R> {
+    if (property.type !== "Property" || getString(property, "kind") !== "init") {
+      throw new InterpreterRuntimeError("Unsupported object destructuring property.", property)
+    }
+    const keyNode = getNode(property, "key")
+    if (getBoolean(property, "computed")) {
+      return Effect.map(this.evaluateExpression(keyNode), (value) => this.toPropertyKey(value, keyNode))
+    }
+    return Effect.succeed(keyNode.type === "Identifier" ? getString(keyNode, "name") : String(keyNode.value))
+  }
+
+  private destructuringPropertyValue(source: SafeObject | Array<unknown>, key: string | number): unknown {
+    if (!Array.isArray(source)) return source[String(key)]
+    if (key === "length") return source.length
+    if (typeof key === "number") return source[key]
+    if (Object.hasOwn(source, key)) return (source as Record<string, unknown> & Array<unknown>)[key]
+    if (arrayMethods.has(key)) return new IntrinsicReference(source, key)
+    return undefined
+  }
+
   private evaluateExpression(node: AstNode): Effect.Effect<unknown, unknown, R> {
     switch (node.type) {
       case "Literal": {
@@ -1279,6 +1289,7 @@ export class Interpreter<R> {
   private evaluateUnaryExpression(node: AstNode): Effect.Effect<unknown, unknown, R> {
     const operator = getString(node, "operator")
     const argument = getNode(node, "argument")
+    if (operator === "delete") return this.evaluateDeleteExpression(argument)
     // Undeclared names short-circuit, but declared TDZ bindings must still throw.
     if (operator === "typeof" && argument.type === "Identifier" && !this.scopes.resolve(getString(argument, "name"))) {
       return Effect.succeed("undefined")
@@ -1286,6 +1297,7 @@ export class Interpreter<R> {
     return Effect.map(this.evaluateExpression(argument), (value) => {
       if (operator === "typeof") return typeofValue(value)
       if (operator === "!") return !value
+      if (operator === "void") return undefined
       if (containsOpaqueReference(value)) {
         throw new InterpreterRuntimeError("Unary operators require data values in CodeMode.", node, "InvalidDataValue")
       }
@@ -1729,6 +1741,7 @@ export class Interpreter<R> {
 
   private getMemberReference(
     node: AstNode,
+    operation: "read" | "delete" = "read",
   ): Effect.Effect<
     | MemberReference
     | ToolReference
@@ -1875,6 +1888,7 @@ export class Interpreter<R> {
       }
 
       if (Array.isArray(objectValue)) {
+        if (operation === "delete") return { target: objectValue, key }
         if (
           key !== "length" &&
           !(typeof key === "string" && arrayMethods.has(key)) &&
@@ -1923,6 +1937,29 @@ export class Interpreter<R> {
     return this.modifyMember(node, () => Effect.succeed({ write: true, next: value, result: value }))
   }
 
+  private evaluateDeleteExpression(argument: AstNode): Effect.Effect<boolean, unknown, R> {
+    const target = argument.type === "ChainExpression" ? getNode(argument, "expression") : argument
+    if (target.type !== "MemberExpression") {
+      throw new InterpreterRuntimeError("Only data fields may be deleted in CodeMode.", argument)
+    }
+    return Effect.map(this.getMemberReference(target, "delete"), (reference) => {
+      if (reference === OptionalShortCircuit) return true
+      if (
+        reference instanceof ComputedValue ||
+        reference === undefined ||
+        reference instanceof ToolReference ||
+        reference instanceof PromiseMethodReference ||
+        reference instanceof PromiseInstanceMethodReference ||
+        reference instanceof IntrinsicReference ||
+        reference instanceof GlobalMethodReference ||
+        reference.target instanceof CodeModeURL
+      ) {
+        throw new InterpreterRuntimeError("Only data fields may be deleted in CodeMode.", target, "InvalidDataValue")
+      }
+      return Reflect.deleteProperty(reference.target, reference.key)
+    })
+  }
+
   // Resolve side-effecting object and key expressions exactly once.
   private modifyMember(
     node: AstNode,

packages/codemode/src/stdlib/object.ts

@@ -1,4 +1,5 @@
 import { type AstNode, InterpreterRuntimeError } from "../interpreter/model.js"
+import { containsOpaqueReference } from "../interpreter/references.js"
 import { isBlockedMember } from "../tool-runtime.js"
 import { isCodeModeValue, CodeModeMap, CodeModePromise, CodeModeSet, CodeModeURLSearchParams } from "../values.js"
 import { boundedData, coerceToString } from "./value.js"
@@ -44,6 +45,11 @@ export const invokeObjectMethod = (name: string, args: Array<unknown>, node: Ast
       return Object.entries(requireObject()).map(([key, item]) => [key, item])
     case "hasOwn":
       return Object.hasOwn(requireObject(), String(args[1]))
+    case "is":
+      if (containsOpaqueReference(args[0]) || containsOpaqueReference(args[1])) {
+        throw new InterpreterRuntimeError("Object.is requires data values in CodeMode.", node, "InvalidDataValue")
+      }
+      return Object.is(args[0], args[1])
     case "assign": {
       const target = args[0]
       if (target === null || typeof target !== "object" || Array.isArray(target) || isCodeModeValue(target)) {

packages/codemode/test/parity.test.ts

@@ -99,6 +99,84 @@ describe("H4: typeof on an undeclared identifier is 'undefined'", () => {
   })
 })
 
+describe("unary void", () => {
+  test("evaluates its operand and returns undefined", async () => {
+    expect(await value(`let count = 0; const result = void (count += 1); return [count, result === undefined]`)).toEqual([
+      1,
+      true,
+    ])
+  })
+
+  test("discards opaque values", async () => {
+    expect(await value(`return void tools === undefined`)).toBe(true)
+  })
+})
+
+describe("property deletion", () => {
+  test("deletes plain object fields and reports missing fields as successful", async () => {
+    expect(
+      await value(`
+        const object = { keep: 1, remove: 2 }
+        return [delete object.remove, delete object.missing, object]
+      `),
+    ).toEqual([true, true, { keep: 1 }])
+  })
+
+  test("evaluates computed object and key expressions once", async () => {
+    expect(
+      await value(`
+        const object = { remove: true }
+        let objectReads = 0
+        let keyReads = 0
+        function getObject() { objectReads++; return object }
+        function getKey() { keyReads++; return "remove" }
+        const removed = delete getObject()[getKey()]
+        return [removed, objectReads, keyReads, Object.hasOwn(object, "remove")]
+      `),
+    ).toEqual([true, 1, 1, false])
+  })
+
+  test("deleting an array index creates a hole without changing its length", async () => {
+    expect(await value(`const values = [1, 2, 3]; const removed = delete values[1]; return [removed, values.length, 1 in values, values]`)).toEqual([
+      true,
+      3,
+      false,
+      [1, null, 3],
+    ])
+  })
+
+  test("array length is not configurable", async () => {
+    expect(await value(`const values = [1, 2]; return [delete values.length, values.length]`)).toEqual([false, 2])
+  })
+
+  test("does not broaden unsupported array property assignment", async () => {
+    expect(
+      await value(`
+        const values = []
+        let rightHandSideRuns = 0
+        function next() { rightHandSideRuns++; return 1 }
+        try { values.field = next() } catch {}
+        return rightHandSideRuns
+      `),
+    ).toBe(0)
+  })
+
+  test("optional deletion short-circuits without evaluating the key", async () => {
+    expect(
+      await value(`let keyReads = 0; const object = null; return [delete object?.[keyReads++], keyReads]`),
+    ).toEqual([true, 0])
+  })
+
+  test("rejects deletion from opaque runtime references", async () => {
+    expect((await error(`return delete tools.example`)).kind).toBe("InvalidDataValue")
+  })
+
+  test("keeps blocked property names unavailable", async () => {
+    expect((await error(`const object = {}; return delete object.__proto__`)).kind).toBe("ExecutionFailure")
+    expect((await error(`const values = []; return delete values["constructor"]`)).kind).toBe("ExecutionFailure")
+  })
+})
+
 describe("H1: NaN/Infinity flow as intermediates and normalize to null at the boundary", () => {
   test("guards run instead of the program crashing on a transient NaN", async () => {
     expect(await value(`return parseInt("abc") || 0`)).toBe(0)
@@ -470,4 +548,71 @@ describe("destructuring assignment", () => {
   test("returns the assigned value", async () => {
     expect(await value(`let a = 0; const result = ([a] = [7]); return [a, result]`)).toEqual([7, [7]])
   })
+
+  test("supports computed object keys and evaluates them once", async () => {
+    expect(
+      await value(`
+        let calls = 0
+        const field = () => { calls++; return "name" }
+        const { [field()]: name, ...rest } = { name: "Ada", role: "engineer" }
+        return { calls, name, rest }
+      `),
+    ).toEqual({ calls: 1, name: "Ada", rest: { role: "engineer" } })
+  })
+
+  test("supports object patterns over arrays", async () => {
+    expect(
+      await value(`
+        const { 0: first, length, slice, ...rest } = ["a", "b", "c"]
+        return { first, length, sliced: slice(1), rest }
+      `),
+    ).toEqual({ first: "a", length: 3, sliced: ["b", "c"], rest: { 1: "b", 2: "c" } })
+  })
+
+  test("preserves exact computed property names on arrays", async () => {
+    expect(
+      await value(`
+        const { ["01"]: item, ...rest } = [10, 20]
+        return { missing: item === undefined, rest }
+      `),
+    ).toEqual({ missing: true, rest: { 0: 10, 1: 20 } })
+  })
+
+  test("supports array patterns over strings, Maps, Sets, and URLSearchParams", async () => {
+    expect(
+      await value(`
+        const [letter, ...letters] = "A😀B"
+        const [[mapKey, mapValue]] = new Map([["key", 1]])
+        const [setFirst, setSecond] = new Set([2, 3])
+        const [[queryKey, queryValue]] = new URLSearchParams("q=test&page=2")
+        return { letter, letters, mapKey, mapValue, setFirst, setSecond, queryKey, queryValue }
+      `),
+    ).toEqual({
+      letter: "A",
+      letters: ["😀", "B"],
+      mapKey: "key",
+      mapValue: 1,
+      setFirst: 2,
+      setSecond: 3,
+      queryKey: "q",
+      queryValue: "test",
+    })
+  })
+
+  test("supports iterable patterns in assignment and parameters", async () => {
+    expect(
+      await value(`
+        let first
+        let rest
+        ;[first, ...rest] = new Set([1, 2, 3])
+        const read = ([[key, value]]) => key + value
+        return { first, rest, entry: read(new Map([["a", 4]])) }
+      `),
+    ).toEqual({ first: 1, rest: [2, 3], entry: "a4" })
+  })
+
+  test("rejects computed keys that are not confined property keys", async () => {
+    const err = await error(`const key = {}; const { [key]: value } = {}`)
+    expect(err.message).toContain("Property key must be a string or number")
+  })
 })

packages/codemode/test/stdlib.test.ts

@@ -593,6 +593,24 @@ describe("Set", () => {
 })
 
 describe("stdlib integration", () => {
+  test("Object.is uses SameValue semantics", async () => {
+    expect(
+      await value(`
+        const object = {}
+        return [
+          Object.is(NaN, NaN),
+          Object.is(0, -0),
+          Object.is(object, object),
+          Object.is({}, {}),
+        ]
+      `),
+    ).toEqual([true, false, true, false])
+  })
+
+  test("Object.is rejects opaque runtime references", async () => {
+    expect((await error(`return Object.is(Math.max, Math.max)`)).kind).toBe("InvalidDataValue")
+  })
+
   test("Object values and entries accept arrays", async () => {
     expect(await value(`return [Object.values(["a", "b"]), Object.entries(["a", "b"])]`)).toEqual([
       ["a", "b"],

packages/core/schema.json

@@ -1,9 +1,9 @@
 {
   "version": "7",
   "dialect": "sqlite",
-  "id": "5f0a1db8-d4bf-42c3-becb-96b46fe66bed",
+  "id": "a4ba73b4-21bc-41ab-a415-94e2ca38d798",
   "prevIds": [
-    "666138ef-82cb-4a9a-a765-e6669a436ff3"
+    "5f0a1db8-d4bf-42c3-becb-96b46fe66bed"
   ],
   "ddl": [
     {
@@ -38,6 +38,10 @@
       "name": "event",
       "entityType": "tables"
     },
+    {
+      "name": "kv",
+      "entityType": "tables"
+    },
     {
       "name": "permission",
       "entityType": "tables"
@@ -556,6 +560,46 @@
       "entityType": "columns",
       "table": "event"
     },
+    {
+      "type": "text",
+      "notNull": false,
+      "autoincrement": false,
+      "default": null,
+      "generated": null,
+      "name": "key",
+      "entityType": "columns",
+      "table": "kv"
+    },
+    {
+      "type": "text",
+      "notNull": true,
+      "autoincrement": false,
+      "default": null,
+      "generated": null,
+      "name": "value",
+      "entityType": "columns",
+      "table": "kv"
+    },
+    {
+      "type": "integer",
+      "notNull": true,
+      "autoincrement": false,
+      "default": null,
+      "generated": null,
+      "name": "time_created",
+      "entityType": "columns",
+      "table": "kv"
+    },
+    {
+      "type": "integer",
+      "notNull": true,
+      "autoincrement": false,
+      "default": null,
+      "generated": null,
+      "name": "time_updated",
+      "entityType": "columns",
+      "table": "kv"
+    },
     {
       "type": "text",
       "notNull": false,
@@ -1844,6 +1888,15 @@
       "table": "event",
       "entityType": "pks"
     },
+    {
+      "columns": [
+        "key"
+      ],
+      "nameExplicit": false,
+      "name": "kv_pk",
+      "table": "kv",
+      "entityType": "pks"
+    },
     {
       "columns": [
         "id"

packages/core/script/wellknown-server.ts

@@ -0,0 +1,112 @@
+const token = "dummy-wellknown-token"
+const port = Number(process.env.PORT ?? 8787)
+const config = {
+  $schema: "https://opencode.ai/config.json",
+  share: "manual",
+  model: "example-primary/example-chat",
+  enabled_providers: ["example-primary", "example-secondary"],
+  disabled_providers: ["opencode", "anthropic", "openai", "google", "xai", "amazon-bedrock", "azure"],
+  provider: {
+    "example-primary": {
+      name: "Example Primary",
+      npm: "@ai-sdk/openai-compatible",
+      whitelist: ["example-chat", "example-code"],
+      options: {
+        baseURL: "https://models.example.com/v1",
+        apiKey: "{env:TOKEN}",
+      },
+      models: {
+        "example-chat": {
+          name: "Example Chat",
+          reasoning: true,
+          tool_call: true,
+          attachment: true,
+          modalities: {
+            input: ["text", "image"],
+            output: ["text"],
+          },
+          limit: {
+            context: 128000,
+            output: 16000,
+          },
+        },
+        "example-code": {
+          name: "Example Code",
+          reasoning: true,
+          tool_call: true,
+          limit: {
+            context: 200000,
+            output: 32000,
+          },
+        },
+      },
+    },
+    "example-secondary": {
+      name: "Example Secondary",
+      npm: "@ai-sdk/openai-compatible",
+      whitelist: ["example-fast"],
+      options: {
+        baseURL: "https://inference.example.org/v1",
+        apiKey: "{env:TOKEN}",
+      },
+      models: {
+        "example-fast": {
+          name: "Example Fast",
+          tool_call: true,
+          limit: {
+            context: 64000,
+            output: 8000,
+          },
+        },
+      },
+    },
+  },
+  mcp: {
+    "example-tools": {
+      type: "remote",
+      url: "https://tools.example.net/mcp",
+      enabled: false,
+    },
+  },
+  permission: {
+    bash: "ask",
+    edit: "ask",
+    webfetch: "ask",
+    read: {
+      "*": "allow",
+      "*.env": "deny",
+      "*.env.*": "deny",
+      "*.env.example": "allow",
+    },
+  },
+}
+
+const server = Bun.serve({
+  hostname: "127.0.0.1",
+  port,
+  async fetch(request) {
+    const url = new URL(request.url)
+    if (url.pathname === "/.well-known/opencode") {
+      return Response.json({
+        auth: {
+          command: ["bun", "-e", `await Bun.sleep(5000); process.stdout.write(${JSON.stringify(token)})`],
+          env: "TOKEN",
+        },
+        remote_config: {
+          url: `${url.origin}/config/opencode.json`,
+          headers: { authorization: "Bearer {env:TOKEN}" },
+        },
+      })
+    }
+    if (url.pathname === "/config/opencode.json") {
+      if (request.headers.get("authorization") !== `Bearer ${token}`) {
+        return new Response("Unauthorized", { status: 401 })
+      }
+      return Response.json(config)
+    }
+    return new Response("Not found", { status: 404 })
+  },
+})
+
+console.log(`Well-known fixture listening at ${server.url.origin}`)
+console.log(`Test with: bun run dev auth connect ${server.url.origin}`)

packages/core/src/config.ts

@@ -2,10 +2,13 @@ export * as Config from "./config"
 
 import { makeLocationNode } from "./effect/app-node"
 import path from "path"
+import { isDeepStrictEqual } from "node:util"
 import { type ParseError, parse } from "jsonc-parser"
-import { Context, Effect, Fiber, Layer, Option, PubSub, Schema, Stream } from "effect"
+import { Context, Effect, Fiber, Layer, Option, PubSub, Schema, Semaphore, Stream } from "effect"
 import { Permission } from "@opencode-ai/schema/permission"
 import { Config as ConfigSchema } from "@opencode-ai/schema/config"
+import { Integration } from "@opencode-ai/schema/integration"
+import { Credential } from "./credential"
 import { EventV2 } from "./event"
 import { Watcher } from "./filesystem/watcher"
 import { FSUtil } from "./fs-util"
@@ -16,6 +19,7 @@ import { ConfigAgent } from "./config/agent"
 import { ConfigAttachments } from "./config/attachments"
 import { ConfigCompaction } from "./config/compaction"
 import { ConfigCommand } from "./config/command"
+import { ConfigExperimental } from "./config/experimental"
 import { ConfigFormatter } from "./config/formatter"
 import { ConfigLSP } from "./config/lsp"
 import { ConfigMCP } from "./config/mcp"
@@ -28,6 +32,7 @@ import { ConfigVariable } from "./config/variable"
 import { ConfigWatcher } from "./config/watcher"
 import { ConfigV1 } from "./v1/config/config"
 import { ConfigMigrateV1 } from "./v1/config/migrate"
+import { WellKnown } from "./wellknown"
 
 export class Info extends Schema.Class<Info>("Config.Info")({
   $schema: Schema.optional(Schema.String).annotate({
@@ -106,6 +111,7 @@ export class Info extends Schema.Class<Info>("Config.Info")({
     description: "Ordered plugin enablement directives and external package declarations",
   }),
   providers: Schema.Record(Schema.String, ConfigProvider.Info).pipe(Schema.optional),
+  experimental: ConfigExperimental.Info.pipe(Schema.optional),
 }) {}
 
 export class Document extends Schema.Class<Document>("Config.Document")({
@@ -157,29 +163,68 @@ const layer = Layer.effect(
     const location = yield* Location.Service
     const watcher = yield* Watcher.Service
     const events = yield* EventV2.Service
+    const credentials = yield* Credential.Service
+    const wellknown = yield* WellKnown.Service
     const names = ["opencode.json", "opencode.jsonc"]
+    const reloadLock = Semaphore.makeUnsafe(1)
     const decodeOptions = { errors: "all", onExcessProperty: "ignore", propertyOrder: "original" } as const
     const decodeInfo = Schema.decodeUnknownOption(Info, decodeOptions)
     const decodeV1Info = Schema.decodeUnknownOption(ConfigV1.Info, decodeOptions)
 
+    const parseInfo = (text: string) => {
+      const errors: ParseError[] = []
+      const input: unknown = parse(text, errors, { allowTrailingComma: true })
+      if (errors.length) return
+      return Option.getOrUndefined(
+        ConfigMigrateV1.isV1(input)
+          ? decodeV1Info(input).pipe(Option.map(ConfigMigrateV1.migrate), Option.flatMap(decodeInfo))
+          : decodeInfo(input),
+      )
+    }
+
     const loadFile = Effect.fnUntraced(function* (filepath: string) {
       const text = yield* fs.readFileStringSafe(filepath)
       if (!text) return
       const substituted = yield* ConfigVariable.substitute({ type: "path", path: filepath, text })
-
-      const errors: ParseError[] = []
-      const input: unknown = parse(substituted, errors, { allowTrailingComma: true })
-      if (errors.length) return
-
-      const info = Option.getOrUndefined(
-        ConfigMigrateV1.isV1(input)
-          ? decodeV1Info(input).pipe(Option.map(ConfigMigrateV1.migrate), Option.flatMap(decodeInfo))
-          : decodeInfo(input),
-      )
+      const info = parseInfo(substituted)
       if (!info) return
       return new Document({ type: "document", path: filepath, info })
     })
 
+    const loadWellknown = Effect.fn("Config.loadWellknown")(function* () {
+      const entries = yield* wellknown
+        .entries()
+        .pipe(
+          Effect.catch((error) =>
+            Effect.logWarning("failed to discover wellknown config", { error }).pipe(Effect.as([] as const)),
+          ),
+        )
+      return yield* Effect.forEach(entries, (entry) =>
+        Effect.gen(function* () {
+          const auth = entry.manifest.auth
+          if (!auth) return []
+          const credential = (yield* credentials.list(entry.integrationID)).findLast(
+            (credential) => credential.value.type === "key",
+          )
+          if (!credential || credential.value.type !== "key") return []
+          const variables = { [auth.env]: credential.value.key }
+          const configs = yield* wellknown.resolve(entry, variables).pipe(Effect.orDie)
+          return yield* Effect.forEach(configs, (config) =>
+            ConfigVariable.substitute({
+              type: "virtual",
+              source: entry.origin,
+              dir: entry.origin,
+              text: JSON.stringify(config),
+              env: variables,
+            }).pipe(
+              Effect.map(parseInfo),
+              Effect.map((info) => (info ? new Document({ type: "document", info }) : undefined)),
+            ),
+          ).pipe(Effect.map((documents) => documents.filter((document) => document !== undefined)))
+        }),
+      ).pipe(Effect.map((documents) => documents.flat()))
+    })
+
     const loadDirectory = Effect.fnUntraced(function* (directory: AbsolutePath) {
       return [
         ...(yield* Effect.forEach(names, (file) => loadFile(path.join(directory, file))).pipe(
@@ -201,7 +246,7 @@ const layer = Layer.effect(
               targets: [".opencode", ".claude", ".agents", ...names.toReversed()],
               start: location.directory,
             })
-          .pipe(Effect.orDie)
+            .pipe(Effect.orDie)
 
       // We load certain files from a few other folders in the ecosystem
       const claude = [
@@ -244,7 +289,14 @@ const layer = Layer.effect(
       )
 
       const supplementary = yield* Effect.forEach(directories, loadDirectory).pipe(Effect.orDie)
-      return [...claude, ...agents, ...(supplementary[0] ?? []), ...direct, ...supplementary.slice(1).flat()]
+      return [
+        ...claude,
+        ...agents,
+        ...(supplementary[0] ?? []),
+        ...direct,
+        ...supplementary.slice(1).flat(),
+        ...(yield* loadWellknown().pipe(Effect.orDie)),
+      ]
     })
 
     const initial = yield* discover()
@@ -276,16 +328,60 @@ const layer = Layer.effect(
       }
     })
 
-    yield* Stream.fromPubSub(updates).pipe(
-      Stream.debounce("100 millis"),
-      Stream.runForEach((update) =>
+    const reload = Effect.fn("Config.reload")(() =>
+      reloadLock.withPermit(
         Effect.gen(function* () {
           const next = yield* discover()
+          if (isDeepStrictEqual(configs, next)) return
           configs = next
           yield* reconcile(next)
           yield* events.publish(ConfigSchema.Event.Updated, {})
-        }).pipe(Effect.catchCause((cause) => Effect.logError("failed to reload config", { path: update.path, cause }))),
+        }),
       ),
+    )
+
+    yield* Stream.fromPubSub(updates).pipe(
+      Stream.debounce("100 millis"),
+      Stream.runForEach((update) =>
+        reload().pipe(
+          Effect.catchCause((cause) => Effect.logError("failed to reload config", { path: update.path, cause })),
+        ),
+      ),
+      Effect.forkScoped({ startImmediately: true }),
+    )
+    yield* events.subscribe(Integration.Event.ConnectionUpdated).pipe(
+      Stream.filterEffect((event) =>
+        wellknown.entries().pipe(
+          Effect.map((entries) => entries.some((entry) => entry.integrationID === event.data.integrationID)),
+          Effect.catch(() => Effect.succeed(false)),
+        ),
+      ),
+      Stream.runForEach(() =>
+        reload().pipe(Effect.catchCause((cause) => Effect.logError("failed to reload wellknown config", { cause }))),
+      ),
+      Effect.forkScoped({ startImmediately: true }),
+    )
+    yield* events.subscribe(WellKnown.Event.Updated).pipe(
+      Stream.runForEach(() =>
+        reload().pipe(Effect.catchCause((cause) => Effect.logError("failed to reload wellknown sources", { cause }))),
+      ),
+      Effect.forkScoped({ startImmediately: true }),
+    )
+    yield* Effect.sleep("10 minutes").pipe(
+      Effect.andThen(
+        Effect.suspend(() => {
+          if (!wellknown.snapshot().length) return Effect.void
+          return Effect.gen(function* () {
+            const changed = yield* wellknown.refresh().pipe(
+              Effect.catch((error) =>
+                Effect.logWarning("failed to refresh wellknown manifests", { error }).pipe(Effect.as(false)),
+              ),
+            )
+            if (!changed) yield* reload()
+          }).pipe(Effect.catchCause((cause) => Effect.logWarning("failed to refresh wellknown config", { cause })))
+        }),
+      ),
+      Effect.forever,
       Effect.forkScoped({ startImmediately: true }),
     )
     yield* reconcile(initial)
@@ -301,5 +397,5 @@ const layer = Layer.effect(
 export const node = makeLocationNode({
   service: Service,
   layer,
-  deps: [Watcher.node, EventV2.node, FSUtil.node, Global.node, Location.node],
+  deps: [Watcher.node, EventV2.node, FSUtil.node, Global.node, Location.node, Credential.node, WellKnown.node],
 })

packages/core/src/config/experimental.ts

@@ -0,0 +1,14 @@
+export * as ConfigExperimental from "./experimental"
+
+import { Schema } from "effect"
+import { NonNegativeInt } from "../schema"
+import { ConfigPolicy } from "./policy"
+
+export class Info extends Schema.Class<Info>("ConfigExperimental.Info")({
+  subagent_depth: NonNegativeInt.pipe(Schema.optional).annotate({
+    description: "Maximum subagent nesting depth. Defaults to 1.",
+  }),
+  policies: ConfigPolicy.Info.pipe(Schema.Array, Schema.optional).annotate({
+    description: "Ordered policies controlling access to configured resources",
+  }),
+}) {}

packages/core/src/config/plugin/policy.ts

@@ -0,0 +1,35 @@
+export * as ConfigPolicyPlugin from "./policy"
+
+import { define } from "@opencode-ai/plugin/v2/effect/plugin"
+import { Effect, Stream } from "effect"
+import { Config } from "../../config"
+import { Wildcard } from "../../util/wildcard"
+
+export const Plugin = define({
+  id: "opencode.config.policy",
+  effect: Effect.fn(function* (ctx) {
+    const config = yield* Config.Service
+    const loaded = { entries: yield* config.entries() }
+    yield* ctx.catalog.transform((catalog) => {
+      // User-global policy takes priority over policy authored by a repository.
+      const policies = loaded.entries
+        .filter((entry): entry is Config.Document => entry.type === "document")
+        .toReversed()
+        .flatMap((entry) => entry.info.experimental?.policies ?? [])
+      for (const record of catalog.provider.list()) {
+        const policy = policies.findLast((policy) => Wildcard.match(record.provider.id, policy.resource))
+        if (policy?.effect === "deny") catalog.provider.remove(record.provider.id)
+      }
+    })
+    yield* ctx.event.subscribe().pipe(
+      Stream.filter((event) => event.type === "config.updated"),
+      Stream.runForEach(() =>
+        config.entries().pipe(
+          Effect.tap((entries) => Effect.sync(() => (loaded.entries = entries))),
+          Effect.andThen(ctx.catalog.reload()),
+        ),
+      ),
+      Effect.forkScoped({ startImmediately: true }),
+    )
+  }),
+})

packages/core/src/config/policy.ts

@@ -0,0 +1,13 @@
+export * as ConfigPolicy from "./policy"
+
+import { Schema } from "effect"
+
+export const Effect = Schema.Literals(["allow", "deny"])
+export type Effect = typeof Effect.Type
+
+export const Info = Schema.Struct({
+  action: Schema.Literal("provider.use"),
+  resource: Schema.String,
+  effect: Effect,
+})
+export type Info = typeof Info.Type

packages/core/src/database/migration.gen.ts

@@ -54,5 +54,6 @@ export const migrations = (
     import("./migration/20260709163752_time_suspended"),
     import("./migration/20260709190621_session_pending_table"),
     import("./migration/20260710025429_instruction_sync"),
+    import("./migration/20260716020354_kv"),
   ])
 ).map((module) => module.default) satisfies DatabaseMigration.Migration[]

packages/core/src/database/migration/20260716020354_kv.ts

@@ -0,0 +1,18 @@
+import { Effect } from "effect"
+import type { DatabaseMigration } from "../migration"
+
+export default {
+  id: "20260716020354_kv",
+  up(tx) {
+    return Effect.gen(function* () {
+      yield* tx.run(`
+        CREATE TABLE \`kv\` (
+          \`key\` text PRIMARY KEY,
+          \`value\` text NOT NULL,
+          \`time_created\` integer NOT NULL,
+          \`time_updated\` integer NOT NULL
+        );
+      `)
+    })
+  },
+} satisfies DatabaseMigration.Migration

packages/core/src/database/schema.gen.ts

@@ -87,6 +87,14 @@ export default {
           CONSTRAINT \`fk_event_aggregate_id_event_sequence_aggregate_id_fk\` FOREIGN KEY (\`aggregate_id\`) REFERENCES \`event_sequence\`(\`aggregate_id\`) ON DELETE CASCADE
         );
       `)
+      yield* tx.run(`
+        CREATE TABLE \`kv\` (
+          \`key\` text PRIMARY KEY,
+          \`value\` text NOT NULL,
+          \`time_created\` integer NOT NULL,
+          \`time_updated\` integer NOT NULL
+        );
+      `)
       yield* tx.run(`
         CREATE TABLE \`permission\` (
           \`id\` text PRIMARY KEY,

packages/core/src/integration.ts

@@ -8,6 +8,7 @@ import {
   Duration,
   Effect,
   Exit,
+  Fiber,
   Layer,
   Schedule,
   Schema,
@@ -617,34 +618,40 @@ const layer = Layer.effect(
         }),
       )
 
-      yield* processes
-        .runStream(
+      yield* Effect.gen(function* () {
+        const handle = yield* processes.spawn(
           ChildProcess.make(method.command[0], method.command.slice(1), {
             extendEnv: true,
             stdin: "ignore",
           }),
-          { okExitCodes: [0] },
         )
-        .pipe(
-          Stream.tap((line) =>
+        const stdout = yield* AppProcess.collectStream(handle.stdout, undefined).pipe(Effect.forkScoped)
+        yield* handle.stderr.pipe(
+          Stream.decodeText,
+          Stream.tap((chunk) =>
             SynchronizedRef.update(commandAttempts, (current) => {
               const attempt = current.get(attemptID)
               if (!attempt || attempt.status !== "pending") return current
-              const message = attempt.message ? `${attempt.message}\n${line}` : line
+              const message = (attempt.message ?? "") + chunk
               return new Map(current).set(attemptID, { ...attempt, message })
             }),
           ),
-          Stream.runCollect,
-          Effect.flatMap((lines) => {
-            const credential = Array.from(lines).at(-1)
-            return credential
-              ? Effect.succeed(credential)
-              : Effect.fail(new Error("Authentication command returned no credential"))
-          }),
-          Effect.exit,
-          Effect.flatMap((exit) => settleCommand(attemptID, exit)),
-          Effect.forkIn(attemptScope, { startImmediately: true }),
+          Stream.runDrain,
         )
+        const exitCode = yield* handle.exitCode
+        if (exitCode !== 0) {
+          const attempt = (yield* SynchronizedRef.get(commandAttempts)).get(attemptID)
+          return yield* Effect.fail(new Error(attempt?.message?.trim() || `Authentication command exited ${exitCode}`))
+        }
+        const credential = (yield* Fiber.join(stdout)).buffer.toString("utf8").trim()
+        if (!credential) return yield* Effect.fail(new Error("Authentication command returned no credential"))
+        return credential
+      }).pipe(
+        Scope.provide(attemptScope),
+        Effect.exit,
+        Effect.flatMap((exit) => settleCommand(attemptID, exit)),
+        Effect.forkIn(attemptScope, { startImmediately: true }),
+      )
 
       return CommandAttempt.make({ attemptID, time })
     })

packages/core/src/kv.ts

@@ -0,0 +1,47 @@
+export * as KV from "./kv"
+
+import { eq } from "drizzle-orm"
+import { Context, Effect, Layer, Schema } from "effect"
+import { Database } from "./database/database"
+import { makeGlobalNode } from "./effect/app-node"
+import { KVTable } from "./kv/sql"
+
+export type Value = Schema.Json
+
+export interface Interface {
+  readonly get: (key: string) => Effect.Effect<Value | undefined>
+  readonly set: (key: string, value: Value) => Effect.Effect<void>
+  readonly remove: (key: string) => Effect.Effect<void>
+}
+
+export class Service extends Context.Service<Service, Interface>()("@opencode/v2/KV") {}
+
+const layer = Layer.effect(
+  Service,
+  Effect.gen(function* () {
+    const { db } = yield* Database.Service
+    return Service.of({
+      get: Effect.fn("KV.get")(function* (key) {
+        return (yield* db
+          .select({ value: KVTable.value })
+          .from(KVTable)
+          .where(eq(KVTable.key, key))
+          .get()
+          .pipe(Effect.orDie))?.value
+      }),
+      set: Effect.fn("KV.set")(function* (key, value) {
+        yield* db
+          .insert(KVTable)
+          .values({ key, value })
+          .onConflictDoUpdate({ target: KVTable.key, set: { value, time_updated: Date.now() } })
+          .run()
+          .pipe(Effect.orDie)
+      }),
+      remove: Effect.fn("KV.remove")(function* (key) {
+        yield* db.delete(KVTable).where(eq(KVTable.key, key)).run().pipe(Effect.orDie)
+      }),
+    })
+  }),
+)
+
+export const node = makeGlobalNode({ service: Service, layer, deps: [Database.node] })

packages/core/src/kv/sql.ts

@@ -0,0 +1,9 @@
+import { sqliteTable, text } from "drizzle-orm/sqlite-core"
+import { Timestamps } from "../database/schema.sql"
+import type { KV } from "../kv"
+
+export const KVTable = sqliteTable("kv", {
+  key: text().primaryKey(),
+  value: text({ mode: "json" }).$type<KV.Value>().notNull(),
+  ...Timestamps,
+})

packages/core/src/location-services.ts

@@ -27,18 +27,19 @@ import { Pty } from "./pty"
 import { QuestionV2 } from "./question"
 import { Shell } from "./shell"
 import { Reference } from "./reference"
-import { ReferenceGuidance } from "./reference/guidance"
+import { ReferenceInstructions } from "./reference/instructions"
 import { SessionRunnerLLM } from "./session/runner/llm"
 import { SessionRunnerModel } from "./session/runner/model"
 import { SessionCompaction } from "./session/compaction"
 import { SessionTitle } from "./session/title"
 import { SkillV2 } from "./skill"
-import { SkillGuidance } from "./skill/guidance"
+import { SkillInstructions } from "./skill/instructions"
 import { Snapshot } from "./snapshot"
 import { InstructionDiscovery } from "./instruction-discovery"
 import { InstructionBuiltIns } from "./instructions/builtins"
 import { InstructionEntry } from "./session/instruction-entry"
 import { SessionInstructions } from "./session/instructions"
+import { SessionGenerateNode } from "./session/generate-node"
 import { McpTool } from "./tool/mcp"
 import { ReadToolFileSystem } from "./tool/read-filesystem"
 import { ToolRegistry } from "./tool/registry"
@@ -75,12 +76,13 @@ const locationServiceNodes = [
   ToolRegistry.node,
   ToolRegistry.toolsNode,
   Image.node,
-  SkillGuidance.node,
-  ReferenceGuidance.node,
+  SkillInstructions.node,
+  ReferenceInstructions.node,
   InstructionEntry.node,
   Form.node,
   QuestionV2.node,
   Generate.node,
+  SessionGenerateNode.node,
   ReadToolFileSystem.node,
   McpTool.node,
   SessionInstructions.node,

packages/core/src/mcp/instructions.ts

@@ -1,4 +1,4 @@
-export * as McpGuidance from "./guidance"
+export * as McpInstructions from "./instructions"
 
 import { makeLocationNode } from "../effect/app-node"
 import { Context, Effect, Layer, Schema } from "effect"
@@ -54,7 +54,7 @@ export interface Interface {
   readonly load: (agent: AgentV2.Selection) => Effect.Effect<Instructions.Instructions>
 }
 
-export class Service extends Context.Service<Service, Interface>()("@opencode/v2/McpGuidance") {}
+export class Service extends Context.Service<Service, Interface>()("@opencode/v2/McpInstructions") {}
 
 export const layer = Layer.effect(
   Service,
@@ -62,7 +62,7 @@ export const layer = Layer.effect(
     const mcp = yield* MCP.Service
 
     return Service.of({
-      load: Effect.fn("McpGuidance.load")(function* (selection) {
+      load: Effect.fn("McpInstructions.load")(function* (selection) {
         const agent = selection.info
         if (!agent) return Instructions.empty
         const source = (value: ReadonlyArray<Summary> | Instructions.Removed) =>

packages/core/src/plugin/host.ts

@@ -425,6 +425,7 @@ export const make = Effect.fn("PluginHost.make")(function* (plugin: PluginV2.Int
       get: (input) => runtime.session.get(input.sessionID),
       prompt: runtime.session.prompt,
       command: runtime.session.command,
+      synthetic: runtime.session.synthetic,
       interrupt: (input) => runtime.session.interrupt(input.sessionID),
     },
   } satisfies Plugin.Context

packages/core/src/plugin/internal.ts

@@ -10,6 +10,7 @@ import { Config } from "../config"
 import { ConfigAgentPlugin } from "../config/plugin/agent"
 import { ConfigCommandPlugin } from "../config/plugin/command"
 import { ConfigProviderPlugin } from "../config/plugin/provider"
+import { ConfigPolicyPlugin } from "../config/plugin/policy"
 import { ConfigReferencePlugin } from "../config/plugin/reference"
 import { ConfigSkillPlugin } from "../config/plugin/skill"
 import { EventV2 } from "../event"
@@ -43,6 +44,7 @@ import { SubagentTool } from "../tool/subagent"
 import { Tools } from "../tool/tools"
 import { WebFetchTool } from "../tool/webfetch"
 import { WebSearchTool } from "../tool/websearch"
+import { WellKnown } from "../wellknown"
 import { WriteTool } from "../tool/write"
 import { AgentPlugin } from "./agent"
 import { CommandPlugin } from "./command"
@@ -52,6 +54,7 @@ import { PluginRuntime } from "./runtime"
 import { SkillPlugin } from "./skill"
 import { SystemPromptPlugin } from "./system-prompt"
 import { VariantPlugin } from "./variant"
+import { WellKnownPlugin } from "../wellknown/plugin"
 
 const services = Effect.fn("PluginInternal.services")(function* () {
   const agent = yield* AgentV2.Service
@@ -81,6 +84,7 @@ const services = Effect.fn("PluginInternal.services")(function* () {
   const skill = yield* SkillV2.Service
   const tools = yield* Tools.Service
   const websearch = yield* WebSearchTool.ConfigService
+  const wellknown = yield* WellKnown.Service
   return Context.mergeAll(
     Context.make(AgentV2.Service, agent),
     Context.make(Catalog.Service, catalog),
@@ -109,6 +113,7 @@ const services = Effect.fn("PluginInternal.services")(function* () {
     Context.make(SkillV2.Service, skill),
     Context.make(Tools.Service, tools),
     Context.make(WebSearchTool.ConfigService, websearch),
+    Context.make(WellKnown.Service, wellknown),
   )
 })
 
@@ -119,6 +124,7 @@ export type Requirements = ContextServices<Effect.Success<ReturnType<typeof serv
 export type InternalPlugin = Plugin<Requirements | Scope.Scope>
 
 const pre = [
+  WellKnownPlugin.Plugin,
   AgentPlugin.Plugin,
   CommandPlugin.Plugin,
   SkillPlugin.Plugin,
@@ -146,6 +152,7 @@ const post = [
   ConfigSkillPlugin.Plugin,
   ConfigProviderPlugin.Plugin,
   VariantPlugin.Plugin,
+  ConfigPolicyPlugin.Plugin,
 ] as const satisfies readonly InternalPlugin[]
 
 export const list = Effect.fn("PluginInternal.list")(function* () {

packages/core/src/plugin/promise.ts

@@ -244,6 +244,17 @@ export function fromPromise(plugin: Plugin) {
                   resume: input.resume ?? undefined,
                 }),
               ),
+            synthetic: (input) =>
+              run(
+                host.session.synthetic({
+                  ...input,
+                  sessionID: Session.ID.make(input.sessionID),
+                  id: input.id == null ? undefined : SessionMessage.ID.make(input.id),
+                  description: input.description ?? undefined,
+                  delivery: input.delivery ?? undefined,
+                  resume: input.resume ?? undefined,
+                }),
+              ),
             interrupt: (input) => run(host.session.interrupt({ sessionID: Session.ID.make(input.sessionID) })),
           },
         }

packages/core/src/plugin/supervisor.ts

@@ -35,6 +35,7 @@ import { SkillV2 } from "../skill"
 import { ReadToolFileSystem } from "../tool/read-filesystem"
 import { ToolRegistry } from "../tool/registry"
 import { WebSearchTool } from "../tool/websearch"
+import { WellKnown } from "../wellknown"
 import { PluginInternal } from "./internal"
 import { PluginRuntime } from "./runtime"
 import { SdkPlugins } from "./sdk"
@@ -163,9 +164,7 @@ const load = Effect.fn("PluginSupervisor.load")(function* (operation: Extract<Op
   if (!entrypoint) return
   // Bun currently ignores query parameters when caching file:// imports.
   const source =
-    operation.mtime === undefined
-      ? entrypoint
-      : `${operation.target.replaceAll("\\", "/")}?mtime=${operation.mtime}`
+    operation.mtime === undefined ? entrypoint : `${operation.target.replaceAll("\\", "/")}?mtime=${operation.mtime}`
   yield* Effect.log({ msg: "loading plugin", id: operation.target, entrypoint: source })
   const mod = yield* Effect.promise(() => import(source))
   const value = (yield* Schema.decodeUnknownEffect(PluginModule)(mod)).default
@@ -296,6 +295,7 @@ export const node = makeLocationNode({
     SkillV2.node,
     ToolRegistry.toolsNode,
     WebSearchTool.configNode,
+    WellKnown.node,
   ],
 })
 

packages/core/src/reference/instructions.ts

@@ -1,4 +1,4 @@
-export * as ReferenceGuidance from "./guidance"
+export * as ReferenceInstructions from "./instructions"
 
 import { makeLocationNode } from "../effect/app-node"
 import { Context, Effect, Layer, Schema } from "effect"
@@ -57,7 +57,7 @@ export interface Interface {
   readonly load: () => Effect.Effect<Instructions.Instructions>
 }
 
-export class Service extends Context.Service<Service, Interface>()("@opencode/v2/ReferenceGuidance") {}
+export class Service extends Context.Service<Service, Interface>()("@opencode/v2/ReferenceInstructions") {}
 
 const layer = Layer.effect(
   Service,
@@ -65,7 +65,7 @@ const layer = Layer.effect(
     const references = yield* Reference.Service
 
     return Service.of({
-      load: Effect.fn("ReferenceGuidance.load")(function* () {
+      load: Effect.fn("ReferenceInstructions.load")(function* () {
         const available = (yield* references.list())
           .filter((reference) => reference.description !== undefined)
           .map((reference) => ({

packages/core/src/session.ts

@@ -33,10 +33,12 @@ import { LocationServiceMap } from "./location-service-map"
 import { MessageDecodeError } from "./session/error"
 import { SessionEvent } from "./session/event"
 import { SessionPending } from "./session/pending"
+import { SessionGenerate } from "./session/generate"
 import { Snapshot } from "./snapshot"
 import { SessionRevert } from "./session/revert"
 import { Session } from "@opencode-ai/schema/session"
 import { FSUtil } from "./fs-util"
+import { Image } from "./image"
 import { Mime } from "./mime"
 import type { EventLog } from "@opencode-ai/schema/event-log"
 import { SkillV2 } from "./skill"
@@ -175,6 +177,7 @@ export type Error =
   | CommandV2.NotFoundError
   | CommandV2.EvaluationError
   | MessageNotFoundError
+  | SessionGenerate.Error
 
 export interface Interface {
   readonly list: (input?: ListInput) => Effect.Effect<{
@@ -243,6 +246,11 @@ export interface Interface {
     delivery?: SessionPending.Delivery
     resume?: boolean
   }) => Effect.Effect<SessionPending.User, NotFoundError | PromptConflictError | AttachmentError>
+  /** Generates text from current Session context without admitting input or mutating history. */
+  readonly generate: (input: {
+    sessionID: SessionSchema.ID
+    prompt: string
+  }) => Effect.Effect<string, NotFoundError | SessionGenerate.Error>
   readonly command: (input: {
     id?: SessionMessage.ID
     sessionID: SessionSchema.ID
@@ -531,9 +539,13 @@ const layer = Layer.effect(
             // continues from the reverted boundary rather than stale post-boundary history.
             if (session.revert)
               yield* SessionRevert.commit(session).pipe(Effect.provideService(EventV2.Service, events))
-            const prompt = yield* resolvePrompt({ text: input.text, files: input.files, agents: input.agents }).pipe(
-              Effect.provideService(FSUtil.Service, fs),
-            )
+            // Resolved lazily so prompt admission only boots location services when an
+            // image attachment actually needs the resizer.
+            const image = Image.Service.pipe(Effect.provide(locations.get(session.location)))
+            const prompt = yield* resolvePrompt(
+              { text: input.text, files: input.files, agents: input.agents },
+              image,
+            ).pipe(Effect.provideService(FSUtil.Service, fs))
             const messageID = input.id ?? SessionMessage.ID.create()
             const admittedInput = SessionPending.Message.make({
               type: "user",
@@ -564,6 +576,11 @@ const layer = Layer.effect(
           }),
         ),
       ),
+      generate: Effect.fn("V2Session.generate")(function* (input) {
+        const session = yield* result.get(input.sessionID)
+        const generate = yield* SessionGenerate.Service.pipe(Effect.provide(locations.get(session.location)))
+        return yield* generate.generate(input)
+      }),
       command: Effect.fn("V2Session.command")(function* (input) {
         const session = yield* result.get(input.sessionID)
         const commands = yield* CommandV2.Service.pipe(Effect.provide(locations.get(session.location)))
@@ -859,10 +876,13 @@ function synthesizeTerminalShellInfo(started: ShellSchema.Info): ShellSchema.Inf
   }
 }
 
-const resolvePrompt = Effect.fn("V2Session.resolvePrompt")(function* (input: PromptInput.Prompt) {
+const resolvePrompt = Effect.fn("V2Session.resolvePrompt")(function* (
+  input: PromptInput.Prompt,
+  image: Effect.Effect<Image.Interface>,
+) {
   const fs = yield* FSUtil.Service
   const files = input.files
-    ? yield* Effect.forEach(input.files, (file) => materializeAttachment(fs, file), { concurrency: 8 })
+    ? yield* Effect.forEach(input.files, (file) => materializeAttachment(fs, file, image), { concurrency: 8 })
     : undefined
   return Prompt.make({ text: input.text, agents: input.agents, files })
 })
@@ -872,6 +892,7 @@ const MAX_ATTACHMENT_BYTES = 20 * 1024 * 1024
 const materializeAttachment = Effect.fn("V2Session.materializeAttachment")(function* (
   fs: FSUtil.Interface,
   input: PromptInput.FileAttachment,
+  image: Effect.Effect<Image.Interface>,
 ) {
   const resolved = input.uri.startsWith("data:")
     ? {
@@ -900,9 +921,15 @@ const materializeAttachment = Effect.fn("V2Session.materializeAttachment")(funct
             .join("\n"),
         )
       : resolved.bytes
-  return FileAttachment.create({
-    data: Base64.make(Buffer.from(content).toString("base64")),
+  const normalized = yield* normalizeImageAttachment(
+    input,
+    Base64.make(Buffer.from(content).toString("base64")),
     mime,
+    image,
+  )
+  return FileAttachment.create({
+    data: normalized.data,
+    mime: normalized.mime,
     source: resolved.source,
     name: input.name ?? resolved.name,
     description: input.description,
@@ -910,6 +937,23 @@ const materializeAttachment = Effect.fn("V2Session.materializeAttachment")(funct
   })
 })
 
+const normalizeImageAttachment = Effect.fn("V2Session.normalizeImageAttachment")(function* (
+  input: PromptInput.FileAttachment,
+  data: Base64,
+  mime: string,
+  image: Effect.Effect<Image.Interface>,
+) {
+  if (!mime.startsWith("image/")) return { data, mime }
+  const service = yield* image
+  const label = input.name ?? (input.uri.startsWith("data:") ? "inline attachment" : input.uri)
+  const content = { uri: label, content: data, encoding: "base64" as const, mime }
+  const normalized = yield* service.normalize(label, content).pipe(
+    Effect.catchTag("Image.ResizerUnavailableError", () => Effect.succeed(content)),
+    Effect.mapError((error) => new AttachmentError({ uri: label, message: error.message })),
+  )
+  return { data: Base64.make(normalized.content), mime: normalized.mime }
+})
+
 const readFileAttachment = Effect.fn("V2Session.readFileAttachment")(function* (fs: FSUtil.Interface, uri: string) {
   const url = yield* Effect.try({
     try: () => new URL(uri),

packages/core/src/session/context.ts

@@ -8,10 +8,10 @@ import { InstructionDiscovery } from "../instruction-discovery"
 import { Instructions } from "../instructions/index"
 import { InstructionBuiltIns } from "../instructions/builtins"
 import { Location } from "../location"
-import { McpGuidance } from "../mcp/guidance"
+import { McpInstructions } from "../mcp/instructions"
 import { PluginSupervisor } from "../plugin/supervisor"
-import { ReferenceGuidance } from "../reference/guidance"
-import { SkillGuidance } from "../skill/guidance"
+import { ReferenceInstructions } from "../reference/instructions"
+import { SkillInstructions } from "../skill/instructions"
 import { AgentNotFoundError } from "./error"
 import { SessionHistory } from "./history"
 import { InstructionEntry } from "./instruction-entry"
@@ -34,13 +34,19 @@ export interface Loaded {
   readonly messages: ReadonlyArray<SessionMessage.Info>
 }
 
+/**
+ * Resolves model-request state in two phases: `select` fixes the Session,
+ * agent, and instruction sources; `load` adds the model and active history for
+ * that selection. This module does not build or execute the model request.
+ */
 export interface Interface {
-  /** Resolves the Session, selected agent, and current instruction sources before durable Step preparation. */
+  /** Selects the Session, agent, and instruction sources used by subsequent work. */
   readonly select: (sessionID: SessionSchema.ID) => Effect.Effect<Selection, AgentNotFoundError>
-  /** Loads the selected model and active history after instruction sync and pending-input promotion. */
+  /** Resolves the model and active history for that selection. */
   readonly load: (selection: Selection) => Effect.Effect<Loaded, SessionRunnerModel.Error>
 }
 
+/** Location-scoped model-context loader for durable Session Steps. */
 export class Service extends Context.Service<Service, Interface>()("@opencode/v2/SessionContext") {}
 
 const layer = Layer.effect(
@@ -52,11 +58,11 @@ const layer = Layer.effect(
     const discovery = yield* InstructionDiscovery.Service
     const entries = yield* InstructionEntry.Service
     const location = yield* Location.Service
-    const mcpGuidance = yield* McpGuidance.Service
+    const mcpInstructions = yield* McpInstructions.Service
     const models = yield* SessionRunnerModel.Service
     const plugins = yield* PluginSupervisor.Service
-    const referenceGuidance = yield* ReferenceGuidance.Service
-    const skillGuidance = yield* SkillGuidance.Service
+    const referenceInstructions = yield* ReferenceInstructions.Service
+    const skillInstructions = yield* SkillInstructions.Service
     const store = yield* SessionStore.Service
 
     const select = Effect.fn("SessionContext.select")(function* (sessionID: SessionSchema.ID) {
@@ -72,9 +78,9 @@ const layer = Layer.effect(
         [
           builtins.load(sessionID),
           discovery.load(),
-          skillGuidance.load(agent),
-          referenceGuidance.load(),
-          mcpGuidance.load(agent),
+          skillInstructions.load(agent),
+          referenceInstructions.load(),
+          mcpInstructions.load(agent),
           entries.load(sessionID),
         ],
         { concurrency: "unbounded" },
@@ -108,11 +114,11 @@ export const node = makeLocationNode({
     InstructionDiscovery.node,
     InstructionEntry.node,
     Location.node,
-    McpGuidance.node,
+    McpInstructions.node,
     PluginSupervisor.node,
-    ReferenceGuidance.node,
+    ReferenceInstructions.node,
     SessionRunnerModel.node,
     SessionStore.node,
-    SkillGuidance.node,
+    SkillInstructions.node,
   ],
 })

packages/core/src/session/generate-node.ts

@@ -0,0 +1,69 @@
+export * as SessionGenerateNode from "./generate-node"
+
+import { LLM, LLMClient, Message, SystemPart } from "@opencode-ai/ai"
+import { Effect, Layer } from "effect"
+import { Database } from "../database/database"
+import { makeLocationNode } from "../effect/app-node"
+import { llmClient } from "../effect/app-node-platform"
+import { PluginHooks } from "../plugin/hooks"
+import { SessionContext } from "./context"
+import { SessionGenerate } from "./generate"
+import { SessionHistory } from "./history"
+import { SessionModelHeaders } from "./model-headers"
+import { SessionRunnerModel } from "./runner/model"
+import PROMPT_DEFAULT from "./runner/prompt/base.txt"
+import { toLLMMessages } from "./runner/to-llm-message"
+
+const layer = Layer.effect(
+  SessionGenerate.Service,
+  Effect.gen(function* () {
+    const context = yield* SessionContext.Service
+    const database = yield* Database.Service
+    const hooks = yield* PluginHooks.Service
+    const llm = yield* LLMClient.Service
+    const models = yield* SessionRunnerModel.Service
+
+    return SessionGenerate.Service.of({
+      generate: Effect.fn("SessionGenerate.generate")(function* (input) {
+        const selection = yield* context.select(input.sessionID)
+        const model = yield* models.resolve(selection.session)
+        const history = yield* SessionHistory.preview(database.db, selection.session.id, selection.instructions)
+        const providerMetadataKey = model.model.route.providerMetadataKey ?? model.model.provider
+        const promptCacheKey = /^ses_[0-9a-f]{64}$/.test(selection.session.id)
+          ? selection.session.id.slice(4)
+          : selection.session.id
+        const contextEvent = yield* hooks.trigger("session", "context", {
+          sessionID: selection.session.id,
+          agent: selection.agent.id,
+          model: model.ref,
+          system: [selection.agent.info.system ? selection.agent.info.system : PROMPT_DEFAULT, history.initial]
+            .filter((part) => part.length > 0)
+            .map(SystemPart.make),
+          messages: [
+            ...toLLMMessages(history.messages, model.ref, providerMetadataKey),
+            ...(history.instructionUpdate ? [Message.system(history.instructionUpdate)] : []),
+            Message.user(input.prompt),
+          ],
+          tools: {},
+        })
+        return (yield* llm.generate(
+          LLM.request({
+            model: model.model,
+            http: { headers: SessionModelHeaders.make(selection.session) },
+            providerOptions: { openai: { promptCacheKey } },
+            system: contextEvent.system,
+            messages: contextEvent.messages,
+            tools: [],
+            toolChoice: "none",
+          }),
+        )).text
+      }),
+    })
+  }),
+)
+
+export const node = makeLocationNode({
+  service: SessionGenerate.Service,
+  layer,
+  deps: [SessionContext.node, Database.node, PluginHooks.node, SessionRunnerModel.node, llmClient],
+})

packages/core/src/session/generate.ts

@@ -0,0 +1,21 @@
+export * as SessionGenerate from "./generate"
+
+import type { LLMError } from "@opencode-ai/ai"
+import { Context, type Effect } from "effect"
+import type { Instructions } from "../instructions"
+import type { AgentNotFoundError } from "./error"
+import type { SessionRunnerModel } from "./runner/model"
+import type { SessionSchema } from "./schema"
+
+export type Error = AgentNotFoundError | Instructions.InitializationBlocked | SessionRunnerModel.Error | LLMError
+
+export interface Interface {
+  /** Generates text from current Session context without mutating the Session. */
+  readonly generate: (input: {
+    readonly sessionID: SessionSchema.ID
+    readonly prompt: string
+  }) => Effect.Effect<string, Error>
+}
+
+/** Location-scoped transient generation from Session context. */
+export class Service extends Context.Service<Service, Interface>()("@opencode/v2/SessionGenerate") {}

packages/core/src/session/history.ts

@@ -60,13 +60,17 @@ const decodeMessageRow = (row: typeof SessionMessageTable.$inferSelect) =>
     ),
   )
 
-export const load = Effect.fn("SessionHistory.load")(function* (db: DatabaseService, sessionID: SessionSchema.ID) {
-  return yield* Effect.forEach(
-    yield* messageRows(db, sessionID, yield* latestCompaction(db, sessionID)),
-    decodeMessageRow,
+const messageEntries = Effect.fnUntraced(function* (db: DatabaseService, sessionID: SessionSchema.ID) {
+  const rows = yield* messageRows(db, sessionID, yield* latestCompaction(db, sessionID))
+  return yield* Effect.forEach(rows, (row) =>
+    decodeMessageRow(row).pipe(Effect.map((message) => ({ seq: row.seq, message }))),
   )
 })
 
+export const load = Effect.fn("SessionHistory.load")(function* (db: DatabaseService, sessionID: SessionSchema.ID) {
+  return (yield* messageEntries(db, sessionID)).map((entry) => entry.message)
+})
+
 export const entriesForRunner = Effect.fn("SessionHistory.entriesForRunner")(function* (
   db: DatabaseService,
   sessionID: SessionSchema.ID,
@@ -75,10 +79,7 @@ export const entriesForRunner = Effect.fn("SessionHistory.entriesForRunner")(fun
   return yield* db
     .transaction(() =>
       Effect.gen(function* () {
-        const rows = yield* messageRows(db, sessionID, yield* latestCompaction(db, sessionID))
-        const messages = yield* Effect.forEach(rows, (row) =>
-          decodeMessageRow(row).pipe(Effect.map((message) => ({ seq: row.seq, message }))),
-        )
+        const messages = yield* messageEntries(db, sessionID)
         const assembled = yield* InstructionState.assemble(db, sessionID, instructions)
         return {
           initial: assembled.initial,
@@ -89,6 +90,33 @@ export const entriesForRunner = Effect.fn("SessionHistory.entriesForRunner")(fun
     .pipe(Effect.orDie)
 })
 
+export const preview = Effect.fn("SessionHistory.preview")(function* (
+  db: DatabaseService,
+  sessionID: SessionSchema.ID,
+  instructions: Instructions.Instructions,
+) {
+  const observed = yield* Instructions.read(instructions)
+  return yield* db
+    .transaction(() =>
+      Effect.gen(function* () {
+        const messages = yield* messageEntries(db, sessionID)
+        // An active assistant may contain an unresolved tool call, so only preview the settled prefix.
+        const unsettled = messages.findIndex(
+          (entry) => entry.message.type === "assistant" && entry.message.time.completed === undefined,
+        )
+        const settled = unsettled === -1 ? messages : messages.slice(0, unsettled)
+        const assembled = yield* InstructionState.preview(db, sessionID, instructions, observed)
+        const entries = [...settled, ...assembled.updates].toSorted((a, b) => a.seq - b.seq)
+        return {
+          initial: assembled.initial,
+          messages: entries.map((entry) => entry.message),
+          instructionUpdate: assembled.update,
+        }
+      }),
+    )
+    .pipe(Effect.catch((error) => (error instanceof Instructions.InitializationBlocked ? error : Effect.die(error))))
+})
+
 /** Returns the session's sole user message, or `undefined` once a second one exists. */
 export const firstUserMessageIfOnly = Effect.fn("SessionHistory.firstUserMessageIfOnly")(function* (
   db: DatabaseService,

packages/core/src/session/instruction-state.ts

@@ -15,27 +15,52 @@ type DatabaseService = Database.Interface["db"]
 
 const decodeInstructionsUpdated = Schema.decodeUnknownSync(SessionEvent.InstructionsUpdated.data)
 
+export interface Observation extends Instructions.Admission {
+  readonly sessionID: SessionSchema.ID
+  readonly initial: boolean
+  readonly current: Instructions.Values
+}
+
+export const observe = Effect.fn("InstructionState.observe")(function* (
+  db: DatabaseService,
+  instructions: Instructions.Instructions,
+  sessionID: SessionSchema.ID,
+): Effect.fn.Return<Observation, Instructions.InitializationBlocked> {
+  const [observed, stored] = yield* Effect.all([Instructions.read(instructions), ensure(db, sessionID)], {
+    concurrency: "unbounded",
+  })
+  const result = yield* observeAgainst(observed, stored?.current_values)
+  return {
+    sessionID,
+    initial: !stored,
+    ...result,
+  }
+})
+
+export const commit = Effect.fn("InstructionState.commit")(function* (
+  db: DatabaseService,
+  events: EventV2.Interface,
+  observation: Observation,
+) {
+  if (!observation.initial && Object.keys(observation.delta).length === 0) return
+  yield* events.publish(
+    SessionEvent.InstructionsUpdated,
+    { sessionID: observation.sessionID, delta: observation.delta },
+    {
+      // Initial sync establishes the baseline; unlike later deltas it is not chronological history.
+      ...(observation.initial ? { metadata: { instructions: { initial: true } } } : {}),
+      commit: () => insertBlobs(db, observation.blobs),
+    },
+  )
+})
+
 export const prepare = Effect.fn("InstructionState.prepare")(function* (
   db: DatabaseService,
   events: EventV2.Interface,
   instructions: Instructions.Instructions,
   sessionID: SessionSchema.ID,
 ) {
-  const [observed, stored] = yield* Effect.all([Instructions.read(instructions), ensure(db, sessionID)], {
-    concurrency: "unbounded",
-  })
-  const admission = yield* Instructions.diff(observed, stored?.current_values)
-  if (!stored || Object.keys(admission.delta).length > 0) {
-    yield* events.publish(
-      SessionEvent.InstructionsUpdated,
-      { sessionID, delta: admission.delta },
-      {
-        // Initial sync establishes the baseline; unlike later deltas it is not chronological history.
-        ...(!stored ? { metadata: { instructions: { initial: true } } } : {}),
-        commit: () => insertBlobs(db, admission.blobs),
-      },
-    )
-  }
+  yield* commit(db, events, yield* observe(db, instructions, sessionID))
 })
 
 export const apply = Effect.fn("InstructionState.apply")(function* (
@@ -97,28 +122,21 @@ export const rebuild = Effect.fn("InstructionState.rebuild")(function* (
   db: DatabaseService,
   sessionID: SessionSchema.ID,
 ) {
-  const folded = fold(yield* instructionEvents(db, sessionID))
-  if (!folded) {
+  const state = yield* stateFromEvents(db, sessionID)
+  if (!state) {
     yield* reset(db, sessionID)
     return undefined
   }
-  const state = {
-    session_id: sessionID,
-    epoch_start: folded.epochStart,
-    through_seq: folded.throughSeq,
-    initial_values: folded.initial,
-    current_values: folded.current,
-  }
   yield* db
     .insert(InstructionStateTable)
     .values(state)
     .onConflictDoUpdate({
       target: InstructionStateTable.session_id,
       set: {
-        epoch_start: folded.epochStart,
-        through_seq: folded.throughSeq,
-        initial_values: folded.initial,
-        current_values: folded.current,
+        epoch_start: state.epoch_start,
+        through_seq: state.through_seq,
+        initial_values: state.initial_values,
+        current_values: state.current_values,
       },
     })
     .run()
@@ -126,13 +144,12 @@ export const rebuild = Effect.fn("InstructionState.rebuild")(function* (
   return state
 })
 
-export const assemble = Effect.fn("InstructionState.assemble")(function* (
+const assembleState = Effect.fnUntraced(function* (
   db: DatabaseService,
   sessionID: SessionSchema.ID,
   instructions: Instructions.Instructions,
+  state: typeof InstructionStateTable.$inferSelect,
 ) {
-  const state = yield* find(db, sessionID)
-  if (!state) return yield* Effect.die(new Error(`Instruction state not found during assembly: ${sessionID}`))
   const rows = yield* instructionUpdatesAfter(db, sessionID, state.epoch_start)
   const updates = rows.map((row) => ({
     row,
@@ -162,7 +179,49 @@ export const assemble = Effect.fn("InstructionState.assemble")(function* (
       })
     values = Instructions.applyDelta(values, delta)
   }
-  return { initial: Instructions.renderInitial(instructions, valuesAtStart), updates: result }
+  return { initial: Instructions.renderInitial(instructions, valuesAtStart), updates: result, current: values }
+})
+
+export const assemble = Effect.fn("InstructionState.assemble")(function* (
+  db: DatabaseService,
+  sessionID: SessionSchema.ID,
+  instructions: Instructions.Instructions,
+) {
+  const state = yield* find(db, sessionID)
+  if (!state) return yield* Effect.die(new Error(`Instruction state not found during assembly: ${sessionID}`))
+  const assembled = yield* assembleState(db, sessionID, instructions, state)
+  return { initial: assembled.initial, updates: assembled.updates }
+})
+
+export const preview = Effect.fn("InstructionState.preview")(function* (
+  db: DatabaseService,
+  sessionID: SessionSchema.ID,
+  instructions: Instructions.Instructions,
+  observed: Instructions.ReadResult,
+) {
+  const state = yield* readState(db, sessionID)
+  const result = yield* observeAgainst(observed, state?.current_values)
+  const blobs = new Map<Instructions.Hash, Schema.Json>(
+    Object.entries(result.blobs).map(([hash, value]) => [Instructions.Hash.make(hash), value]),
+  )
+  if (!state) {
+    const values = dereference(result.current, blobs)
+    return { initial: Instructions.renderInitial(instructions, values), updates: [], update: "" }
+  }
+  const assembled = yield* assembleState(db, sessionID, instructions, state)
+  return {
+    initial: assembled.initial,
+    updates: assembled.updates,
+    update: Instructions.renderUpdate(instructions, assembled.current, dereferenceDelta(result.delta, blobs)),
+  }
+})
+
+const observeAgainst = Effect.fnUntraced(function* (observed: Instructions.ReadResult, previous?: Instructions.Values) {
+  const admission = yield* Instructions.diff(observed, previous)
+  return {
+    current: Instructions.applyHashDelta(previous ?? {}, admission.delta),
+    ...admission,
+  }
 })
 
 const find = Effect.fnUntraced(function* (db: DatabaseService, sessionID: SessionSchema.ID) {
@@ -177,7 +236,26 @@ const find = Effect.fnUntraced(function* (db: DatabaseService, sessionID: Sessio
 const ensure = Effect.fnUntraced(function* (db: DatabaseService, sessionID: SessionSchema.ID) {
   const stored = yield* find(db, sessionID)
   if (!stored) return yield* rebuild(db, sessionID)
-  const latest = yield* db
+  const latest = yield* latestRelevantSequence(db, sessionID)
+  if (!latest || latest.seq <= stored.through_seq) return stored
+  return yield* rebuild(db, sessionID)
+})
+
+const readState = Effect.fnUntraced(function* (db: DatabaseService, sessionID: SessionSchema.ID) {
+  const stored = yield* find(db, sessionID)
+  if (!stored) return yield* stateFromEvents(db, sessionID)
+  const latest = yield* latestRelevantSequence(db, sessionID)
+  if (!latest || latest.seq <= stored.through_seq) return stored
+  return yield* stateFromEvents(db, sessionID)
+})
+
+const stateFromEvents = Effect.fnUntraced(function* (db: DatabaseService, sessionID: SessionSchema.ID) {
+  const folded = fold(yield* instructionEvents(db, sessionID))
+  return folded ? foldedState(sessionID, folded) : undefined
+})
+
+const latestRelevantSequence = Effect.fnUntraced(function* (db: DatabaseService, sessionID: SessionSchema.ID) {
+  return yield* db
     .select({ seq: EventTable.seq })
     .from(EventTable)
     .where(and(eq(EventTable.aggregate_id, sessionID), inArray(EventTable.type, relevantEventTypes)))
@@ -185,8 +263,6 @@ const ensure = Effect.fnUntraced(function* (db: DatabaseService, sessionID: Sess
     .limit(1)
     .get()
     .pipe(Effect.orDie)
-  if (!latest || latest.seq <= stored.through_seq) return stored
-  return yield* rebuild(db, sessionID)
 })
 
 const insertBlobs = Effect.fnUntraced(function* (db: DatabaseService, blobs: Readonly<Record<string, Schema.Json>>) {
@@ -338,3 +414,13 @@ function fold(rows: ReadonlyArray<InstructionEventRow>) {
       : { epochStart: row.seq, throughSeq: row.seq, initial: current, current }
   }, undefined)
 }
+
+function foldedState(sessionID: SessionSchema.ID, folded: NonNullable<ReturnType<typeof fold>>) {
+  return {
+    session_id: sessionID,
+    epoch_start: folded.epochStart,
+    through_seq: folded.throughSeq,
+    initial_values: folded.initial,
+    current_values: folded.current,
+  }
+}

packages/core/src/session/model-request.ts

@@ -0,0 +1,119 @@
+export * as SessionModelRequest from "./model-request"
+
+import { LLM, Message, SystemPart, type LLMRequest } from "@opencode-ai/ai"
+import { SessionError } from "@opencode-ai/schema/session-error"
+import { Context, Effect, Layer } from "effect"
+import { makeLocationNode } from "../effect/app-node"
+import { PluginHooks } from "../plugin/hooks"
+import { ToolRegistry } from "../tool/registry"
+import { SessionContext } from "./context"
+import { SessionModelHeaders } from "./model-headers"
+import { MAX_STEPS_PROMPT } from "./runner/max-steps"
+import PROMPT_DEFAULT from "./runner/prompt/base.txt"
+import { toLLMMessages } from "./runner/to-llm-message"
+
+type ToolCallResolution =
+  | { readonly type: "reject"; readonly error: SessionError.Error }
+  | { readonly type: "settle"; readonly settle: ToolRegistry.Materialization["settle"] }
+
+interface Prepared {
+  readonly request: LLMRequest
+  readonly resolveToolCall: (name: string) => ToolCallResolution
+}
+
+interface PrepareInput {
+  readonly context: SessionContext.Loaded
+  readonly step: number
+}
+
+/**
+ * Builds an outbound model request and captures the tool-call capability that
+ * must remain paired with it. It does not execute the request or mutate
+ * Session state.
+ */
+export interface Interface {
+  /** Builds one outbound model request and its matching tool-call capability. */
+  readonly prepare: (input: PrepareInput) => Effect.Effect<Prepared>
+}
+
+/** Location-scoped outbound model-request preparation. */
+export class Service extends Context.Service<Service, Interface>()("@opencode/v2/SessionModelRequest") {}
+
+const layer = Layer.effect(
+  Service,
+  Effect.gen(function* () {
+    const hooks = yield* PluginHooks.Service
+    const registry = yield* ToolRegistry.Service
+
+    const prepare = Effect.fn("SessionModelRequest.prepare")(function* (input: PrepareInput) {
+      const session = input.context.session
+      const agent = input.context.agent
+      const resolved = input.context.model
+      const model = resolved.model
+      const providerMetadataKey = model.route.providerMetadataKey ?? model.provider
+      const stepLimitReached = agent.info.steps !== undefined && input.step >= agent.info.steps
+      const executableTools = stepLimitReached ? undefined : yield* registry.materialize(agent.info.permissions)
+      const promptCacheKey = /^ses_[0-9a-f]{64}$/.test(session.id) ? session.id.slice(4) : session.id
+      const system = [agent.info.system ? agent.info.system : PROMPT_DEFAULT, input.context.initial]
+        .filter((part) => part.length > 0)
+        .map(SystemPart.make)
+      const history = toLLMMessages(input.context.messages, resolved.ref, providerMetadataKey)
+      const messages = stepLimitReached ? [...history, Message.assistant(MAX_STEPS_PROMPT)] : history
+      const toolDefinitions = executableTools?.definitions ?? []
+      const toolsByName = new Map(toolDefinitions.map((tool) => [tool.name, tool]))
+      // Hooks may reshape available definitions but cannot advertise tools omitted by permissions or the Step limit.
+      const contextEvent = yield* hooks.trigger("session", "context", {
+        sessionID: session.id,
+        agent: agent.id,
+        model: resolved.ref,
+        system,
+        messages,
+        tools: Object.fromEntries(
+          toolDefinitions.map((tool) => [tool.name, { description: tool.description, input: { ...tool.inputSchema } }]),
+        ),
+      })
+      const hookedTools = Object.entries(contextEvent.tools).flatMap(([name, tool]) => {
+        const registered = toolsByName.get(name)
+        return registered
+          ? [Object.assign({}, registered, { description: tool.description, inputSchema: tool.input })]
+          : []
+      })
+      const request = LLM.request({
+        model,
+        http: {
+          headers: SessionModelHeaders.make(session),
+        },
+        providerOptions: { openai: { promptCacheKey } },
+        system: contextEvent.system,
+        messages: contextEvent.messages,
+        tools: hookedTools,
+        toolChoice: stepLimitReached ? "none" : undefined,
+      })
+      const resolveToolCall = (name: string): ToolCallResolution => {
+        if (!executableTools)
+          return {
+            type: "reject",
+            error: { type: "tool.execution", message: "Tools are disabled after the maximum agent steps" },
+          }
+        if (toolsByName.has(name) && !Object.hasOwn(contextEvent.tools, name))
+          return {
+            type: "reject",
+            error: { type: "tool.execution", message: `Tool is not available for this request: ${name}` },
+          }
+        return { type: "settle", settle: executableTools.settle }
+      }
+      return {
+        request,
+        resolveToolCall,
+      }
+    })
+
+    return Service.of({ prepare })
+  }),
+)
+
+export const node = makeLocationNode({
+  service: Service,
+  layer,
+  deps: [PluginHooks.node, ToolRegistry.node],
+})

packages/core/src/session/runner/llm.ts

@@ -1,16 +1,6 @@
 export * as SessionRunnerLLM from "./llm"
 
-import {
-  LLM,
-  LLMClient,
-  LLMError,
-  LLMEvent,
-  Message,
-  SystemPart,
-  isContextOverflowFailure,
-  type ProviderErrorEvent,
-} from "@opencode-ai/ai"
-import type { SessionHooks } from "@opencode-ai/plugin/v2/effect/session"
+import { LLMClient, LLMError, LLMEvent, isContextOverflowFailure, type ProviderErrorEvent } from "@opencode-ai/ai"
 import { SessionError } from "@opencode-ai/schema/session-error"
 import { Money } from "@opencode-ai/schema/money"
 import { Cause, Effect, Exit, Fiber, FiberSet, Layer, Option, Semaphore, Stream } from "effect"
@@ -19,30 +9,25 @@ import { EventV2 } from "../../event"
 import { ModelV2 } from "../../model"
 import { PermissionV2 } from "../../permission"
 import { QuestionTool } from "../../tool/question"
-import { ToolRegistry } from "../../tool/registry"
 import { ToolOutputStore } from "../../tool-output-store"
 import { InstructionState } from "../instruction-state"
 import { SessionCompaction } from "../compaction"
 import { SessionContext } from "../context"
 import { SessionEvent } from "../event"
 import { SessionPending } from "../pending"
+import { SessionModelRequest } from "../model-request"
 import { SessionMessage } from "../message"
 import { SessionSchema } from "../schema"
 import { SessionStore } from "../store"
 import { SessionTitle } from "../title"
 import { Service } from "./index"
 import { createLLMEventPublisher } from "./publish-llm-event"
-import { toLLMMessages } from "./to-llm-message"
-import { MAX_STEPS_PROMPT } from "./max-steps"
-import PROMPT_DEFAULT from "./prompt/base.txt"
 import { Snapshot } from "../../snapshot"
 import { makeLocationNode } from "../../effect/app-node"
 import { llmClient } from "../../effect/app-node-platform"
 import { StepFailedError } from "../error"
 import { toSessionError } from "../to-session-error"
 import { SessionRunnerRetry } from "./retry"
-import { PluginHooks } from "../../plugin/hooks"
-import { SessionModelHeaders } from "../model-headers"
 
 type StepTokens = {
   readonly input: number
@@ -68,20 +53,14 @@ export function calculateCost(costs: ModelV2.Info["cost"], tokens: StepTokens) {
   )
 }
 
-/**
- * Runs one durable coding-agent Session until it settles. Each step reloads projected history,
- * materializes tools, makes one model request, and settles local calls before continuation.
- */
-
 const layer = Layer.effect(
   Service,
   Effect.gen(function* () {
     const events = yield* EventV2.Service
     const llm = yield* LLMClient.Service
-    const tools = yield* ToolRegistry.Service
-    const hooks = yield* PluginHooks.Service
     const store = yield* SessionStore.Service
     const context = yield* SessionContext.Service
+    const modelRequests = yield* SessionModelRequest.Service
     const snapshots = yield* Snapshot.Service
     const db = (yield* Database.Service).db
     const compaction = yield* SessionCompaction.Service
@@ -146,59 +125,18 @@ const layer = Layer.effect(
       const loaded = yield* context.load(selected)
       const session = loaded.session
       const agent = loaded.agent
-      const agentInfo = agent.info
       const resolved = loaded.model
       const model = resolved.model
-      const providerMetadataKey = model.route.providerMetadataKey ?? model.provider
       const compactionInput = { session, messages: loaded.messages, model }
       if (compaction.required(compactionInput) && !(yield* SessionPending.compaction(db, session.id))) {
         const compacted = yield* compaction.compact(compactionInput)
         if (compacted.status === "completed") return { _tag: "RestartAfterCompaction", step: currentStep } as const
         return yield* new StepFailedError({ error: compacted.error })
       }
-      const isLastStep = agentInfo.steps !== undefined && currentStep >= agentInfo.steps
-      const toolMaterialization = isLastStep ? undefined : yield* tools.materialize(agentInfo.permissions)
-      const promptCacheKey = /^ses_[0-9a-f]{64}$/.test(session.id) ? session.id.slice(4) : session.id
-      const request = LLM.request({
-        model,
-        http: {
-          headers: SessionModelHeaders.make(session),
-        },
-        providerOptions: { openai: { promptCacheKey } },
-        system: [agentInfo.system ? agentInfo.system : PROMPT_DEFAULT, loaded.initial]
-          .filter((part): part is string => part !== undefined && part.length > 0)
-          .map(SystemPart.make),
-        messages: [
-          ...toLLMMessages(loaded.messages, resolved.ref, providerMetadataKey),
-          ...(isLastStep ? [Message.assistant(MAX_STEPS_PROMPT)] : []),
-        ],
-        tools: toolMaterialization?.definitions ?? [],
-        toolChoice: isLastStep ? "none" : undefined,
+      const prepared = yield* modelRequests.prepare({
+        context: loaded,
+        step: currentStep,
       })
-      const availableTools = new Map(request.tools.map((tool) => [tool.name, tool]))
-      const contextEvent: SessionHooks["context"] = {
-        sessionID: session.id,
-        agent: agent.id,
-        model: resolved.ref,
-        system: [...request.system],
-        messages: [...request.messages],
-        tools: Object.fromEntries(
-          request.tools.map((tool) => [tool.name, { description: tool.description, input: { ...tool.inputSchema } }]),
-        ),
-      }
-      // Plugins may reshape the draft but cannot advertise tools excluded by
-      // permissions, registration state, or the selected agent's step limit.
-      yield* hooks.trigger("session", "context", contextEvent)
-      const hookedRequest = LLM.updateRequest(request, {
-        system: contextEvent.system,
-        messages: contextEvent.messages,
-        tools: Object.entries(contextEvent.tools).flatMap(([name, tool]) => {
-          const registered = availableTools.get(name)
-          if (!registered) return []
-          return [{ ...registered, description: tool.description, inputSchema: tool.input }]
-        }),
-      })
-      const advertisedTools = new Set(hookedRequest.tools.map((tool) => tool.name))
       const toolFibers = yield* FiberSet.make<void, ToolOutputStore.Error>()
       const ownedToolFibers: Array<Fiber.Fiber<void, ToolOutputStore.Error>> = []
       let needsContinuation = false
@@ -209,7 +147,7 @@ const layer = Layer.effect(
         // The selected catalog identity, not model.id: route-level ids are provider API
         // model ids (for example gpt-5.5-fast resolves to api id gpt-5.5).
         model: resolved.ref,
-        providerMetadataKey,
+        providerMetadataKey: model.route.providerMetadataKey ?? model.provider,
         snapshot: startSnapshot,
         assistantMessageID,
       })
@@ -219,7 +157,7 @@ const layer = Layer.effect(
       const serialized = <A, E, R>(effect: Effect.Effect<A, E, R>) => publication.withPermit(effect)
       const publish = (event: LLMEvent, error?: SessionError.Error) => serialized(publisher.publish(event, error))
       let overflowFailure: ProviderErrorEvent | undefined
-      const providerStream = llm.stream(hookedRequest).pipe(
+      const providerStream = llm.stream(prepared.request).pipe(
         Stream.runForEach((event) =>
           Effect.gen(function* () {
             if (overflowFailure || publisher.hasProviderError()) return
@@ -231,15 +169,9 @@ const layer = Layer.effect(
             }
             yield* publish(event)
             if (event.type !== "tool-call" || event.providerExecuted) return
-            if (!toolMaterialization || (availableTools.has(event.name) && !advertisedTools.has(event.name))) {
-              yield* serialized(
-                publisher.failUnsettledTools({
-                  type: "tool.execution",
-                  message: toolMaterialization
-                    ? `Tool is not available for this request: ${event.name}`
-                    : "Tools are disabled after the maximum agent steps",
-                }),
-              )
+            const tool = prepared.resolveToolCall(event.name)
+            if (tool.type === "reject") {
+              yield* serialized(publisher.failUnsettledTools(tool.error))
               return
             }
             needsContinuation = true
@@ -247,7 +179,7 @@ const layer = Layer.effect(
             ownedToolFibers.push(
               yield* Effect.uninterruptibleMask((restore) =>
                 restore(
-                  toolMaterialization.settle({
+                  tool.settle({
                     sessionID: session.id,
                     agent: agent.id,
                     messageID: assistantMessageID,
@@ -337,11 +269,7 @@ const layer = Layer.effect(
           const llmFailure = streamFailure instanceof LLMError ? streamFailure : undefined
           if (llmFailure && !publisher.hasProviderError()) {
             const error = toSessionError(llmFailure)
-            if (
-              SessionRunnerRetry.isRetryable(llmFailure) &&
-              !publisher.hasRetryEvidence() &&
-              (agentInfo.steps === undefined || currentStep < agentInfo.steps)
-            ) {
+            if (SessionRunnerRetry.isRetryable(llmFailure) && !publisher.hasRetryEvidence()) {
               return yield* new SessionRunnerRetry.RetryableFailure({
                 cause: llmFailure,
                 assistantMessageID: yield* publisher.startAssistant(),
@@ -457,7 +385,7 @@ const layer = Layer.effect(
           Effect.tapError((error) =>
             error instanceof SessionRunnerRetry.RetryableFailure
               ? Effect.sync(() => {
-                  currentStep = error.step + 1
+                  currentStep = error.step
                   assistantMessageID = error.assistantMessageID
                   currentPromotion = undefined
                 })
@@ -571,9 +499,8 @@ export const node = makeLocationNode({
   deps: [
     EventV2.node,
     llmClient,
-    ToolRegistry.node,
-    PluginHooks.node,
     SessionContext.node,
+    SessionModelRequest.node,
     SessionStore.node,
     SessionCompaction.node,
     SessionTitle.node,

packages/core/src/skill/instructions.ts

@@ -1,4 +1,4 @@
-export * as SkillGuidance from "./guidance"
+export * as SkillInstructions from "./instructions"
 
 import { makeLocationNode } from "../effect/app-node"
 import { Context, Effect, Layer, Schema } from "effect"
@@ -60,7 +60,7 @@ export interface Interface {
   readonly load: (agent: AgentV2.Selection) => Effect.Effect<Instructions.Instructions>
 }
 
-export class Service extends Context.Service<Service, Interface>()("@opencode/v2/SkillGuidance") {}
+export class Service extends Context.Service<Service, Interface>()("@opencode/v2/SkillInstructions") {}
 
 const layer = Layer.effect(
   Service,
@@ -68,7 +68,7 @@ const layer = Layer.effect(
     const skills = yield* SkillV2.Service
 
     return Service.of({
-      load: Effect.fn("SkillGuidance.load")(function* (selection) {
+      load: Effect.fn("SkillInstructions.load")(function* (selection) {
         const agent = selection.info
         if (!agent) return Instructions.empty
         const permitted = SkillV2.available(yield* skills.list(), agent)

packages/core/src/state.ts

@@ -1,14 +1,14 @@
 export * as State from "./state"
 
-import { Context, Effect, Scope, Semaphore } from "effect"
+import { Clock, Context, Deferred, Effect, Scope, Semaphore } from "effect"
 
 /**
  * A replayable transform applied to a draft during reload.
  *
  * Domain drafts expose readable and writable state while preserving concise
- * plugin/config code. Transforms may perform Effects before returning.
+ * plugin/config code. Transforms synchronously rebuild derived state.
  */
-type TransformCallback<DraftApi> = (draft: DraftApi) => Effect.Effect<void> | void
+type TransformCallback<DraftApi> = (draft: DraftApi) => void
 export type MakeDraft<State, DraftApi> = (state: State) => DraftApi
 
 export interface Registration {
@@ -34,6 +34,7 @@ type Batch = {
 const CurrentBatch = Context.Reference<Batch | undefined>("@opencode/State/CurrentBatch", {
   defaultValue: () => undefined,
 })
+const reloadDebounce = 500
 
 export function batch<A, E, R>(effect: Effect.Effect<A, E, R>) {
   return Effect.gen(function* () {
@@ -73,6 +74,10 @@ export interface Interface<State, DraftApi> extends Transformable<DraftApi> {
 export function create<State, DraftApi>(options: Options<State, DraftApi>): Interface<State, DraftApi> {
   let state = options.initial()
   let transforms: { run: TransformCallback<DraftApi> }[] = []
+  let generation = 0
+  let requestedAt = 0
+  let running = false
+  let waiters: { generation: number; done: Deferred.Deferred<void> }[] = []
   const semaphore = Semaphore.makeUnsafe(1)
 
   const commit = Effect.fn("State.commit")(function* (next: State) {
@@ -82,9 +87,8 @@ export function create<State, DraftApi>(options: Options<State, DraftApi>): Inte
   })
 
   const apply = (transform: TransformCallback<DraftApi>, draft: DraftApi) =>
-    Effect.suspend(() => {
-      const result = transform(draft)
-      return Effect.isEffect(result) ? Effect.asVoid(result).pipe(Effect.orDie) : Effect.void
+    Effect.sync(() => {
+      transform(draft)
     })
 
   const materialize = Effect.fnUntraced(function* () {
@@ -97,7 +101,39 @@ export function create<State, DraftApi>(options: Options<State, DraftApi>): Inte
     yield* commit(next)
   })
 
-  const reload = () => semaphore.withPermit(materialize())
+  const materializeReload = () => semaphore.withPermit(materialize())
+
+  const rebuild = (): Effect.Effect<void> =>
+    Effect.gen(function* () {
+      const clock = yield* Clock.Clock
+      const remaining = requestedAt + reloadDebounce - clock.currentTimeMillisUnsafe()
+      if (remaining > 0) yield* Effect.sleep(remaining)
+      if (clock.currentTimeMillisUnsafe() < requestedAt + reloadDebounce) return yield* rebuild()
+
+      const target = generation
+      const exit = yield* materializeReload().pipe(Effect.exit)
+      const completed = waiters.filter((waiter) => waiter.generation <= target)
+      waiters = waiters.filter((waiter) => waiter.generation > target)
+      yield* Effect.forEach(completed, (waiter) => Deferred.done(waiter.done, exit), {
+        concurrency: "unbounded",
+        discard: true,
+      })
+      if (generation > target) return yield* rebuild()
+      running = false
+    })
+
+  const reload = Effect.fnUntraced(function* () {
+    const done = Deferred.makeUnsafe<void>()
+    const clock = yield* Clock.Clock
+    generation++
+    requestedAt = clock.currentTimeMillisUnsafe()
+    waiters.push({ generation, done })
+    if (!running) {
+      running = true
+      yield* rebuild().pipe(Effect.forkDetach)
+    }
+    return yield* Deferred.await(done)
+  })
 
   const result: Interface<State, DraftApi> = {
     get: () => state,
@@ -117,7 +153,7 @@ export function create<State, DraftApi>(options: Options<State, DraftApi>): Inte
                 return Effect.gen(function* () {
                   const batch = yield* CurrentBatch
                   if (batch?.active) {
-                    batch.reloads.add(reload)
+                    batch.reloads.add(materializeReload)
                     return
                   }
                   yield* materialize()
@@ -132,8 +168,8 @@ export function create<State, DraftApi>(options: Options<State, DraftApi>): Inte
           )
           yield* Scope.addFinalizer(scope, dispose)
           const batch = yield* CurrentBatch
-          if (batch?.active) batch.reloads.add(reload)
-          else yield* reload()
+          if (batch?.active) batch.reloads.add(materializeReload)
+          else yield* materializeReload()
           return { dispose }
         }),
       )

packages/core/src/tool/read.ts

@@ -6,7 +6,6 @@ import { ToolFailure } from "@opencode-ai/ai"
 import { Effect, Schema } from "effect"
 import { FileSystem } from "../filesystem"
 import { FSUtil } from "../fs-util"
-import { Image } from "../image"
 import { Location } from "../location"
 import { LocationMutation } from "../location-mutation"
 import { PermissionV2 } from "../permission"
@@ -35,7 +34,6 @@ export const Plugin = {
   effect: Effect.fn("ReadTool.Plugin")(function* (ctx: PluginContext) {
     const reader = yield* ReadToolFileSystem.Service
     const mutation = yield* LocationMutation.Service
-    const image = yield* Image.Service
     const permission = yield* PermissionV2.Service
     const sessionInstructions = yield* SessionInstructions.Service
     const fs = yield* FSUtil.Service
@@ -50,6 +48,12 @@ export const Plugin = {
               "Read a text file or supported image, page through a large UTF-8 text file by line offset, or list a directory page. Relative paths resolve from the current location; absolute paths inside it are accepted, while external absolute paths require external_directory approval.",
             input: Input,
             output: Output,
+            structured: Schema.toEncoded(Output),
+            // Image base64 reaches the model through content items (normalized generically
+            // at tool settlement); persisting a second copy in structured would store the
+            // original unresized bytes in the message row.
+            toStructuredOutput: ({ output }) =>
+              "encoding" in output && output.encoding === "base64" ? { ...output, content: "" } : output,
             toModelOutput: ({ input, output }) => {
               if (!("encoding" in output) || output.encoding !== "base64" || !SUPPORTED_IMAGE_MIMES.has(output.mime))
                 return []
@@ -117,21 +121,14 @@ export const Plugin = {
                   Effect.catch(() => Effect.void),
                   Effect.catchDefect(() => Effect.void),
                 )
-                if ("encoding" in content && content.encoding === "base64" && SUPPORTED_IMAGE_MIMES.has(content.mime)) {
-                  return yield* image
-                    .normalize(resource, { ...content, encoding: "base64" })
-                    .pipe(Effect.catchTag("Image.ResizerUnavailableError", () => Effect.succeed(content)))
-                }
-                if ("encoding" in content && content.encoding === "base64")
+                if ("encoding" in content && content.encoding === "base64" && !SUPPORTED_IMAGE_MIMES.has(content.mime))
                   return yield* Effect.fail(new ReadToolFileSystem.BinaryFileError({ resource }))
                 return content
               }).pipe(
                 Effect.mapError((error) => {
                   const message =
                     error instanceof ReadToolFileSystem.BinaryFileError ||
-                    error instanceof ReadToolFileSystem.MediaIngestLimitError ||
-                    error instanceof Image.DecodeError ||
-                    error instanceof Image.SizeError
+                    error instanceof ReadToolFileSystem.MediaIngestLimitError
                       ? error.message
                       : `Unable to read ${input.path}`
                   return new ToolFailure({ message, error })

packages/core/src/tool/registry.ts

@@ -3,6 +3,7 @@ export * as ToolRegistry from "./registry"
 import { ToolOutput, type ToolCall, type ToolDefinition, type ToolResultValue } from "@opencode-ai/ai"
 import { Context, Effect, Layer, Scope } from "effect"
 import type { AgentV2 } from "../agent"
+import { Image } from "../image"
 import { PermissionV2 } from "../permission"
 import { SessionMessage } from "../session/message"
 import { SessionSchema } from "../session/schema"
@@ -57,6 +58,40 @@ const registryLayer = Layer.effect(
   Effect.gen(function* () {
     const resources = yield* ToolOutputStore.Service
     const toolHooks = yield* ToolHooks.Service
+    const image = yield* Image.Service
+
+    type NormalizedItem = ToolOutput["content"][number] | "decode" | "size"
+    const normalizeImages = Effect.fn("ToolRegistry.normalizeImages")(function* (content: ToolOutput["content"]) {
+      const normalized = yield* Effect.forEach(content, (item): Effect.Effect<NormalizedItem> => {
+        if (item.type !== "file" || !item.mime.startsWith("image/")) return Effect.succeed(item)
+        // RFC 2397 permits parameters between the mime and ";base64".
+        const base64 = /^data:[^,]*;base64,(.*)$/s.exec(item.uri)?.[1]
+        if (base64 === undefined) return Effect.succeed(item)
+        const resource = item.name ?? `${item.mime} tool output`
+        return image
+          .normalize(resource, { uri: resource, content: base64, encoding: "base64", mime: item.mime })
+          .pipe(
+            Effect.map((result) => ({
+              ...item,
+              uri: `data:${result.mime};base64,${result.content}`,
+              mime: result.mime,
+            })),
+            Effect.catchTag("Image.ResizerUnavailableError", () => Effect.succeed(item)),
+            Effect.catchTag("Image.DecodeError", () => Effect.succeed("decode" as const)),
+            Effect.catchTag("Image.SizeError", () => Effect.succeed("size" as const)),
+          )
+      })
+      const note = (reason: "decode" | "size", text: string) => {
+        const count = normalized.filter((item) => item === reason).length
+        if (count === 0) return []
+        return [{ type: "text" as const, text: `[${count} image${count === 1 ? "" : "s"} omitted: ${text}]` }]
+      }
+      return [
+        ...normalized.filter((item) => typeof item !== "string"),
+        ...note("decode", "could not be decoded."),
+        ...note("size", "could not be resized below the image size limit."),
+      ]
+    })
     type Registration = {
       readonly tool: AnyTool
       readonly name: string
@@ -84,10 +119,11 @@ const registryLayer = Layer.effect(
           agent: input.agent,
           messageID: input.messageID,
           callID: input.call.id,
-          progress: (update) =>
-            input.progress?.({
-              structured: update.structured,
-              content: (update.content ?? []).map((part) =>
+          progress: (update) => {
+            const progress = input.progress
+            if (!progress) return Effect.void
+            return normalizeImages(
+              (update.content ?? []).map((part) =>
                 part.type === "text"
                   ? { type: "text" as const, text: part.text }
                   : {
@@ -97,7 +133,8 @@ const registryLayer = Layer.effect(
                       name: part.name,
                     },
               ),
-            }) ?? Effect.void,
+            ).pipe(Effect.flatMap((content) => progress({ structured: update.structured, content })))
+          },
         },
       ).pipe(
         Effect.map((output) => ({ output })),
@@ -115,7 +152,7 @@ const registryLayer = Layer.effect(
         const bounded = yield* resources.bound({
           sessionID: input.sessionID,
           callID: input.call.id,
-          output: pending.output,
+          output: { structured: pending.output.structured, content: yield* normalizeImages(pending.output.content) },
         })
         const result = ToolOutput.toResultValue(bounded.output)
         settlement =
@@ -232,11 +269,11 @@ function whollyDisabled(action: string, rules: PermissionV2.Ruleset) {
 export const node = makeLocationNode({
   service: Service,
   layer,
-  deps: [ToolOutputStore.node, ToolHooks.node],
+  deps: [ToolOutputStore.node, ToolHooks.node, Image.node],
 })
 
 export const toolsNode = makeLocationNode({
   service: Tools.Service,
   layer,
-  deps: [ToolOutputStore.node, ToolHooks.node],
+  deps: [ToolOutputStore.node, ToolHooks.node, Image.node],
 })

packages/core/src/tool/shell.ts

@@ -17,6 +17,7 @@ export const name = "shell"
 export const DEFAULT_TIMEOUT_MS = 2 * 60 * 1_000
 export const MAX_TIMEOUT_MS = 10 * 60 * 1_000
 export const MAX_CAPTURE_BYTES = 1024 * 1024
+export const PROGRESS_LINES = 25
 
 const BACKGROUND_STARTED = "The command was moved to the background."
 const BACKGROUND_INSTRUCTION =
@@ -210,6 +211,23 @@ export const Plugin = {
                   }
                 })
 
+                const captureProgress = Effect.fn("ShellTool.captureProgress")(function* () {
+                  const latest = yield* shell.output(info.id, { cursor: Number.MAX_SAFE_INTEGER })
+                  const start = Math.max(0, latest.size - MAX_CAPTURE_BYTES)
+                  const page = yield* shell.output(info.id, { cursor: start, limit: MAX_CAPTURE_BYTES })
+                  const trailingNewline = page.output.endsWith("\n")
+                  const lines = trailingNewline ? page.output.split("\n").slice(0, -1) : page.output.split("\n")
+                  const truncated = start > 0 || lines.length > PROGRESS_LINES
+                  const output = lines.slice(-PROGRESS_LINES).join("\n") + (trailingNewline ? "\n" : "")
+                  const notice = truncated
+                    ? `[output truncated; showing last ${PROGRESS_LINES} lines. Full output saved to: ${info.file}]\n\n`
+                    : ""
+                  return {
+                    output: `${notice}${output || "(no output)"}`,
+                    truncated,
+                  }
+                })
+
                 const settleShell = Effect.fn("ShellTool.settleShell")(function* () {
                   const final = yield* shell.wait(info.id)
 
@@ -258,7 +276,7 @@ export const Plugin = {
 
                 const progress = yield* Effect.sleep("1 second").pipe(
                   Effect.andThen(
-                    captureShell().pipe(
+                    captureProgress().pipe(
                       Effect.flatMap((capture) =>
                         context.progress({
                           structured: { truncated: capture.truncated },

packages/core/src/tool/subagent.ts

@@ -4,6 +4,7 @@ import { ToolFailure } from "@opencode-ai/ai"
 import type { Context as PluginContext } from "@opencode-ai/plugin/v2/effect/plugin"
 import { Effect, Schema, Scope } from "effect"
 import { AgentV2 } from "../agent"
+import { Config } from "../config"
 import { PluginRuntime } from "../plugin/runtime"
 import { PermissionV2 } from "../permission"
 import { SessionSchema } from "../session/schema"
@@ -43,6 +44,7 @@ export const Plugin = {
   effect: Effect.fn("SubagentTool.Plugin")(function* (ctx: PluginContext) {
     const runtime = yield* PluginRuntime.Service
     const agents = yield* AgentV2.Service
+    const config = yield* Config.Service
     const permission = yield* PermissionV2.Service
     const scope = yield* Scope.Scope
 
@@ -123,6 +125,23 @@ export const Plugin = {
                       (error) => new ToolFailure({ message: `Parent session not found: ${context.sessionID}`, error }),
                     ),
                   )
+                let current = parent
+                let depth = 0
+                while (current.parentID) {
+                  depth++
+                  current = yield* runtime.session
+                    .get(current.parentID)
+                    .pipe(
+                      Effect.mapError(
+                        (error) => new ToolFailure({ message: `Parent session not found: ${current.parentID}`, error }),
+                      ),
+                    )
+                }
+                const limit = Config.latest(yield* config.entries(), "experimental")?.subagent_depth ?? 1
+                if (depth >= limit)
+                  return yield* new ToolFailure({
+                    message: `Subagent depth limit reached (${limit}). Increase "experimental.subagent_depth" to allow nested subagents.`,
+                  })
                 const agent = yield* agents.resolve(input.agent)
                 if (agent === undefined) return yield* new ToolFailure({ message: `Unknown agent: ${input.agent}` })
                 if (agent.mode === "primary")

packages/core/src/v1/config/config.ts

@@ -175,6 +175,9 @@ export const Info = Schema.Struct({
       primary_tools: Schema.optional(Schema.mutable(Schema.Array(Schema.String))).annotate({
         description: "Tools that should only be available to primary agents.",
       }),
+      subagent_depth: Schema.optional(NonNegativeInt).annotate({
+        description: "Maximum subagent nesting depth. Defaults to 1.",
+      }),
       continue_loop_on_deny: Schema.optional(Schema.Boolean).annotate({
         description: "Continue the agent loop when a tool call is denied",
       }),

packages/core/src/v1/config/migrate.ts

@@ -77,6 +77,7 @@ export function migrate(info: typeof ConfigV1.Info.Type) {
     commands: commands(info.command),
     instructions: info.instructions,
     references: info.references ?? info.reference,
+    experimental: experimental(info),
     plugins: info.plugin?.map((plugin) =>
       typeof plugin === "string" ? plugin : { package: plugin[0], options: plugin[1] },
     ),
@@ -84,6 +85,31 @@ export function migrate(info: typeof ConfigV1.Info.Type) {
   }
 }
 
+function experimental(info: typeof ConfigV1.Info.Type) {
+  const policies = [
+    ...(info.enabled_providers === undefined
+      ? []
+      : [
+          { action: "provider.use" as const, resource: "*", effect: "deny" as const },
+          ...info.enabled_providers.map((resource) => ({
+            action: "provider.use" as const,
+            resource,
+            effect: "allow" as const,
+          })),
+        ]),
+    ...(info.disabled_providers ?? []).map((resource) => ({
+      action: "provider.use" as const,
+      resource,
+      effect: "deny" as const,
+    })),
+  ]
+  if (info.experimental?.subagent_depth === undefined && !policies.length) return
+  return {
+    subagent_depth: info.experimental?.subagent_depth,
+    policies: policies.length ? policies : undefined,
+  }
+}
+
 function permissions(info?: ConfigPermissionV1.Info, tools?: Readonly<Record<string, boolean>>) {
   const rules: Array<{ action: string; resource: string; effect: ConfigPermissionV1.Action }> = Object.entries(
     tools ?? {},

packages/core/src/wellknown.ts

@@ -0,0 +1,194 @@
+export * as WellKnown from "./wellknown"
+
+import { Integration } from "@opencode-ai/schema/integration"
+import { Context, Effect, Layer, Ref, Schema, Semaphore } from "effect"
+import { HttpClient, HttpClientRequest, HttpClientResponse } from "effect/unstable/http"
+import { isDeepStrictEqual } from "node:util"
+import { makeGlobalNode } from "./effect/app-node"
+import { httpClient } from "./effect/app-node-platform"
+import { EventV2 } from "./event"
+import { KV } from "./kv"
+
+export interface Auth extends Schema.Schema.Type<typeof Auth> {}
+export const Auth = Schema.Struct({
+  command: Schema.Array(Schema.String),
+  env: Schema.String,
+}).annotate({ identifier: "WellKnown.Auth" })
+
+export interface RemoteConfig extends Schema.Schema.Type<typeof RemoteConfig> {}
+export const RemoteConfig = Schema.Struct({
+  url: Schema.String,
+  headers: Schema.optional(Schema.Record(Schema.String, Schema.String)),
+}).annotate({ identifier: "WellKnown.RemoteConfig" })
+
+export interface Config extends Schema.Schema.Type<typeof Config> {}
+export const Config = Schema.Record(Schema.String, Schema.Json).annotate({ identifier: "WellKnown.Config" })
+
+export interface Manifest extends Schema.Schema.Type<typeof Manifest> {}
+export const Manifest = Schema.Struct({
+  auth: Schema.optional(Auth),
+  config: Schema.optional(Schema.NullOr(Config)),
+  remote_config: Schema.optional(RemoteConfig),
+}).annotate({ identifier: "WellKnown.Manifest" })
+
+export interface ResolveInput {
+  readonly origin: string
+  readonly variables?: Readonly<Record<string, string>>
+}
+
+export interface Entry {
+  readonly origin: string
+  readonly integrationID: Integration.ID
+  readonly manifest: Manifest
+}
+
+export interface Interface {
+  readonly entries: () => Effect.Effect<readonly Entry[], Error>
+  readonly snapshot: () => readonly Entry[]
+  readonly refresh: () => Effect.Effect<boolean, Error>
+  readonly add: (origin: string) => Effect.Effect<Entry, Error>
+  readonly remove: (origin: string) => Effect.Effect<void>
+  readonly resolve: (entry: Entry, variables: Readonly<Record<string, string>>) => Effect.Effect<Config[], Error>
+}
+
+export class Service extends Context.Service<Service, Interface>()("@opencode/v2/WellKnown") {}
+
+export const Event = {
+  Updated: EventV2.ephemeral({ type: "wellknown.updated", schema: {} }),
+}
+
+export const inspect = Effect.fn("WellKnown.inspect")(function* (origin: string) {
+  const url = `${origin.replace(/\/+$/, "")}/.well-known/opencode`
+  const http = HttpClient.filterStatusOk(yield* HttpClient.HttpClient)
+  return yield* http.execute(HttpClientRequest.get(url).pipe(HttpClientRequest.acceptJson)).pipe(
+    Effect.flatMap(HttpClientResponse.schemaBodyJson(Manifest)),
+    Effect.mapError((cause) => new Error(`Failed to load wellknown manifest from ${url}`, { cause })),
+  )
+})
+
+export const resolve = Effect.fn("WellKnown.resolve")(function* (input: ResolveInput) {
+  const manifest = yield* inspect(input.origin)
+  return yield* resolveEntry(
+    { origin: input.origin, integrationID: Integration.ID.make(input.origin.replace(/\/+$/, "")), manifest },
+    input.variables ?? {},
+  )
+})
+
+const resolveEntry = Effect.fnUntraced(function* (entry: Entry, variables: Readonly<Record<string, string>>) {
+  const configs = entry.manifest.config ? [entry.manifest.config] : []
+  if (!entry.manifest.remote_config) return configs
+
+  const substitute = (value: string) =>
+    value.replace(/\{env:([^}]+)\}/g, (_, name: string) => variables[name] ?? process.env[name] ?? "")
+  const url = substitute(entry.manifest.remote_config.url)
+  const headers = Object.fromEntries(
+    Object.entries(entry.manifest.remote_config.headers ?? {}).map(([key, value]) => [key, substitute(value)]),
+  )
+  const http = HttpClient.filterStatusOk(yield* HttpClient.HttpClient)
+  const remote = yield* http
+    .execute(HttpClientRequest.get(url).pipe(HttpClientRequest.acceptJson, HttpClientRequest.setHeaders(headers)))
+    .pipe(
+      Effect.flatMap(HttpClientResponse.schemaBodyJson(Config)),
+      Effect.mapError((cause) => new Error(`Failed to load wellknown remote config from ${url}`, { cause })),
+    )
+  if (Schema.is(Config)(remote.config)) return [...configs, remote.config]
+  return [...configs, remote]
+})
+
+const sourcesKey = "wellknown:sources"
+const Sources = Schema.Array(Schema.String)
+
+const layer = Layer.effect(
+  Service,
+  Effect.gen(function* () {
+    const http = yield* HttpClient.HttpClient
+    const kv = yield* KV.Service
+    const events = yield* EventV2.Service
+    const cache = yield* Ref.make(new Map<string, Entry>())
+    const lock = Semaphore.makeUnsafe(1)
+
+    const load = Effect.fn("WellKnown.load")(function* () {
+      const value = yield* kv.get(sourcesKey)
+      const origins = Schema.is(Sources)(value) ? value : []
+      const current = yield* Ref.get(cache)
+      const entries = yield* Effect.forEach(origins, (origin) => {
+        const cached = current.get(origin)
+        if (cached) return Effect.succeed(cached)
+        return inspect(origin).pipe(
+          Effect.provideService(HttpClient.HttpClient, http),
+          Effect.map((manifest) => ({ origin, integrationID: Integration.ID.make(origin), manifest })),
+        )
+      })
+      yield* Ref.set(cache, new Map(entries.map((entry) => [entry.origin, entry])))
+      return entries
+    })
+
+    const refresh = Effect.fn("WellKnown.refresh")(function* () {
+      return yield* lock.withPermit(
+        Effect.gen(function* () {
+          const value = yield* kv.get(sourcesKey)
+          const origins = Schema.is(Sources)(value) ? value : []
+          if (!origins.length) return false
+          const entries = yield* Effect.forEach(origins, (origin) =>
+            inspect(origin).pipe(
+              Effect.provideService(HttpClient.HttpClient, http),
+              Effect.map((manifest) => ({ origin, integrationID: Integration.ID.make(origin), manifest })),
+            ),
+          )
+          const next = new Map(entries.map((entry) => [entry.origin, entry]))
+          const changed = !isDeepStrictEqual(Ref.getUnsafe(cache), next)
+          if (!changed) return false
+          yield* Ref.set(cache, next)
+          yield* events.publish(Event.Updated, {})
+          return true
+        }),
+      )
+    })
+
+    return Service.of({
+      entries: load,
+      snapshot: () => Array.from(Ref.getUnsafe(cache).values()),
+      refresh,
+      add: Effect.fn("WellKnown.add")(function* (value) {
+        return yield* lock.withPermit(
+          Effect.gen(function* () {
+            const origin = value.replace(/\/+$/, "")
+            const manifest = yield* inspect(origin).pipe(Effect.provideService(HttpClient.HttpClient, http))
+            if (!manifest.auth) return yield* Effect.fail(new Error(`No authentication method found at ${origin}`))
+            const entry = { origin, integrationID: Integration.ID.make(origin), manifest }
+            const sources = yield* kv.get(sourcesKey)
+            const origins = Schema.is(Sources)(sources) ? sources : []
+            yield* kv.set(sourcesKey, Array.from(new Set([...origins, origin])))
+            yield* Ref.update(cache, (current) => new Map(current).set(origin, entry))
+            yield* events.publish(Event.Updated, {})
+            return entry
+          }),
+        )
+      }),
+      remove: Effect.fn("WellKnown.remove")(function* (value) {
+        yield* lock.withPermit(
+          Effect.gen(function* () {
+            const origin = value.replace(/\/+$/, "")
+            const sources = yield* kv.get(sourcesKey)
+            const origins = Schema.is(Sources)(sources) ? sources : []
+            yield* kv.set(
+              sourcesKey,
+              origins.filter((item) => item !== origin),
+            )
+            yield* Ref.update(cache, (current) => {
+              const next = new Map(current)
+              next.delete(origin)
+              return next
+            })
+            yield* events.publish(Event.Updated, {})
+          }),
+        )
+      }),
+      resolve: Effect.fn("WellKnown.resolveEntry")(function* (entry, variables) {
+        return yield* resolveEntry(entry, variables).pipe(Effect.provideService(HttpClient.HttpClient, http))
+      }),
+    })
+  }),
+)
+
+export const node = makeGlobalNode({ service: Service, layer, deps: [httpClient, KV.node, EventV2.node] })

packages/core/src/wellknown/plugin.ts

@@ -0,0 +1,36 @@
+export * as WellKnownPlugin from "./plugin"
+
+import { define } from "@opencode-ai/plugin/v2/effect/plugin"
+import { Effect, Stream } from "effect"
+import { EventV2 } from "../event"
+import { WellKnown } from "../wellknown"
+
+export const Plugin = define({
+  id: "opencode.wellknown",
+  effect: Effect.fn(function* (ctx) {
+    const events = yield* EventV2.Service
+    const wellknown = yield* WellKnown.Service
+    yield* wellknown.entries().pipe(Effect.orDie)
+    yield* ctx.integration.transform((draft) => {
+      wellknown.snapshot().forEach((entry) => {
+        if (!entry.manifest.auth) return
+        draft.update(entry.integrationID, (integration) => {
+          integration.name = new URL(entry.origin).hostname
+        })
+        draft.method.update({
+          integrationID: entry.integrationID,
+          method: {
+            id: "login",
+            type: "command",
+            label: "Log in",
+            command: [...entry.manifest.auth.command],
+          },
+        })
+      })
+    })
+    yield* events.subscribe(WellKnown.Event.Updated).pipe(
+      Stream.runForEach(() => ctx.integration.reload()),
+      Effect.forkScoped({ startImmediately: true }),
+    )
+  }),
+})

packages/core/test/agent.test.ts

@@ -1,5 +1,6 @@
 import { describe, expect } from "bun:test"
 import { Effect, Exit, Fiber, Layer, Scope, Stream } from "effect"
+import { TestClock } from "effect/testing"
 import { AgentV2 } from "@opencode-ai/core/agent"
 import { EventV2 } from "@opencode-ai/core/event"
 import { AppNodeBuilder } from "@opencode-ai/core/effect/app-node-builder"
@@ -76,7 +77,9 @@ describe("AgentV2", () => {
       )
       description = "New description"
       hidden = false
-      yield* agent.reload()
+      const reload = yield* agent.reload().pipe(Effect.forkChild({ startImmediately: true }))
+      yield* TestClock.adjust("500 millis")
+      yield* Fiber.join(reload)
 
       expect(yield* agent.get(id)).toMatchObject({ description: "New description", hidden: false })
     }),

packages/core/test/catalog.test.ts

@@ -1,6 +1,7 @@
 import { describe, expect } from "bun:test"
 import { Money } from "@opencode-ai/schema/money"
 import { Effect, Fiber, Layer, Stream } from "effect"
+import { TestClock } from "effect/testing"
 import { Catalog } from "@opencode-ai/core/catalog"
 import { Integration } from "@opencode-ai/core/integration"
 import { Credential } from "@opencode-ai/core/credential"
@@ -261,7 +262,9 @@ describe("CatalogV2", () => {
       expect((yield* catalog.model.default())?.id).toBe(old)
 
       configured = false
-      yield* catalog.reload()
+      const reload = yield* catalog.reload().pipe(Effect.forkChild({ startImmediately: true }))
+      yield* TestClock.adjust("500 millis")
+      yield* Fiber.join(reload)
       expect((yield* catalog.model.default())?.id).toBe(newest)
     }),
   )

packages/core/test/config/config.test.ts

@@ -8,7 +8,9 @@ import { ConfigModel } from "@opencode-ai/core/config/model"
 import { Config as ConfigSchema } from "@opencode-ai/schema/config"
 import { ConfigProvider } from "@opencode-ai/core/config/provider"
 import { AppNodeBuilder } from "@opencode-ai/core/effect/app-node-builder"
+import { makeGlobalNode } from "@opencode-ai/core/effect/app-node"
 import { LayerNode } from "@opencode-ai/core/effect/layer-node"
+import { Credential } from "@opencode-ai/core/credential"
 import { ConfigMigrateV1 } from "@opencode-ai/core/v1/config/migrate"
 import { ConfigV1 } from "@opencode-ai/core/v1/config/config"
 import { FSUtil } from "@opencode-ai/core/fs-util"
@@ -19,6 +21,8 @@ import { Location } from "@opencode-ai/core/location"
 import { Project } from "@opencode-ai/core/project"
 import { ProviderV2 } from "@opencode-ai/core/provider"
 import { AbsolutePath } from "@opencode-ai/core/schema"
+import { WellKnown } from "@opencode-ai/core/wellknown"
+import { Integration } from "@opencode-ai/schema/integration"
 import { location } from "../fixture/location"
 import { tmpdir } from "../fixture/tmpdir"
 import { testEffect } from "../lib/effect"
@@ -26,12 +30,46 @@ import { testEffect } from "../lib/effect"
 const it = testEffect(Layer.empty)
 const selection = Schema.decodeUnknownSync(ConfigModel.Selection)
 
+const emptyCredentialNode = makeGlobalNode({
+  service: Credential.Service,
+  layer: Layer.succeed(
+    Credential.Service,
+    Credential.Service.of({
+      all: () => Effect.succeed([]),
+      list: () => Effect.succeed([]),
+      get: () => Effect.succeed(undefined),
+      create: () => Effect.die("unused Credential.create"),
+      update: () => Effect.die("unused Credential.update"),
+      remove: () => Effect.die("unused Credential.remove"),
+    }),
+  ),
+  deps: [],
+})
+
+const emptyWellknownNode = makeGlobalNode({
+  service: WellKnown.Service,
+  layer: Layer.succeed(
+    WellKnown.Service,
+    WellKnown.Service.of({
+      entries: () => Effect.succeed([]),
+      snapshot: () => [],
+      refresh: () => Effect.succeed(false),
+      add: () => Effect.die("unused Wellknown.add"),
+      remove: () => Effect.die("unused Wellknown.remove"),
+      resolve: () => Effect.die("unused Wellknown.resolve"),
+    }),
+  ),
+  deps: [],
+})
+
 function testLayer(
   directory: string,
   globalDirectory = path.join(directory, "global"),
   projectDirectory = directory,
   vcs?: Project.Vcs,
   watcher?: Layer.Layer<Watcher.Service>,
+  credentialNode = emptyCredentialNode,
+  wellknownNode = emptyWellknownNode,
 ) {
   const locationLayer = Layer.succeed(
     Location.Service,
@@ -45,6 +83,8 @@ function testLayer(
   return AppNodeBuilder.build(LayerNode.group([Config.node, EventV2.node]), [
     [Location.node, locationLayer],
     [Global.node, Global.layerWith({ config: globalDirectory, home: path.join(globalDirectory, "home") })],
+    [Credential.node, credentialNode],
+    [WellKnown.node, wellknownNode],
     ...(watcher ? ([[Watcher.node, watcher]] as const) : []),
   ])
 }
@@ -125,6 +165,86 @@ describe("Config", () => {
     }),
   )
 
+  it.live("loads authenticated wellknown config at highest priority", () =>
+    Effect.acquireUseRelease(
+      Effect.promise(() => tmpdir()),
+      (tmp) =>
+        Effect.gen(function* () {
+          const global = path.join(tmp.path, "global")
+          const project = path.join(tmp.path, "project")
+          yield* Effect.promise(async () => {
+            await fs.mkdir(global, { recursive: true })
+            await fs.mkdir(project, { recursive: true })
+            await fs.writeFile(path.join(global, "opencode.json"), JSON.stringify({ shell: "global" }))
+            await fs.writeFile(path.join(project, "opencode.json"), JSON.stringify({ shell: "project" }))
+          })
+
+          const integrationID = Integration.ID.make("https://example.com")
+          let key = "secret"
+          const credentialNode = makeGlobalNode({
+            service: Credential.Service,
+            layer: Layer.succeed(
+              Credential.Service,
+              Credential.Service.of({
+                all: () => Effect.die("unused Credential.all"),
+                list: () =>
+                  Effect.succeed([
+                    new Credential.Info({
+                      id: Credential.ID.create(),
+                      integrationID,
+                      label: "default",
+                      value: Credential.Key.make({ type: "key", key }),
+                    }),
+                  ]),
+                get: () => Effect.die("unused Credential.get"),
+                create: () => Effect.die("unused Credential.create"),
+                update: () => Effect.die("unused Credential.update"),
+                remove: () => Effect.die("unused Credential.remove"),
+              }),
+            ),
+            deps: [],
+          })
+          const entry: WellKnown.Entry = {
+            origin: "https://example.com",
+            integrationID,
+            manifest: { auth: { command: ["login"], env: "TOKEN" } },
+          }
+          const wellknownNode = makeGlobalNode({
+            service: WellKnown.Service,
+            layer: Layer.succeed(
+              WellKnown.Service,
+              WellKnown.Service.of({
+                entries: () => Effect.succeed([entry]),
+                snapshot: () => [entry],
+                refresh: () => Effect.succeed(false),
+                add: () => Effect.die("unused Wellknown.add"),
+                remove: () => Effect.die("unused Wellknown.remove"),
+                resolve: (_entry, variables) => Effect.succeed([{ shell: variables.TOKEN }]),
+              }),
+            ),
+            deps: [],
+          })
+
+          return yield* Effect.gen(function* () {
+            const config = yield* Config.Service
+            const events = yield* EventV2.Service
+            expect(Config.latest(yield* config.entries(), "shell")).toBe("secret")
+            const updated = yield* events
+              .subscribe(ConfigSchema.Event.Updated)
+              .pipe(Stream.take(1), Stream.runCollect, Effect.forkScoped)
+            yield* Effect.yieldNow
+            key = "next"
+            yield* events.publish(Integration.Event.ConnectionUpdated, { integrationID })
+            expect(yield* Fiber.join(updated)).toHaveLength(1)
+            expect(Config.latest(yield* config.entries(), "shell")).toBe("next")
+          }).pipe(
+            Effect.provide(testLayer(project, global, project, undefined, undefined, credentialNode, wellknownNode)),
+          )
+        }),
+      (tmp) => Effect.promise(() => tmp[Symbol.asyncDispose]()),
+    ),
+  )
+
   it.effect("detects v1 configuration from any v1-only top-level key", () =>
     Effect.sync(() => {
       expect(ConfigMigrateV1.isV1({ snapshot: false })).toBe(true)
@@ -163,6 +283,31 @@ describe("Config", () => {
     }),
   )
 
+  it.effect("migrates the v1 experimental subagent depth", () =>
+    Effect.sync(() => {
+      expect(ConfigMigrateV1.migrate({ experimental: { subagent_depth: 2 } }).experimental?.subagent_depth).toBe(2)
+    }),
+  )
+
+  it.effect("migrates v1 provider lists to policies", () =>
+    Effect.sync(() => {
+      expect(
+        ConfigMigrateV1.migrate({
+          enabled_providers: ["anthropic", "openai"],
+          disabled_providers: ["openai"],
+        }).experimental?.policies,
+      ).toEqual([
+        { action: "provider.use", resource: "*", effect: "deny" },
+        { action: "provider.use", resource: "anthropic", effect: "allow" },
+        { action: "provider.use", resource: "openai", effect: "allow" },
+        { action: "provider.use", resource: "openai", effect: "deny" },
+      ])
+      expect(ConfigMigrateV1.migrate({ enabled_providers: [] }).experimental?.policies).toEqual([
+        { action: "provider.use", resource: "*", effect: "deny" },
+      ])
+    }),
+  )
+
   it.effect("migrates v1 provider setup options into AISDK settings", () =>
     Effect.sync(() => {
       const migrated = ConfigMigrateV1.migrate({
@@ -282,9 +427,7 @@ describe("Config", () => {
             const config = yield* Config.Service
             yield* config.entries()
 
-            expect(targets).toEqual([
-              { type: "directory", path: AbsolutePath.make(path.join(tmp.path, "global")) },
-            ])
+            expect(targets).toEqual([{ type: "directory", path: AbsolutePath.make(path.join(tmp.path, "global")) }])
           }).pipe(Effect.provide(testLayer(tmp.path, undefined, undefined, undefined, watcher)))
         }),
       ),

packages/core/test/config/policy.test.ts

@@ -0,0 +1,95 @@
+import { describe, expect } from "bun:test"
+import { Config as ConfigSchema } from "@opencode-ai/schema/config"
+import { Catalog } from "@opencode-ai/core/catalog"
+import { Config } from "@opencode-ai/core/config"
+import { ConfigPolicyPlugin } from "@opencode-ai/core/config/plugin/policy"
+import { EventV2 } from "@opencode-ai/core/event"
+import { PluginV2 } from "@opencode-ai/core/plugin"
+import { PluginHost } from "@opencode-ai/core/plugin/host"
+import { ProviderV2 } from "@opencode-ai/core/provider"
+import { Effect, Schema } from "effect"
+import { testEffect } from "../lib/effect"
+import { PluginTestLayer } from "../plugin/fixture"
+
+const it = testEffect(PluginTestLayer)
+const decode = Schema.decodeUnknownSync(Config.Info)
+
+const policies = (...items: { effect: "allow" | "deny"; resource: string }[]) =>
+  new Config.Document({
+    type: "document",
+    info: decode({
+      experimental: {
+        policies: items.map((item) => ({ action: "provider.use", ...item })),
+      },
+    }),
+  })
+
+const addPlugin = Effect.fn(function* (entries: () => Config.Entry[]) {
+  const plugin = yield* PluginV2.Service
+  const host = yield* PluginHost.make(plugin)
+  yield* ConfigPolicyPlugin.Plugin.effect(host).pipe(
+    Effect.provideService(Config.Service, Config.Service.of({ entries: () => Effect.sync(entries) })),
+  )
+})
+
+describe("ConfigPolicyPlugin.Plugin", () => {
+  it.effect("filters plugin-provided providers with ordered wildcard policies", () =>
+    Effect.gen(function* () {
+      const catalog = yield* Catalog.Service
+      yield* catalog.transform((catalog) => {
+        catalog.provider.update(ProviderV2.ID.openai, () => {})
+        catalog.provider.update(ProviderV2.ID.anthropic, () => {})
+        catalog.provider.update(ProviderV2.ID.make("company-internal"), () => {})
+      })
+      yield* addPlugin(() => [
+        policies(
+          { effect: "deny", resource: "*" },
+          { effect: "allow", resource: "anthropic" },
+          { effect: "allow", resource: "company-*" },
+        ),
+      ])
+
+      expect(yield* catalog.provider.get(ProviderV2.ID.openai)).toBeUndefined()
+      expect(yield* catalog.provider.get(ProviderV2.ID.anthropic)).toBeDefined()
+      expect(yield* catalog.provider.get(ProviderV2.ID.make("company-internal"))).toBeDefined()
+    }),
+  )
+
+  it.effect("prevents project policy from overriding user-global policy", () =>
+    Effect.gen(function* () {
+      const catalog = yield* Catalog.Service
+      yield* catalog.transform((catalog) => catalog.provider.update(ProviderV2.ID.openai, () => {}))
+      yield* addPlugin(() => [
+        policies({ effect: "deny", resource: "openai" }),
+        policies({ effect: "allow", resource: "openai" }),
+      ])
+
+      expect(yield* catalog.provider.get(ProviderV2.ID.openai)).toBeUndefined()
+    }),
+  )
+
+  it.live("reloads changed policies", () =>
+    Effect.gen(function* () {
+      const catalog = yield* Catalog.Service
+      const events = yield* EventV2.Service
+      let entries: Config.Entry[] = [policies({ effect: "deny", resource: "openai" })]
+      yield* catalog.transform((catalog) => catalog.provider.update(ProviderV2.ID.openai, () => {}))
+      yield* addPlugin(() => entries)
+      expect(yield* catalog.provider.get(ProviderV2.ID.openai)).toBeUndefined()
+
+      entries = [policies({ effect: "allow", resource: "openai" })]
+      yield* events.publish(ConfigSchema.Event.Updated, {})
+      yield* waitUntil(
+        catalog.provider.get(ProviderV2.ID.openai).pipe(Effect.map((provider) => provider !== undefined)),
+      )
+    }),
+  )
+})
+
+const waitUntil = Effect.fnUntraced(function* (condition: Effect.Effect<boolean>) {
+  for (let attempt = 0; attempt < 200; attempt++) {
+    if (yield* condition) return
+    yield* Effect.sleep("10 millis")
+  }
+  return yield* Effect.die("Timed out waiting for policy reload")
+})

packages/core/test/instruction-state.test.ts

@@ -0,0 +1,425 @@
+import { describe, expect } from "bun:test"
+import { and, asc, eq } from "drizzle-orm"
+import { Effect, Schema } from "effect"
+import { Database } from "@opencode-ai/core/database/database"
+import { AppNodeBuilder } from "@opencode-ai/core/effect/app-node-builder"
+import { LayerNode } from "@opencode-ai/core/effect/layer-node"
+import { EventV2 } from "@opencode-ai/core/event"
+import { EventTable } from "@opencode-ai/core/event/sql"
+import { Instructions } from "@opencode-ai/core/instructions"
+import { Project } from "@opencode-ai/core/project"
+import { ProjectTable } from "@opencode-ai/core/project/sql"
+import { AbsolutePath } from "@opencode-ai/core/schema"
+import { InstructionState } from "@opencode-ai/core/session/instruction-state"
+import { SessionProjector } from "@opencode-ai/core/session/projector"
+import { SessionSchema } from "@opencode-ai/core/session/schema"
+import { InstructionBlobTable, InstructionStateTable, SessionTable } from "@opencode-ai/core/session/sql"
+import { testEffect } from "./lib/effect"
+
+const it = testEffect(AppNodeBuilder.build(LayerNode.group([Database.node, EventV2.node, SessionProjector.node])))
+
+const source = (name: string, read: Effect.Effect<string | Instructions.Unavailable | Instructions.Removed>) =>
+  Instructions.make({
+    key: Instructions.Key.make(name),
+    codec: Schema.toCodecJson(Schema.String),
+    read,
+    render: {
+      initial: String,
+      changed: (_previous, current) => current,
+      removed: (previous) => `Removed ${previous}`,
+    },
+  })
+
+const setup = (sessionID: SessionSchema.ID) =>
+  Effect.gen(function* () {
+    const { db } = yield* Database.Service
+    yield* db
+      .insert(ProjectTable)
+      .values({ id: Project.ID.global, worktree: AbsolutePath.make("/project"), sandboxes: [] })
+      .onConflictDoNothing()
+      .run()
+      .pipe(Effect.orDie)
+    yield* db
+      .insert(SessionTable)
+      .values({
+        id: sessionID,
+        project_id: Project.ID.global,
+        slug: "instruction-state-test",
+        directory: "/project",
+        title: "Instruction state test",
+        version: "test",
+      })
+      .run()
+      .pipe(Effect.orDie)
+    return { db, events: yield* EventV2.Service }
+  })
+
+const instructionEvents = (db: Database.Interface["db"], sessionID: SessionSchema.ID) =>
+  db
+    .select()
+    .from(EventTable)
+    .where(and(eq(EventTable.aggregate_id, sessionID), eq(EventTable.type, "session.instructions.updated.2")))
+    .orderBy(asc(EventTable.seq))
+    .all()
+    .pipe(Effect.orDie)
+
+const preview = (db: Database.Interface["db"], sessionID: SessionSchema.ID, instructions: Instructions.Instructions) =>
+  Instructions.read(instructions).pipe(
+    Effect.flatMap((observed) => InstructionState.preview(db, sessionID, instructions, observed)),
+  )
+
+describe("InstructionState", () => {
+  it.effect("observes each source once without publishing events or inserting blobs", () =>
+    Effect.gen(function* () {
+      const sessionID = SessionSchema.ID.make("ses_instruction_observe")
+      const { db, events } = yield* setup(sessionID)
+      const reads = { first: 0, second: 0 }
+      const instructions = Instructions.combine([
+        source(
+          "test/first",
+          Effect.sync(() => {
+            reads.first++
+            return "first"
+          }),
+        ),
+        source(
+          "test/second",
+          Effect.sync(() => {
+            reads.second++
+            return "second"
+          }),
+        ),
+      ])
+      const published: EventV2.Payload[] = []
+      const unsubscribe = yield* events.listen((event) =>
+        Effect.sync(() => {
+          if (event.type === "session.instructions.updated") published.push(event)
+        }),
+      )
+
+      const observation = yield* InstructionState.observe(db, instructions, sessionID)
+      yield* unsubscribe
+
+      expect(reads).toEqual({ first: 1, second: 1 })
+      expect(observation).toEqual({
+        sessionID,
+        initial: true,
+        current: {
+          "test/first": Instructions.hash("first"),
+          "test/second": Instructions.hash("second"),
+        },
+        delta: {
+          "test/first": Instructions.hash("first"),
+          "test/second": Instructions.hash("second"),
+        },
+        blobs: {
+          [Instructions.hash("first")]: "first",
+          [Instructions.hash("second")]: "second",
+        },
+      })
+      expect(published).toEqual([])
+      expect(yield* instructionEvents(db, sessionID)).toEqual([])
+      expect(yield* db.select().from(InstructionBlobTable).all().pipe(Effect.orDie)).toEqual([])
+    }),
+  )
+
+  it.effect("commits initial metadata and changed and removed deltas without rereading sources", () =>
+    Effect.gen(function* () {
+      const sessionID = SessionSchema.ID.make("ses_instruction_commit")
+      const { db, events } = yield* setup(sessionID)
+      let current = "initial"
+      let retired: string | Instructions.Removed = "retired"
+      let reads = 0
+      const instructions = Instructions.combine([
+        source(
+          "test/current",
+          Effect.sync(() => {
+            reads++
+            return current
+          }),
+        ),
+        source(
+          "test/retired",
+          Effect.sync(() => {
+            reads++
+            return retired
+          }),
+        ),
+      ])
+      const published: EventV2.Payload[] = []
+      const unsubscribe = yield* events.listen((event) =>
+        Effect.sync(() => {
+          if (event.type === "session.instructions.updated") published.push(event)
+        }),
+      )
+
+      const initial = yield* InstructionState.observe(db, instructions, sessionID)
+      expect(reads).toBe(2)
+      yield* InstructionState.commit(db, events, initial)
+      expect(reads).toBe(2)
+
+      current = "changed"
+      retired = Instructions.removed
+      const changed = yield* InstructionState.observe(db, instructions, sessionID)
+      expect(reads).toBe(4)
+      expect(changed).toMatchObject({
+        sessionID,
+        initial: false,
+        current: { "test/current": Instructions.hash("changed") },
+        delta: {
+          "test/current": Instructions.hash("changed"),
+          "test/retired": "removed",
+        },
+        blobs: { [Instructions.hash("changed")]: "changed" },
+      })
+      yield* InstructionState.commit(db, events, changed)
+      expect(reads).toBe(4)
+      yield* unsubscribe
+
+      expect(published).toHaveLength(2)
+      expect(published[0]?.metadata).toEqual({ instructions: { initial: true } })
+      expect(published[1]?.metadata).toBeUndefined()
+      expect((yield* instructionEvents(db, sessionID)).map((event) => event.data.delta)).toEqual([
+        {
+          "test/current": Instructions.hash("initial"),
+          "test/retired": Instructions.hash("retired"),
+        },
+        {
+          "test/current": Instructions.hash("changed"),
+          "test/retired": "removed",
+        },
+      ])
+      expect(yield* db.select().from(InstructionStateTable).get().pipe(Effect.orDie)).toMatchObject({
+        initial_values: {
+          "test/current": Instructions.hash("initial"),
+          "test/retired": Instructions.hash("retired"),
+        },
+        current_values: { "test/current": Instructions.hash("changed") },
+      })
+      expect(
+        Object.fromEntries(
+          (yield* db.select().from(InstructionBlobTable).all().pipe(Effect.orDie)).map((row) => [row.hash, row.value]),
+        ),
+      ).toEqual({
+        [Instructions.hash("initial")]: "initial",
+        [Instructions.hash("retired")]: "retired",
+        [Instructions.hash("changed")]: "changed",
+      })
+    }),
+  )
+
+  it.effect("keeps no-op observations free of events and blobs", () =>
+    Effect.gen(function* () {
+      const sessionID = SessionSchema.ID.make("ses_instruction_noop")
+      const { db, events } = yield* setup(sessionID)
+      const instructions = source("test/context", Effect.succeed("unchanged"))
+      yield* InstructionState.prepare(db, events, instructions, sessionID)
+      const beforeEvents = yield* instructionEvents(db, sessionID)
+      const beforeBlobs = yield* db.select().from(InstructionBlobTable).all().pipe(Effect.orDie)
+
+      const observation = yield* InstructionState.observe(db, instructions, sessionID)
+      expect(observation).toEqual({
+        sessionID,
+        initial: false,
+        current: { "test/context": Instructions.hash("unchanged") },
+        delta: {},
+        blobs: {},
+      })
+      yield* InstructionState.commit(db, events, observation)
+
+      expect(yield* instructionEvents(db, sessionID)).toEqual(beforeEvents)
+      expect(yield* db.select().from(InstructionBlobTable).all().pipe(Effect.orDie)).toEqual(beforeBlobs)
+    }),
+  )
+
+  it.effect("assembles a fresh private update without repairing a missing cache", () =>
+    Effect.gen(function* () {
+      const sessionID = SessionSchema.ID.make("ses_instruction_generate")
+      const { db, events } = yield* setup(sessionID)
+      let value = "Initial context"
+      const instructions = source(
+        "test/context",
+        Effect.sync(() => value),
+      )
+      yield* InstructionState.prepare(db, events, instructions, sessionID)
+      yield* db
+        .delete(InstructionStateTable)
+        .where(eq(InstructionStateTable.session_id, sessionID))
+        .run()
+        .pipe(Effect.orDie)
+      value = "Changed context"
+      const beforeEvents = yield* instructionEvents(db, sessionID)
+      const beforeBlobs = yield* db.select().from(InstructionBlobTable).all().pipe(Effect.orDie)
+
+      const assembled = yield* preview(db, sessionID, instructions)
+
+      expect(assembled).toEqual({ initial: "Initial context", updates: [], update: "Changed context" })
+      expect(yield* instructionEvents(db, sessionID)).toEqual(beforeEvents)
+      expect(yield* db.select().from(InstructionBlobTable).all().pipe(Effect.orDie)).toEqual(beforeBlobs)
+      expect(
+        yield* db
+          .select()
+          .from(InstructionStateTable)
+          .where(eq(InstructionStateTable.session_id, sessionID))
+          .get()
+          .pipe(Effect.orDie),
+      ).toBeUndefined()
+    }),
+  )
+
+  it.effect("reads through a stale cache without repairing it", () =>
+    Effect.gen(function* () {
+      const sessionID = SessionSchema.ID.make("ses_instruction_generate_stale")
+      const { db, events } = yield* setup(sessionID)
+      let value = "Initial context"
+      const instructions = source(
+        "test/context",
+        Effect.sync(() => value),
+      )
+      yield* InstructionState.prepare(db, events, instructions, sessionID)
+      value = "Committed update"
+      yield* InstructionState.prepare(db, events, instructions, sessionID)
+      yield* db
+        .update(InstructionStateTable)
+        .set({ through_seq: 0, current_values: { "test/context": Instructions.hash("Initial context") } })
+        .where(eq(InstructionStateTable.session_id, sessionID))
+        .run()
+        .pipe(Effect.orDie)
+      value = "Private update"
+      const beforeEvents = yield* instructionEvents(db, sessionID)
+      const beforeBlobs = yield* db.select().from(InstructionBlobTable).all().pipe(Effect.orDie)
+      const beforeState = yield* db.select().from(InstructionStateTable).get().pipe(Effect.orDie)
+
+      const assembled = yield* preview(db, sessionID, instructions)
+
+      expect(assembled.initial).toBe("Initial context")
+      expect(assembled.updates.map((entry) => entry.message.text)).toEqual(["Committed update"])
+      expect(assembled.update).toBe("Private update")
+      expect(yield* instructionEvents(db, sessionID)).toEqual(beforeEvents)
+      expect(yield* db.select().from(InstructionBlobTable).all().pipe(Effect.orDie)).toEqual(beforeBlobs)
+      expect(yield* db.select().from(InstructionStateTable).get().pipe(Effect.orDie)).toEqual(beforeState)
+    }),
+  )
+
+  it.effect("assembles initial instructions without persisting a baseline", () =>
+    Effect.gen(function* () {
+      const sessionID = SessionSchema.ID.make("ses_instruction_generate_initial")
+      const { db } = yield* setup(sessionID)
+      const instructions = source("test/context", Effect.succeed("Initial context"))
+
+      expect(yield* preview(db, sessionID, instructions)).toEqual({
+        initial: "Initial context",
+        updates: [],
+        update: "",
+      })
+      expect(yield* instructionEvents(db, sessionID)).toEqual([])
+      expect(yield* db.select().from(InstructionBlobTable).all().pipe(Effect.orDie)).toEqual([])
+      expect(yield* db.select().from(InstructionStateTable).get().pipe(Effect.orDie)).toBeUndefined()
+    }),
+  )
+
+  it.effect("retains a committed value when fresh instructions are unavailable", () =>
+    Effect.gen(function* () {
+      const sessionID = SessionSchema.ID.make("ses_instruction_generate_unavailable")
+      const { db, events } = yield* setup(sessionID)
+      let value: string | Instructions.Unavailable = "Committed context"
+      const instructions = source(
+        "test/context",
+        Effect.sync(() => value),
+      )
+      yield* InstructionState.prepare(db, events, instructions, sessionID)
+      value = Instructions.unavailable
+      const beforeEvents = yield* instructionEvents(db, sessionID)
+      const beforeBlobs = yield* db.select().from(InstructionBlobTable).all().pipe(Effect.orDie)
+      const beforeState = yield* db.select().from(InstructionStateTable).get().pipe(Effect.orDie)
+
+      expect(yield* preview(db, sessionID, instructions)).toEqual({
+        initial: "Committed context",
+        updates: [],
+        update: "",
+      })
+      expect(yield* instructionEvents(db, sessionID)).toEqual(beforeEvents)
+      expect(yield* db.select().from(InstructionBlobTable).all().pipe(Effect.orDie)).toEqual(beforeBlobs)
+      expect(yield* db.select().from(InstructionStateTable).get().pipe(Effect.orDie)).toEqual(beforeState)
+    }),
+  )
+
+  it.effect("blocks an unavailable initial instruction without persisting a baseline", () =>
+    Effect.gen(function* () {
+      const sessionID = SessionSchema.ID.make("ses_instruction_generate_blocked")
+      const { db } = yield* setup(sessionID)
+      const instructions = source("test/context", Effect.succeed(Instructions.unavailable))
+
+      const error = yield* preview(db, sessionID, instructions).pipe(Effect.flip)
+
+      expect(error).toBeInstanceOf(Instructions.InitializationBlocked)
+      expect(error.keys).toEqual([Instructions.Key.make("test/context")])
+      expect(yield* instructionEvents(db, sessionID)).toEqual([])
+      expect(yield* db.select().from(InstructionBlobTable).all().pipe(Effect.orDie)).toEqual([])
+      expect(yield* db.select().from(InstructionStateTable).get().pipe(Effect.orDie)).toBeUndefined()
+    }),
+  )
+
+  it.effect("keeps prepare equivalent to observe followed by commit", () =>
+    Effect.gen(function* () {
+      const observedSessionID = SessionSchema.ID.make("ses_instruction_composed")
+      const preparedSessionID = SessionSchema.ID.make("ses_instruction_prepared")
+      const { db, events } = yield* setup(observedSessionID)
+      yield* setup(preparedSessionID)
+      let value: string | Instructions.Removed = "initial"
+      let observedReads = 0
+      let preparedReads = 0
+      const observedInstructions = source(
+        "test/context",
+        Effect.sync(() => {
+          observedReads++
+          return value
+        }),
+      )
+      const preparedInstructions = source(
+        "test/context",
+        Effect.sync(() => {
+          preparedReads++
+          return value
+        }),
+      )
+
+      for (const next of ["initial", "changed", "changed", Instructions.removed] as const) {
+        value = next
+        yield* InstructionState.observe(db, observedInstructions, observedSessionID).pipe(
+          Effect.flatMap((observation) => InstructionState.commit(db, events, observation)),
+        )
+        yield* InstructionState.prepare(db, events, preparedInstructions, preparedSessionID)
+      }
+
+      expect(observedReads).toBe(4)
+      expect(preparedReads).toBe(4)
+      expect((yield* instructionEvents(db, observedSessionID)).map((event) => event.data.delta)).toEqual(
+        (yield* instructionEvents(db, preparedSessionID)).map((event) => event.data.delta),
+      )
+      const states = yield* db.select().from(InstructionStateTable).orderBy(asc(InstructionStateTable.session_id)).all()
+      expect(states).toHaveLength(2)
+      expect(
+        states.map((state) => ({
+          epoch_start: state.epoch_start,
+          through_seq: state.through_seq,
+          initial_values: state.initial_values,
+          current_values: state.current_values,
+        })),
+      ).toEqual([
+        {
+          epoch_start: 0,
+          through_seq: 2,
+          initial_values: { "test/context": Instructions.hash("initial") },
+          current_values: {},
+        },
+        {
+          epoch_start: 0,
+          through_seq: 2,
+          initial_values: { "test/context": Instructions.hash("initial") },
+          current_values: {},
+        },
+      ])
+    }),
+  )
+})

packages/core/test/integration.test.ts

@@ -181,7 +181,7 @@ describe("Integration", () => {
             command: [
               process.execPath,
               "-e",
-              'console.log("https://example.com/login"); await Bun.sleep(50); console.log("secret")',
+              'console.error("https://example.com/login"); await Bun.sleep(50); console.log("secret")',
             ],
           },
         }),
@@ -192,7 +192,7 @@ describe("Integration", () => {
         integrations.command.status({ integrationID, attemptID: attempt.attemptID }),
         (status) => status.status === "pending" && status.message?.includes("https://example.com/login") === true,
       )
-      expect(pending).toMatchObject({ status: "pending", message: "https://example.com/login" })
+      expect(pending).toMatchObject({ status: "pending", message: "https://example.com/login\n" })
 
       expect(
         yield* eventually(

packages/core/test/kv.test.ts

@@ -0,0 +1,25 @@
+import { describe, expect } from "bun:test"
+import { Effect } from "effect"
+import { KV } from "@opencode-ai/core/kv"
+import { LayerNode } from "@opencode-ai/core/effect/layer-node"
+import { testEffect } from "./lib/effect"
+
+const it = testEffect(LayerNode.compile(KV.node))
+
+describe("KV", () => {
+  it.effect("stores, replaces, and removes JSON values", () =>
+    Effect.gen(function* () {
+      const kv = yield* KV.Service
+      expect(yield* kv.get("wellknown:sources")).toBeUndefined()
+
+      yield* kv.set("wellknown:sources", ["https://example.com"])
+      expect(yield* kv.get("wellknown:sources")).toEqual(["https://example.com"])
+
+      yield* kv.set("wellknown:sources", ["https://example.com", "https://example.org"])
+      expect(yield* kv.get("wellknown:sources")).toEqual(["https://example.com", "https://example.org"])
+
+      yield* kv.remove("wellknown:sources")
+      expect(yield* kv.get("wellknown:sources")).toBeUndefined()
+    }),
+  )
+})

packages/core/test/lib/image.ts

@@ -0,0 +1,7 @@
+import { Image } from "@opencode-ai/core/image"
+import { Effect, Layer } from "effect"
+
+/** Passthrough resizer for tests that build ToolRegistry.node without a Location. */
+export const imagePassthrough = Layer.mock(Image.Service, {
+  normalize: (_resource, content) => Effect.succeed(content),
+})

packages/core/test/mcp-instructions.test.ts

@@ -0,0 +1,150 @@
+import { describe, expect } from "bun:test"
+import { Effect, Layer } from "effect"
+import { AgentV2 } from "@opencode-ai/core/agent"
+import { AppNodeBuilder } from "@opencode-ai/core/effect/app-node-builder"
+import { MCP } from "@opencode-ai/core/mcp/index"
+import { McpInstructions } from "@opencode-ai/core/mcp/instructions"
+import { PermissionV2 } from "@opencode-ai/core/permission"
+import { McpTool } from "@opencode-ai/core/tool/mcp"
+import { it } from "./lib/effect"
+import { readInitial, readUpdate } from "./lib/instructions"
+
+const build = AgentV2.ID.make("build")
+
+const selection = (permissions: PermissionV2.Ruleset = []) => {
+  const info = AgentV2.Info.make({ ...AgentV2.Info.empty(build), permissions })
+  return { id: info.id, info }
+}
+
+const instructions = (server: string, text: string) =>
+  new MCP.ServerInstructions({ server: MCP.ServerName.make(server), instructions: text })
+
+const tool = (server: string, name = "search") => new MCP.Tool({ server: MCP.ServerName.make(server), name })
+
+const layer = (catalog: () => MCP.ServerInstructions[], tools: () => MCP.Tool[]) =>
+  AppNodeBuilder.build(McpInstructions.node, [
+    [
+      MCP.node,
+      Layer.mock(MCP.Service, {
+        instructions: () => Effect.succeed(catalog()),
+        tools: () => Effect.succeed(tools()),
+      }),
+    ],
+  ])
+
+describe("McpInstructions", () => {
+  it.effect("renders instructions for servers with at least one permitted tool", () =>
+    Effect.gen(function* () {
+      const service = yield* McpInstructions.Service
+      const generation = yield* service
+        .load(
+          selection([
+            { action: McpTool.name("alpha", "restricted"), resource: "*", effect: "deny" },
+            { action: McpTool.name("hidden", "search"), resource: "*", effect: "deny" },
+          ]),
+        )
+        .pipe(Effect.flatMap(readInitial))
+
+      expect(generation.text).toBe(
+        [
+          "<mcp_instructions>",
+          '  <server name="alpha">',
+          '    Use tools from this server through `execute` under `tools["alpha"]`.',
+          "    Alpha line one",
+          "    Alpha line two",
+          "  </server>",
+          '  <server name="beta">',
+          '    Use tools from this server through `execute` under `tools["beta"]`.',
+          "    Beta instructions",
+          "  </server>",
+          "</mcp_instructions>",
+        ].join("\n"),
+      )
+    }).pipe(
+      Effect.provide(
+        layer(
+          () => [
+            instructions("beta", "Beta instructions"),
+            instructions("unused", "No tools"),
+            instructions("hidden", "Denied tool"),
+            instructions("alpha", "Alpha line one\nAlpha line two"),
+          ],
+          () => [tool("alpha"), tool("alpha", "restricted"), tool("beta"), tool("hidden")],
+        ),
+      ),
+    ),
+  )
+
+  it.effect("omits instructions when the agent cannot use execute", () =>
+    Effect.gen(function* () {
+      const service = yield* McpInstructions.Service
+      const generation = yield* service
+        .load(selection([{ action: "execute", resource: "*", effect: "deny" }]))
+        .pipe(Effect.flatMap(readInitial))
+
+      expect(generation.text).toBe("")
+    }).pipe(
+      Effect.provide(
+        layer(
+          () => [instructions("alpha", "Alpha instructions")],
+          () => [tool("alpha")],
+        ),
+      ),
+    ),
+  )
+
+  it.effect("renders additions, changes, and removal", () => {
+    let catalog = [instructions("alpha", "Alpha instructions")]
+    const tools = [tool("alpha"), tool("beta")]
+    return Effect.gen(function* () {
+      const service = yield* McpInstructions.Service
+      const initialized = yield* service.load(selection()).pipe(Effect.flatMap(readInitial))
+
+      catalog = [instructions("alpha", "Alpha instructions"), instructions("beta", "Beta instructions")]
+      const added = yield* readUpdate(yield* service.load(selection()), initialized)
+      expect(added.text).toBe(
+        [
+          "New MCP server instructions are available in addition to those previously listed:",
+          '  <server name="beta">',
+          '    Use tools from this server through `execute` under `tools["beta"]`.',
+          "    Beta instructions",
+          "  </server>",
+        ].join("\n"),
+      )
+
+      catalog = [instructions("alpha", "Updated alpha"), instructions("beta", "Beta instructions")]
+      const changed = yield* readUpdate(yield* service.load(selection()), added)
+      expect(changed.text).toBe(
+        [
+          "The available MCP server instructions have changed. This list supersedes the previous one.",
+          "<mcp_instructions>",
+          '  <server name="alpha">',
+          '    Use tools from this server through `execute` under `tools["alpha"]`.',
+          "    Updated alpha",
+          "  </server>",
+          '  <server name="beta">',
+          '    Use tools from this server through `execute` under `tools["beta"]`.',
+          "    Beta instructions",
+          "  </server>",
+          "</mcp_instructions>",
+        ].join("\n"),
+      )
+
+      catalog = [instructions("beta", "Beta instructions")]
+      const removed = yield* readUpdate(yield* service.load(selection()), changed)
+      expect(removed.text).toBe("Instructions for the following MCP servers are no longer available: alpha.")
+
+      catalog = []
+      expect((yield* readUpdate(yield* service.load(selection()), removed)).text).toBe(
+        "MCP server instructions are no longer available.",
+      )
+    }).pipe(
+      Effect.provide(
+        layer(
+          () => catalog,
+          () => tools,
+        ),
+      ),
+    )
+  })
+})

packages/core/test/mcp.test.ts

@@ -29,7 +29,9 @@ import { McpTool } from "@opencode-ai/core/tool/mcp"
 import { ToolRegistry } from "@opencode-ai/core/tool/registry"
 import { ToolOutputStore } from "@opencode-ai/core/tool-output-store"
 import { Deferred, Effect, Exit, Fiber, Layer, Schema, Stream } from "effect"
+import { Image } from "@opencode-ai/core/image"
 import { testEffect } from "./lib/effect"
+import { imagePassthrough } from "./lib/image"
 import { location } from "./fixture/location"
 import { settleTool, toolDefinitions, toolIdentity, waitForTool } from "./lib/tool"
 
@@ -250,6 +252,7 @@ const it = testEffect(
     [PermissionV2.node, permissions],
     [EventV2.node, events],
     [ToolOutputStore.node, ToolOutputStore.nodeWithoutConfig],
+    [Image.node, imagePassthrough],
   ]),
 )
 

packages/core/test/plugin/host.ts

@@ -97,6 +97,7 @@ export function host(overrides: Overrides = {}): PluginContext {
       get: overrides.session?.get ?? (() => Effect.die("unused session.get")),
       prompt: overrides.session?.prompt ?? (() => Effect.die("unused session.prompt")),
       command: overrides.session?.command ?? (() => Effect.die("unused session.command")),
+      synthetic: overrides.session?.synthetic ?? (() => Effect.die("unused session.synthetic")),
       interrupt: overrides.session?.interrupt ?? (() => Effect.die("unused session.interrupt")),
     },
   }

packages/core/test/plugin/promise.test.ts

@@ -1,6 +1,6 @@
 import { describe, expect } from "bun:test"
 import { Message, SystemPart } from "@opencode-ai/ai"
-import { Effect, Schema } from "effect"
+import { DateTime, Effect, Schema } from "effect"
 import { AgentV2 } from "@opencode-ai/core/agent"
 import { Catalog } from "@opencode-ai/core/catalog"
 import { ModelV2 } from "@opencode-ai/core/model"
@@ -10,6 +10,7 @@ import { PluginHost } from "@opencode-ai/core/plugin/host"
 import { PluginPromise } from "@opencode-ai/core/plugin/promise"
 import { SessionV2 } from "@opencode-ai/core/session"
 import { SessionMessage } from "@opencode-ai/core/session/message"
+import { SessionPending } from "@opencode-ai/core/session/pending"
 import { ToolRegistry } from "@opencode-ai/core/tool/registry"
 import { ProviderV2 } from "@opencode-ai/core/provider"
 import { Plugin } from "@opencode-ai/plugin/v2"
@@ -18,10 +19,63 @@ import { Model } from "@opencode-ai/schema/model"
 import { Provider } from "@opencode-ai/schema/provider"
 import { testEffect } from "../lib/effect"
 import { PluginTestLayer } from "./fixture"
+import { host as testHost } from "./host"
 
 const it = testEffect(PluginTestLayer)
 
 describe("fromPromise", () => {
+  it.effect("forwards synthetic session input", () =>
+    Effect.gen(function* () {
+      const input = {
+        sessionID: "ses_synthetic",
+        id: "msg_synthetic",
+        text: "Background work completed",
+        description: null,
+        metadata: { shellID: "shell_1" },
+        delivery: null,
+        resume: null,
+      }
+      let seen: unknown
+      const host = testHost({
+        session: {
+          synthetic: (value) => {
+            seen = value
+            return Effect.succeed(
+              SessionPending.Synthetic.make({
+                admittedSeq: 1,
+                id: SessionMessage.ID.make(input.id),
+                sessionID: SessionV2.ID.make(input.sessionID),
+                timeCreated: DateTime.makeUnsafe(0),
+                type: "synthetic",
+                data: {
+                  text: input.text,
+                  metadata: input.metadata,
+                },
+                delivery: "queue",
+              }),
+            )
+          },
+        },
+      })
+
+      yield* PluginPromise.fromPromise(
+        Plugin.define({
+          id: "promise-session-synthetic",
+          setup: async (ctx) => {
+            await ctx.session.synthetic(input)
+          },
+        }),
+      ).effect(host)
+
+      expect(seen).toEqual({
+        ...input,
+        description: undefined,
+        delivery: undefined,
+        resume: undefined,
+      })
+    }),
+  )
+
   it.effect("forwards standard client reads", () =>
     Effect.gen(function* () {
       const plugin = yield* PluginV2.Service

packages/core/test/reference-instructions.test.ts

@@ -3,18 +3,18 @@ import { Effect, Layer } from "effect"
 import { AppNodeBuilder } from "@opencode-ai/core/effect/app-node-builder"
 import { AbsolutePath } from "@opencode-ai/core/schema"
 import { Reference } from "@opencode-ai/core/reference"
-import { ReferenceGuidance } from "@opencode-ai/core/reference/guidance"
+import { ReferenceInstructions } from "@opencode-ai/core/reference/instructions"
 import { it } from "./lib/effect"
 import { readInitial, readUpdate } from "./lib/instructions"
 
-const guidanceLayer = (referenceLayer: Layer.Layer<Reference.Service>) =>
-  AppNodeBuilder.build(ReferenceGuidance.node, [[Reference.node, referenceLayer]])
+const instructionsLayer = (referenceLayer: Layer.Layer<Reference.Service>) =>
+  AppNodeBuilder.build(ReferenceInstructions.node, [[Reference.node, referenceLayer]])
 
-describe("ReferenceGuidance", () => {
+describe("ReferenceInstructions", () => {
   it.effect("lists available references in the instructions", () =>
     Effect.gen(function* () {
-      const guidance = yield* ReferenceGuidance.Service
-      const generation = yield* readInitial(yield* guidance.load())
+      const instructions = yield* ReferenceInstructions.Service
+      const generation = yield* readInitial(yield* instructions.load())
 
       expect(generation.text).toContain("<available_references>")
       expect(generation.text).toContain("<name>docs</name>")
@@ -22,7 +22,7 @@ describe("ReferenceGuidance", () => {
       expect(generation.text).toContain("<description>Use for product documentation</description>")
     }).pipe(
       Effect.provide(
-        guidanceLayer(
+        instructionsLayer(
           Layer.mock(Reference.Service, {
             list: () =>
               Effect.succeed([
@@ -43,22 +43,22 @@ describe("ReferenceGuidance", () => {
     ),
   )
 
-  it.effect("omits guidance when no references are available", () =>
+  it.effect("omits instructions when no references are available", () =>
     Effect.gen(function* () {
-      const guidance = yield* ReferenceGuidance.Service
-      const generation = yield* readInitial(yield* guidance.load())
+      const instructions = yield* ReferenceInstructions.Service
+      const generation = yield* readInitial(yield* instructions.load())
       expect(generation.text).toBe("")
-    }).pipe(Effect.provide(guidanceLayer(Layer.mock(Reference.Service, { list: () => Effect.succeed([]) })))),
+    }).pipe(Effect.provide(instructionsLayer(Layer.mock(Reference.Service, { list: () => Effect.succeed([]) })))),
   )
 
   it.effect("omits references without descriptions", () =>
     Effect.gen(function* () {
-      const guidance = yield* ReferenceGuidance.Service
-      const generation = yield* readInitial(yield* guidance.load())
+      const instructions = yield* ReferenceInstructions.Service
+      const generation = yield* readInitial(yield* instructions.load())
       expect(generation.text).toBe("")
     }).pipe(
       Effect.provide(
-        guidanceLayer(
+        instructionsLayer(
           Layer.mock(Reference.Service, {
             list: () =>
               Effect.succeed([
@@ -84,11 +84,11 @@ describe("ReferenceGuidance", () => {
       })
     let references = [reference("docs", "Use for product documentation")]
     return Effect.gen(function* () {
-      const guidance = yield* ReferenceGuidance.Service
-      const initialized = yield* readInitial(yield* guidance.load())
+      const instructions = yield* ReferenceInstructions.Service
+      const initialized = yield* readInitial(yield* instructions.load())
 
       references = [reference("docs", "Use for product documentation"), reference("examples", "Use for examples")]
-      const added = yield* readUpdate(yield* guidance.load(), initialized)
+      const added = yield* readUpdate(yield* instructions.load(), initialized)
       expect(added.text).toBe(
         [
           "New project references are available in addition to those previously listed:",
@@ -101,9 +101,11 @@ describe("ReferenceGuidance", () => {
       )
 
       references = [reference("examples", "Use for examples")]
-      expect((yield* readUpdate(yield* guidance.load(), added)).text).toBe(
+      expect((yield* readUpdate(yield* instructions.load(), added)).text).toBe(
         "The following project references are no longer available and must not be used: docs.",
       )
-    }).pipe(Effect.provide(guidanceLayer(Layer.mock(Reference.Service, { list: () => Effect.succeed(references) }))))
+    }).pipe(
+      Effect.provide(instructionsLayer(Layer.mock(Reference.Service, { list: () => Effect.succeed(references) }))),
+    )
   })
 })

packages/core/test/session-generate.test.ts

@@ -0,0 +1,303 @@
+import { expect } from "bun:test"
+import { LLMClient, LLMEvent, LLMResponse, Model, SystemPart, type LLMRequest } from "@opencode-ai/ai"
+import { OpenAIChat } from "@opencode-ai/ai/protocols"
+import { AgentV2 } from "@opencode-ai/core/agent"
+import { Database } from "@opencode-ai/core/database/database"
+import { AppNodeBuilder } from "@opencode-ai/core/effect/app-node-builder"
+import { llmClient } from "@opencode-ai/core/effect/app-node-platform"
+import { LayerNode } from "@opencode-ai/core/effect/layer-node"
+import { EventV2 } from "@opencode-ai/core/event"
+import { EventTable } from "@opencode-ai/core/event/sql"
+import { InstructionDiscovery } from "@opencode-ai/core/instruction-discovery"
+import { Instructions } from "@opencode-ai/core/instructions"
+import { InstructionBuiltIns } from "@opencode-ai/core/instructions/builtins"
+import { Location } from "@opencode-ai/core/location"
+import { McpInstructions } from "@opencode-ai/core/mcp/instructions"
+import { ModelV2 } from "@opencode-ai/core/model"
+import { Project } from "@opencode-ai/core/project"
+import { ProjectTable } from "@opencode-ai/core/project/sql"
+import { ProviderV2 } from "@opencode-ai/core/provider"
+import { ReferenceInstructions } from "@opencode-ai/core/reference/instructions"
+import { AbsolutePath } from "@opencode-ai/core/schema"
+import { SessionEvent } from "@opencode-ai/core/session/event"
+import { SessionGenerate } from "@opencode-ai/core/session/generate"
+import { SessionGenerateNode } from "@opencode-ai/core/session/generate-node"
+import { InstructionState } from "@opencode-ai/core/session/instruction-state"
+import { SessionMessage } from "@opencode-ai/core/session/message"
+import { SessionProjector } from "@opencode-ai/core/session/projector"
+import { SessionRunnerModel } from "@opencode-ai/core/session/runner/model"
+import { SessionSchema } from "@opencode-ai/core/session/schema"
+import {
+  InstructionBlobTable,
+  InstructionStateTable,
+  SessionMessageTable,
+  SessionPendingTable,
+  SessionTable,
+} from "@opencode-ai/core/session/sql"
+import { SessionStore } from "@opencode-ai/core/session/store"
+import { SkillInstructions } from "@opencode-ai/core/skill/instructions"
+import { PluginHooks } from "@opencode-ai/core/plugin/hooks"
+import { PluginSupervisor } from "@opencode-ai/core/plugin/supervisor"
+import { asc, eq } from "drizzle-orm"
+import { Effect, Layer, Schema, Stream } from "effect"
+import { testEffect } from "./lib/effect"
+
+const requests: LLMRequest[] = []
+let instruction: string | Instructions.Unavailable = "Initial context"
+const sessionID = SessionSchema.ID.make("ses_generate_test")
+
+const model = Model.make({ id: "generate-model", provider: "test", route: OpenAIChat.route })
+const client = Layer.mock(LLMClient.Service)({
+  prepare: () => Effect.die(new Error("unused")),
+  stream: () => Stream.die(new Error("unused")),
+  generate: (request) =>
+    Effect.sync(() => {
+      requests.push(request)
+      const response = LLMResponse.fromEvents([
+        LLMEvent.stepStart({ index: 0 }),
+        LLMEvent.textStart({ id: "generate" }),
+        LLMEvent.textDelta({ id: "generate", text: "Transient answer" }),
+        LLMEvent.textEnd({ id: "generate" }),
+        LLMEvent.stepFinish({ index: 0, reason: "stop", usage: { inputTokens: 100, outputTokens: 10 } }),
+        LLMEvent.finish({ reason: "stop" }),
+      ])
+      if (!response) throw new Error("Incomplete generate response")
+      return response
+    }),
+})
+const models = SessionRunnerModel.layerWith(() => Effect.succeed(SessionRunnerModel.resolved(model)))
+const builtins = Layer.mock(InstructionBuiltIns.Service, {
+  load: () =>
+    Effect.succeed(
+      Instructions.make({
+        key: Instructions.Key.make("test/context"),
+        codec: Schema.toCodecJson(Schema.String),
+        read: Effect.sync(() => instruction),
+        render: {
+          initial: String,
+          changed: (_previous, current) => current,
+        },
+      }),
+    ),
+})
+const discovery = Layer.mock(InstructionDiscovery.Service, { load: () => Effect.succeed(Instructions.empty) })
+const skills = Layer.mock(SkillInstructions.Service, { load: () => Effect.succeed(Instructions.empty) })
+const references = Layer.mock(ReferenceInstructions.Service, { load: () => Effect.succeed(Instructions.empty) })
+const mcp = Layer.mock(McpInstructions.Service, { load: () => Effect.succeed(Instructions.empty) })
+const plugins = Layer.mock(PluginSupervisor.Service, { flush: Effect.void })
+
+const it = testEffect(
+  AppNodeBuilder.build(
+    LayerNode.group([
+      Database.node,
+      EventV2.node,
+      SessionProjector.node,
+      SessionStore.node,
+      AgentV2.node,
+      InstructionBuiltIns.node,
+      PluginHooks.node,
+      SessionGenerateNode.node,
+    ]),
+    [
+      [llmClient, client],
+      [SessionRunnerModel.node, models],
+      [InstructionBuiltIns.node, builtins],
+      [InstructionDiscovery.node, discovery],
+      [SkillInstructions.node, skills],
+      [ReferenceInstructions.node, references],
+      [McpInstructions.node, mcp],
+      [PluginSupervisor.node, plugins],
+      [Location.node, Location.boundNode({ directory: AbsolutePath.make("/project") })],
+    ],
+  ),
+)
+
+const durableState = (db: Database.Interface["db"], sessionID: SessionSchema.ID) =>
+  Effect.all({
+    sequence: EventV2.latestSequence(db, sessionID),
+    events: db
+      .select()
+      .from(EventTable)
+      .where(eq(EventTable.aggregate_id, sessionID))
+      .orderBy(asc(EventTable.seq))
+      .all()
+      .pipe(Effect.orDie),
+    messages: db
+      .select()
+      .from(SessionMessageTable)
+      .where(eq(SessionMessageTable.session_id, sessionID))
+      .orderBy(asc(SessionMessageTable.seq))
+      .all()
+      .pipe(Effect.orDie),
+    pending: db
+      .select()
+      .from(SessionPendingTable)
+      .where(eq(SessionPendingTable.session_id, sessionID))
+      .orderBy(asc(SessionPendingTable.admitted_seq))
+      .all()
+      .pipe(Effect.orDie),
+    instructions: db
+      .select()
+      .from(InstructionStateTable)
+      .where(eq(InstructionStateTable.session_id, sessionID))
+      .get()
+      .pipe(Effect.orDie),
+    blobs: db.select().from(InstructionBlobTable).orderBy(asc(InstructionBlobTable.hash)).all().pipe(Effect.orDie),
+    session: db.select().from(SessionTable).where(eq(SessionTable.id, sessionID)).get().pipe(Effect.orDie),
+  })
+
+const userTexts = (request: LLMRequest) =>
+  request.messages.flatMap((message) =>
+    message.role === "user"
+      ? message.content.flatMap((content) => (content.type === "text" ? [content.text] : []))
+      : [],
+  )
+
+const setup = Effect.gen(function* () {
+  const { db } = yield* Database.Service
+  const events = yield* EventV2.Service
+  const agents = yield* AgentV2.Service
+  const instructionBuiltIns = yield* InstructionBuiltIns.Service
+  yield* agents.transform((draft) =>
+    draft.update(AgentV2.ID.make("build"), (agent) => {
+      agent.mode = "primary"
+    }),
+  )
+  yield* db
+    .insert(ProjectTable)
+    .values({ id: Project.ID.global, worktree: AbsolutePath.make("/project"), sandboxes: [] })
+    .run()
+    .pipe(Effect.orDie)
+  yield* db
+    .insert(SessionTable)
+    .values({
+      id: sessionID,
+      project_id: Project.ID.global,
+      slug: "generate-test",
+      directory: "/project",
+      title: "Generate test",
+      version: "test",
+      agent: AgentV2.ID.make("build"),
+    })
+    .run()
+    .pipe(Effect.orDie)
+  return { db, events, instructions: yield* instructionBuiltIns.load(sessionID) }
+})
+
+it.effect("generates from fresh settled Session context without durable mutation", () =>
+  Effect.gen(function* () {
+    requests.length = 0
+    instruction = "Initial context"
+    const { db, events, instructions } = yield* setup
+    yield* InstructionState.prepare(db, events, instructions, sessionID)
+    const existing = SessionMessage.ID.create()
+    yield* events.publish(SessionEvent.InputAdmitted, {
+      sessionID,
+      inputID: existing,
+      input: { type: "user", data: { text: "Existing durable context" }, delivery: "steer" },
+    })
+    yield* events.publish(SessionEvent.InputPromoted, { sessionID, inputID: existing })
+    const settledAssistant = SessionMessage.ID.create()
+    yield* events.publish(SessionEvent.Step.Started, {
+      sessionID,
+      assistantMessageID: settledAssistant,
+      agent: AgentV2.ID.make("build"),
+      model: { id: ModelV2.ID.make("generate-model"), providerID: ProviderV2.ID.make("test") },
+    })
+    yield* events.publish(SessionEvent.Text.Started, {
+      sessionID,
+      assistantMessageID: settledAssistant,
+      ordinal: 0,
+    })
+    yield* events.publish(SessionEvent.Text.Ended, {
+      sessionID,
+      assistantMessageID: settledAssistant,
+      ordinal: 0,
+      text: "Settled partial answer",
+    })
+    const activeAssistant = SessionMessage.ID.create()
+    yield* events.publish(SessionEvent.Step.Started, {
+      sessionID,
+      assistantMessageID: activeAssistant,
+      agent: AgentV2.ID.make("build"),
+      model: { id: ModelV2.ID.make("generate-model"), providerID: ProviderV2.ID.make("test") },
+    })
+    yield* events.publish(SessionEvent.Tool.Input.Started, {
+      sessionID,
+      assistantMessageID: activeAssistant,
+      callID: "active-call",
+      name: "echo",
+    })
+    yield* events.publish(SessionEvent.Tool.Input.Ended, {
+      sessionID,
+      assistantMessageID: activeAssistant,
+      callID: "active-call",
+      text: "{}",
+    })
+    yield* events.publish(SessionEvent.Tool.Called, {
+      sessionID,
+      assistantMessageID: activeAssistant,
+      callID: "active-call",
+      input: {},
+      executed: false,
+    })
+    yield* events.publish(SessionEvent.InputAdmitted, {
+      sessionID,
+      inputID: SessionMessage.ID.create(),
+      input: { type: "user", data: { text: "Queued input must remain invisible" }, delivery: "queue" },
+    })
+    instruction = "Changed context"
+    const before = yield* durableState(db, sessionID)
+    const hooks = yield* PluginHooks.Service
+    yield* hooks.register("session", "context", (event) =>
+      Effect.sync(() => {
+        event.system = [SystemPart.make("Hooked system"), ...event.system]
+      }),
+    )
+
+    const generate = yield* SessionGenerate.Service
+    const result = yield* generate.generate({ sessionID, prompt: "Summarize privately" })
+
+    expect(result).toBe("Transient answer")
+    expect(requests).toHaveLength(1)
+    expect(requests[0]?.model).toBe(model)
+    expect(requests[0]?.system[0]?.text).toBe("Hooked system")
+    expect(requests[0]?.system.map((part) => part.text)).toContain("Initial context")
+    expect(requests[0]?.http?.headers).toMatchObject({ "X-Session-Id": sessionID })
+    expect(requests[0]?.providerOptions).toMatchObject({ openai: { promptCacheKey: sessionID } })
+    expect(
+      requests[0]?.messages.flatMap((message) =>
+        message.role === "system"
+          ? message.content.flatMap((content) => (content.type === "text" ? [content.text] : []))
+          : [],
+      ),
+    ).toEqual(["Changed context"])
+    expect(userTexts(requests[0])).toEqual(["Existing durable context", "Summarize privately"])
+    expect(
+      requests[0]?.messages.flatMap((message) =>
+        message.role === "assistant"
+          ? message.content.flatMap((content) => (content.type === "text" ? [content.text] : []))
+          : [],
+      ),
+    ).toEqual(["Settled partial answer"])
+    expect(requests[0]?.tools).toEqual([])
+    expect(requests[0]?.toolChoice).toMatchObject({ type: "none" })
+    expect(yield* durableState(db, sessionID)).toEqual(before)
+  }),
+)
+
+it.effect("blocks unavailable initial instructions before generation", () =>
+  Effect.gen(function* () {
+    requests.length = 0
+    instruction = Instructions.unavailable
+    const { db } = yield* setup
+    const before = yield* durableState(db, sessionID)
+    const generate = yield* SessionGenerate.Service
+
+    const error = yield* generate.generate({ sessionID, prompt: "Summarize privately" }).pipe(Effect.flip)
+
+    expect(error).toBeInstanceOf(Instructions.InitializationBlocked)
+    expect(requests).toEqual([])
+    expect(yield* durableState(db, sessionID)).toEqual(before)
+  }),
+)

packages/core/test/session-prompt.test.ts

@@ -1,5 +1,5 @@
 import { describe, expect } from "bun:test"
-import { DateTime, Effect, Fiber, Layer, Schema, Stream } from "effect"
+import { DateTime, Effect, Fiber, Layer, LayerMap, Schema, Stream } from "effect"
 import { mkdtemp, rm } from "fs/promises"
 import { tmpdir } from "os"
 import path from "path"
@@ -24,7 +24,10 @@ import { SessionExecution } from "@opencode-ai/core/session/execution"
 import { SessionPending } from "@opencode-ai/core/session/pending"
 import { SessionPendingTable, SessionMessageTable, SessionTable } from "@opencode-ai/core/session/sql"
 import { SessionStore } from "@opencode-ai/core/session/store"
+import { LocationServiceMap } from "@opencode-ai/core/location-service-map"
+import type { LocationServices } from "@opencode-ai/core/location-services"
 import { testEffect } from "./lib/effect"
+import { imagePassthrough } from "./lib/image"
 
 const executionCalls: SessionV2.ID[] = []
 const interruptCalls: SessionV2.ID[] = []
@@ -49,10 +52,22 @@ const execution = Layer.succeed(
     awaitIdle: () => Effect.void,
   }),
 )
+const locations = Layer.effect(
+  LocationServiceMap.Service,
+  LayerMap.make(
+    () =>
+      // Attachment admission only needs the location-scoped Image service.
+      // oxlint-disable-next-line typescript-eslint/no-unsafe-type-assertion
+      imagePassthrough as unknown as Layer.Layer<LocationServices>,
+  ),
+)
 const it = testEffect(
   AppNodeBuilder.build(
     LayerNode.group([Database.node, EventV2.node, SessionProjector.node, SessionStore.node, SessionV2.node]),
-    [[SessionExecution.node, execution]],
+    [
+      [SessionExecution.node, execution],
+      [LocationServiceMap.node, locations],
+    ],
   ),
 )
 const sessionID = SessionV2.ID.make("ses_prompt_test")

packages/core/test/session-runner-model.test.ts

@@ -556,7 +556,10 @@ describe("SessionRunnerModel", () => {
       })
       const packages = [
         ["@opencode-ai/ai/providers/google-vertex", "accessToken"],
-        ["@opencode-ai/ai/providers/google-vertex/anthropic", "accessToken"],
+        ["@opencode-ai/ai/providers/google-vertex/gemini", "accessToken"],
+        ["@opencode-ai/ai/providers/google-vertex/chat", "accessToken"],
+        ["@opencode-ai/ai/providers/google-vertex/responses", "accessToken"],
+        ["@opencode-ai/ai/providers/google-vertex/messages", "accessToken"],
         ["@opencode-ai/ai/providers/anthropic", "authToken"],
         ["@opencode-ai/ai/providers/anthropic-compatible", "authToken"],
       ] as const

packages/core/test/session-runner-recorded.test.ts

@@ -33,9 +33,9 @@ import { Location } from "@opencode-ai/core/location"
 import { InstructionBuiltIns } from "@opencode-ai/core/instructions/builtins"
 import { InstructionDiscovery } from "@opencode-ai/core/instruction-discovery"
 import { Instructions } from "@opencode-ai/core/instructions"
-import { SkillGuidance } from "@opencode-ai/core/skill/guidance"
-import { ReferenceGuidance } from "@opencode-ai/core/reference/guidance"
-import { McpGuidance } from "@opencode-ai/core/mcp/guidance"
+import { SkillInstructions } from "@opencode-ai/core/skill/instructions"
+import { ReferenceInstructions } from "@opencode-ai/core/reference/instructions"
+import { McpInstructions } from "@opencode-ai/core/mcp/instructions"
 import { PluginSupervisor } from "@opencode-ai/core/plugin/supervisor"
 import { PluginHooks } from "@opencode-ai/core/plugin/hooks"
 import { SystemPromptPlugin } from "@opencode-ai/core/plugin/system-prompt"
@@ -76,9 +76,11 @@ const model = OpenAIChat.route
 const models = SessionRunnerModel.layerWith(() => Effect.succeed(SessionRunnerModel.resolved(model)))
 const systemContext = Layer.mock(InstructionBuiltIns.Service, { load: () => Effect.succeed(Instructions.empty) })
 const instructionContext = Layer.mock(InstructionDiscovery.Service, { load: () => Effect.succeed(Instructions.empty) })
-const skillGuidance = Layer.mock(SkillGuidance.Service, { load: () => Effect.succeed(Instructions.empty) })
-const referenceGuidance = Layer.mock(ReferenceGuidance.Service, { load: () => Effect.succeed(Instructions.empty) })
-const mcpGuidance = Layer.mock(McpGuidance.Service, { load: () => Effect.succeed(Instructions.empty) })
+const skillInstructions = Layer.mock(SkillInstructions.Service, { load: () => Effect.succeed(Instructions.empty) })
+const referenceInstructions = Layer.mock(ReferenceInstructions.Service, {
+  load: () => Effect.succeed(Instructions.empty),
+})
+const mcpInstructions = Layer.mock(McpInstructions.Service, { load: () => Effect.succeed(Instructions.empty) })
 const config = Layer.succeed(Config.Service, Config.Service.of({ entries: () => Effect.succeed([]) }))
 const pluginSupervisor = Layer.succeed(PluginSupervisor.Service, PluginSupervisor.Service.of({ flush: Effect.void }))
 const promptCatalog = Layer.mock(Catalog.Service, {
@@ -102,9 +104,9 @@ const runnerLayer = AppNodeBuilder.build(SessionRunnerLLM.node, [
   [InstructionBuiltIns.node, systemContext],
   [InstructionDiscovery.node, instructionContext],
   [Location.node, Location.boundNode({ directory: AbsolutePath.make("/project") })],
-  [SkillGuidance.node, skillGuidance],
-  [ReferenceGuidance.node, referenceGuidance],
-  [McpGuidance.node, mcpGuidance],
+  [SkillInstructions.node, skillInstructions],
+  [ReferenceInstructions.node, referenceInstructions],
+  [McpInstructions.node, mcpInstructions],
   [Config.node, config],
   [PermissionV2.node, permission],
   [ToolOutputStore.node, ToolOutputStore.nodeWithoutConfig],
@@ -140,8 +142,8 @@ const it = testEffect(
       SessionRunnerModel.node,
       InstructionBuiltIns.node,
       InstructionDiscovery.node,
-      SkillGuidance.node,
-      ReferenceGuidance.node,
+      SkillInstructions.node,
+      ReferenceInstructions.node,
       Config.node,
       Snapshot.node,
       SessionRunnerLLM.node,
@@ -156,8 +158,8 @@ const it = testEffect(
       [InstructionBuiltIns.node, systemContext],
       [InstructionDiscovery.node, instructionContext],
       [Location.node, Location.boundNode({ directory: AbsolutePath.make("/project") })],
-      [SkillGuidance.node, skillGuidance],
-      [ReferenceGuidance.node, referenceGuidance],
+      [SkillInstructions.node, skillInstructions],
+      [ReferenceInstructions.node, referenceInstructions],
       [Config.node, config],
       [Snapshot.node, Snapshot.noopLayer],
       [PluginSupervisor.node, pluginSupervisor],

packages/core/test/session-runner-tool-registry.test.ts

@@ -3,6 +3,7 @@ import { Tool } from "@opencode-ai/core/tool/tool"
 import { AgentV2 } from "@opencode-ai/core/agent"
 import type { PermissionV2 } from "@opencode-ai/core/permission"
 import { AppNodeBuilder } from "@opencode-ai/core/effect/app-node-builder"
+import { Image } from "@opencode-ai/core/image"
 import { SessionV2 } from "@opencode-ai/core/session"
 import { SessionMessage } from "@opencode-ai/core/session/message"
 import { ToolOutputStore } from "@opencode-ai/core/tool-output-store"
@@ -28,7 +29,28 @@ const outputStore = Layer.mock(ToolOutputStore.Service, {
     )
   },
 })
-const registryLayer = AppNodeBuilder.build(ToolRegistry.node, [[ToolOutputStore.node, outputStore]])
+const imageStore = Layer.mock(Image.Service, {
+  normalize: (resource, content) => {
+    if (resource === "corrupt.png") return Effect.fail(new Image.DecodeError({ resource }))
+    if (resource === "too-large.png")
+      return Effect.fail(
+        new Image.SizeError({
+          resource,
+          width: 9_000,
+          height: 9_000,
+          bytes: content.content.length,
+          maxWidth: 2_000,
+          maxHeight: 2_000,
+          maxBytes: 5,
+        }),
+      )
+    return Effect.succeed({ ...content, content: "bm9ybWFsaXplZA==", mime: "image/jpeg" })
+  },
+})
+const registryLayer = AppNodeBuilder.build(ToolRegistry.node, [
+  [ToolOutputStore.node, outputStore],
+  [Image.node, imageStore],
+])
 const it = testEffect(registryLayer)
 const identity = {
   agent: AgentV2.ID.make("build"),
@@ -255,6 +277,75 @@ describe("ToolRegistry", () => {
     }),
   )
 
+  it.effect("normalizes image tool output at settlement and drops unresizable images", () =>
+    Effect.gen(function* () {
+      const service = yield* ToolRegistry.Service
+      yield* service.register({
+        snapshot: Tool.make({
+          description: "Return images",
+          input: Schema.Struct({ text: Schema.String }),
+          output: Schema.Struct({ text: Schema.String }),
+          execute: ({ text }) => Effect.succeed({ text }),
+          toModelOutput: ({ output }) => [
+            { type: "file", data: "aW1hZ2U=", mime: "image/png", name: "frame.png" },
+            { type: "file", data: "aW1hZ2U=", mime: "image/png", name: "too-large.png" },
+            { type: "file", data: "aW1hZ2U=", mime: "image/png", name: "corrupt.png" },
+            { type: "text", text: output.text },
+          ],
+        }),
+      }, { codemode: false })
+
+      const settlement = yield* settleTool(service, call("snapshot"))
+      expect(settlement.output?.content).toEqual([
+        { type: "file", uri: "data:image/jpeg;base64,bm9ybWFsaXplZA==", mime: "image/jpeg", name: "frame.png" },
+        { type: "text", text: "snapshot" },
+        { type: "text", text: "[1 image omitted: could not be decoded.]" },
+        { type: "text", text: "[1 image omitted: could not be resized below the image size limit.]" },
+      ])
+    }),
+  )
+
+  it.effect("normalizes image progress content before it is published", () =>
+    Effect.gen(function* () {
+      const service = yield* ToolRegistry.Service
+      yield* service.register({
+        progressive: Tool.make({
+          description: "Emit image progress",
+          input: Schema.Struct({ text: Schema.String }),
+          output: Schema.Struct({ text: Schema.String }),
+          execute: ({ text }, context) =>
+            context
+              .progress({
+                structured: { stage: "capture" },
+                content: [
+                  { type: "file", data: "aW1hZ2U=", mime: "image/png", name: "frame.png" },
+                  { type: "file", data: "aW1hZ2U=", mime: "image/png", name: "too-large.png" },
+                ],
+              })
+              .pipe(Effect.as({ text })),
+        }),
+      }, { codemode: false })
+
+      const updates: ToolRegistry.Progress[] = []
+      yield* settleTool(service, {
+        ...call("progressive"),
+        progress: (update) =>
+          Effect.sync(() => {
+            updates.push(update)
+          }),
+      })
+      expect(updates).toEqual([
+        {
+          structured: { stage: "capture" },
+          content: [
+            { type: "file", uri: "data:image/jpeg;base64,bm9ybWFsaXplZA==", mime: "image/jpeg", name: "frame.png" },
+            { type: "text", text: "[1 image omitted: could not be resized below the image size limit.]" },
+          ],
+        },
+      ])
+    }),
+  )
+
   it.effect("enforces transformed codecs at execution and projection boundaries", () =>
     Effect.gen(function* () {
       const service = yield* ToolRegistry.Service

packages/core/test/session-runner.test.ts

@@ -63,9 +63,9 @@ import { SessionStore } from "@opencode-ai/core/session/store"
 import { Instructions } from "@opencode-ai/core/instructions"
 import { InstructionBuiltIns } from "@opencode-ai/core/instructions/builtins"
 import { InstructionDiscovery } from "@opencode-ai/core/instruction-discovery"
-import { SkillGuidance } from "@opencode-ai/core/skill/guidance"
-import { ReferenceGuidance } from "@opencode-ai/core/reference/guidance"
-import { McpGuidance } from "@opencode-ai/core/mcp/guidance"
+import { SkillInstructions } from "@opencode-ai/core/skill/instructions"
+import { ReferenceInstructions } from "@opencode-ai/core/reference/instructions"
+import { McpInstructions } from "@opencode-ai/core/mcp/instructions"
 import { ModelV2 } from "@opencode-ai/core/model"
 import { Location } from "@opencode-ai/core/location"
 import { ProviderV2 } from "@opencode-ai/core/provider"
@@ -318,7 +318,7 @@ const systemContext = Layer.mock(InstructionBuiltIns.Service, {
     ),
 })
 const instructionContext = Layer.mock(InstructionDiscovery.Service, { load: () => Effect.succeed(Instructions.empty) })
-const skillGuidance = Layer.mock(SkillGuidance.Service, {
+const skillInstructions = Layer.mock(SkillInstructions.Service, {
   load: (agent) =>
     Effect.succeed(
       skillBaselines.has(agent.id)
@@ -335,8 +335,10 @@ const skillGuidance = Layer.mock(SkillGuidance.Service, {
         : Instructions.empty,
     ),
 })
-const referenceGuidance = Layer.mock(ReferenceGuidance.Service, { load: () => Effect.succeed(Instructions.empty) })
-const mcpGuidance = Layer.mock(McpGuidance.Service, { load: () => Effect.succeed(Instructions.empty) })
+const referenceInstructions = Layer.mock(ReferenceInstructions.Service, {
+  load: () => Effect.succeed(Instructions.empty),
+})
+const mcpInstructions = Layer.mock(McpInstructions.Service, { load: () => Effect.succeed(Instructions.empty) })
 const config = Layer.succeed(
   Config.Service,
   Config.Service.of({
@@ -382,11 +384,11 @@ const runnerLayer = AppNodeBuilder.build(SessionRunnerLLM.node, [
   [InstructionBuiltIns.node, systemContext],
   [InstructionDiscovery.node, instructionContext],
   [Location.node, Location.boundNode({ directory: AbsolutePath.make("/project") })],
-  [SkillGuidance.node, skillGuidance],
-  [ReferenceGuidance.node, referenceGuidance],
+  [SkillInstructions.node, skillInstructions],
+  [ReferenceInstructions.node, referenceInstructions],
   [PermissionV2.node, permission],
   [Config.node, config],
-  [McpGuidance.node, mcpGuidance],
+  [McpInstructions.node, mcpInstructions],
   [ToolOutputStore.node, ToolOutputStore.nodeWithoutConfig],
   [PluginSupervisor.node, pluginSupervisor],
 ])
@@ -424,8 +426,8 @@ const it = testEffect(
       InstructionBuiltIns.node,
       InstructionDiscovery.node,
       InstructionEntry.node,
-      SkillGuidance.node,
-      ReferenceGuidance.node,
+      SkillInstructions.node,
+      ReferenceInstructions.node,
       Config.node,
       Snapshot.node,
       SessionRunnerLLM.node,
@@ -440,8 +442,8 @@ const it = testEffect(
       [InstructionBuiltIns.node, systemContext],
       [InstructionDiscovery.node, instructionContext],
       [Location.node, Location.boundNode({ directory: AbsolutePath.make("/project") })],
-      [SkillGuidance.node, skillGuidance],
-      [ReferenceGuidance.node, referenceGuidance],
+      [SkillInstructions.node, skillInstructions],
+      [ReferenceInstructions.node, referenceInstructions],
       [Snapshot.node, Snapshot.noopLayer],
       [SessionExecution.node, execution],
       [Config.node, config],
@@ -1424,7 +1426,7 @@ describe("SessionRunnerLLM", () => {
     }),
   )
 
-  it.effect("updates selected-agent skill guidance after an agent switch", () =>
+  it.effect("updates selected-agent skill instructions after an agent switch", () =>
     Effect.gen(function* () {
       const session = yield* setup
       const events = yield* EventV2.Service
@@ -4026,11 +4028,19 @@ describe("SessionRunnerLLM", () => {
         { attempt: 5, at: 30_000 },
       ])
       expect((yield* recordedEventTypes(sessionID)).filter((type) => type === "session.step.started.1")).toHaveLength(5)
-      expect((yield* session.context(sessionID)).filter((message) => message.type === "assistant")).toHaveLength(1)
+      const assistant = requireAssistant(yield* session.context(sessionID))
+      expect(yield* recordedStepSettlementEvents(sessionID, assistant.id)).toMatchObject([
+        { type: "session.step.started.1" },
+        { type: "session.step.started.1" },
+        { type: "session.step.started.1" },
+        { type: "session.step.started.1" },
+        { type: "session.step.started.1" },
+        { type: "session.step.failed.1" },
+      ])
     }),
   )
 
-  it.effect("counts retry attempts against the agent step allowance", () =>
+  it.effect("retries a physical attempt without consuming the logical agent step", () =>
     Effect.gen(function* () {
       const session = yield* setup
       const agents = yield* AgentV2.Service
@@ -4039,21 +4049,36 @@ describe("SessionRunnerLLM", () => {
           agent.steps = 2
         }),
       )
-      yield* admit(session, "Bound retries by steps")
+      yield* admit(session, "Retry without consuming a step")
       const failure = providerUnavailable()
       responseStream = Stream.fail(failure)
-      streamFailure = failure
+      responses = [reply.tool("call-after-retry", "echo", { text: "recovered" }), reply.stop()]
 
       const run = yield* session.resume(sessionID).pipe(Effect.forkChild)
       while (requests.length < 1) yield* Effect.yieldNow
       yield* TestClock.adjust("2 seconds")
-      expect(yield* Fiber.join(run).pipe(Effect.flip)).toBe(failure)
+      yield* Fiber.join(run)
 
-      expect(requests).toHaveLength(2)
+      expect(requests).toHaveLength(3)
+      expect(requests[0]?.toolChoice).toBeUndefined()
+      expect(requests[0]?.tools.map((tool) => tool.name)).toContain("echo")
+      expect(requests[1]?.toolChoice).toBeUndefined()
+      expect(requests[1]?.tools.map((tool) => tool.name)).toContain("echo")
+      expect(requests[1]?.messages.at(-1)).not.toMatchObject({
+        role: "assistant",
+        content: [{ type: "text", text: expect.stringContaining("MAXIMUM STEPS REACHED") }],
+      })
+      expect(requests[2]?.toolChoice).toMatchObject({ type: "none" })
+      expect(requests[2]?.tools).toEqual([])
+      expect(requests[2]?.messages.at(-1)).toMatchObject({
+        role: "assistant",
+        content: [{ type: "text", text: expect.stringContaining("MAXIMUM STEPS REACHED") }],
+      })
+      expect(executions).toEqual(["recovered"])
       const eventTypes = yield* recordedEventTypes(sessionID)
-      expect(eventTypes.filter((type) => type === "session.step.started.1")).toHaveLength(2)
+      expect(eventTypes.filter((type) => type === "session.step.started.1")).toHaveLength(3)
       expect(eventTypes.filter((type) => type === "session.retry.scheduled.1")).toHaveLength(1)
-      expect((yield* session.context(sessionID)).filter((message) => message.type === "assistant")).toHaveLength(1)
+      expect((yield* session.context(sessionID)).filter((message) => message.type === "assistant")).toHaveLength(2)
     }),
   )
 

packages/core/test/skill/instructions.test.ts

@@ -5,7 +5,7 @@ import { AgentV2 } from "@opencode-ai/core/agent"
 import { AppNodeBuilder } from "@opencode-ai/core/effect/app-node-builder"
 import { AbsolutePath } from "@opencode-ai/core/schema"
 import { SkillV2 } from "@opencode-ai/core/skill"
-import { SkillGuidance } from "@opencode-ai/core/skill/guidance"
+import { SkillInstructions } from "@opencode-ai/core/skill/instructions"
 import { it } from "../lib/effect"
 import { readInitial, readUpdate } from "../lib/instructions"
 
@@ -40,11 +40,11 @@ const manual = SkillV2.Info.make({
 })
 
 const layer = (list: () => SkillV2.Info[]) =>
-  AppNodeBuilder.build(SkillGuidance.node, [
+  AppNodeBuilder.build(SkillInstructions.node, [
     [SkillV2.node, Layer.mock(SkillV2.Service, { list: () => Effect.succeed(list()) })],
   ])
 
-describe("SkillGuidance", () => {
+describe("SkillInstructions", () => {
   it.effect("renders described agent skills and updates the complete available list", () => {
     const agent = AgentV2.Info.make({
       ...AgentV2.Info.empty(build),
@@ -52,8 +52,8 @@ describe("SkillGuidance", () => {
     })
     let skills = [hidden, denied, manual, effect]
     return Effect.gen(function* () {
-      const guidance = yield* SkillGuidance.Service
-      const initialized = yield* guidance.load({ id: agent.id, info: agent }).pipe(Effect.flatMap(readInitial))
+      const instructions = yield* SkillInstructions.Service
+      const initialized = yield* instructions.load({ id: agent.id, info: agent }).pipe(Effect.flatMap(readInitial))
 
       expect(initialized.text).toBe(
         [
@@ -72,7 +72,7 @@ describe("SkillGuidance", () => {
 
       skills = []
       expect(
-        yield* guidance
+        yield* instructions
           .load({ id: agent.id, info: agent })
           .pipe(Effect.flatMap((context) => readUpdate(context, initialized))),
       ).toMatchObject({ text: "Skill guidance is no longer available. Do not use any previously listed skill." })
@@ -90,11 +90,11 @@ describe("SkillGuidance", () => {
     })
     let skills = [effect]
     return Effect.gen(function* () {
-      const guidance = yield* SkillGuidance.Service
-      const initialized = yield* guidance.load({ id: agent.id, info: agent }).pipe(Effect.flatMap(readInitial))
+      const instructions = yield* SkillInstructions.Service
+      const initialized = yield* instructions.load({ id: agent.id, info: agent }).pipe(Effect.flatMap(readInitial))
 
       skills = [effect, debugging]
-      const added = yield* guidance
+      const added = yield* instructions
         .load({ id: agent.id, info: agent })
         .pipe(Effect.flatMap((context) => readUpdate(context, initialized)))
       expect(added.text).toBe(
@@ -109,7 +109,7 @@ describe("SkillGuidance", () => {
       )
 
       skills = [debugging]
-      const removed = yield* guidance
+      const removed = yield* instructions
         .load({ id: agent.id, info: agent })
         .pipe(Effect.flatMap((context) => readUpdate(context, added)))
       expect(removed.text).toBe("The following skill IDs are no longer available and must not be used: effect.")
@@ -120,12 +120,12 @@ describe("SkillGuidance", () => {
     const agent = AgentV2.Info.make(AgentV2.Info.empty(build))
     let skills = [effect]
     return Effect.gen(function* () {
-      const guidance = yield* SkillGuidance.Service
-      const initialized = yield* guidance.load({ id: agent.id, info: agent }).pipe(Effect.flatMap(readInitial))
+      const instructions = yield* SkillInstructions.Service
+      const initialized = yield* instructions.load({ id: agent.id, info: agent }).pipe(Effect.flatMap(readInitial))
 
       skills = [SkillV2.Info.make({ ...effect, description: "Build applications with Effect v4" })]
       expect(
-        yield* guidance
+        yield* instructions
           .load({ id: agent.id, info: agent })
           .pipe(Effect.flatMap((context) => readUpdate(context, initialized))),
       ).toMatchObject({
@@ -136,18 +136,18 @@ describe("SkillGuidance", () => {
     }).pipe(Effect.provide(layer(() => skills)))
   })
 
-  it.effect("omits guidance when the selected agent denies all skills", () => {
+  it.effect("omits instructions when the selected agent denies all skills", () => {
     const agent = AgentV2.Info.make({
       ...AgentV2.Info.empty(build),
       permissions: [{ action: "skill", resource: "*", effect: "deny" }],
     })
     return Effect.gen(function* () {
-      const guidance = yield* SkillGuidance.Service
-      expect((yield* guidance.load({ id: agent.id, info: agent }).pipe(Effect.flatMap(readInitial))).text).toBe("")
+      const instructions = yield* SkillInstructions.Service
+      expect((yield* instructions.load({ id: agent.id, info: agent }).pipe(Effect.flatMap(readInitial))).text).toBe("")
     }).pipe(Effect.provide(layer(() => [effect])))
   })
 
-  it.effect("omits guidance when a resource-specific denial follows the global denial", () => {
+  it.effect("omits instructions when a resource-specific denial follows the global denial", () => {
     const agent = AgentV2.Info.make({
       ...AgentV2.Info.empty(build),
       permissions: [
@@ -156,8 +156,8 @@ describe("SkillGuidance", () => {
       ],
     })
     return Effect.gen(function* () {
-      const guidance = yield* SkillGuidance.Service
-      expect((yield* guidance.load({ id: agent.id, info: agent }).pipe(Effect.flatMap(readInitial))).text).toBe("")
+      const instructions = yield* SkillInstructions.Service
+      expect((yield* instructions.load({ id: agent.id, info: agent }).pipe(Effect.flatMap(readInitial))).text).toBe("")
     }).pipe(Effect.provide(layer(() => [effect])))
   })
 
@@ -170,14 +170,14 @@ describe("SkillGuidance", () => {
       ],
     })
     return Effect.gen(function* () {
-      const guidance = yield* SkillGuidance.Service
-      expect((yield* guidance.load({ id: agent.id, info: agent }).pipe(Effect.flatMap(readInitial))).text).toContain(
-        "<name>Effect</name>",
-      )
+      const instructions = yield* SkillInstructions.Service
+      expect(
+        (yield* instructions.load({ id: agent.id, info: agent }).pipe(Effect.flatMap(readInitial))).text,
+      ).toContain("<name>Effect</name>")
     }).pipe(Effect.provide(layer(() => [effect])))
   })
 
-  it.effect("omits guidance when a specifically allowed skill is denied again", () => {
+  it.effect("omits instructions when a specifically allowed skill is denied again", () => {
     const agent = AgentV2.Info.make({
       ...AgentV2.Info.empty(build),
       permissions: [
@@ -187,8 +187,8 @@ describe("SkillGuidance", () => {
       ],
     })
     return Effect.gen(function* () {
-      const guidance = yield* SkillGuidance.Service
-      expect((yield* guidance.load({ id: agent.id, info: agent }).pipe(Effect.flatMap(readInitial))).text).toBe("")
+      const instructions = yield* SkillInstructions.Service
+      expect((yield* instructions.load({ id: agent.id, info: agent }).pipe(Effect.flatMap(readInitial))).text).toBe("")
     }).pipe(Effect.provide(layer(() => [effect])))
   })
 })

packages/core/test/state.test.ts

@@ -1,6 +1,7 @@
 import { describe, expect } from "bun:test"
 import { State } from "@opencode-ai/core/state"
 import { Deferred, Effect, Exit, Fiber, Layer, Scope } from "effect"
+import { TestClock } from "effect/testing"
 import { testEffect } from "./lib/effect"
 
 const it = testEffect(Layer.empty)
@@ -35,7 +36,7 @@ describe("State", () => {
     }),
   )
 
-  it.effect("runs effectful transforms during every reload", () =>
+  it.effect("runs transforms during every reload", () =>
     Effect.gen(function* () {
       let value = "first"
       const state = State.create({
@@ -43,15 +44,15 @@ describe("State", () => {
         draft: (draft) => ({ add: (item: string) => draft.values.push(item) }),
       })
 
-      yield* state.transform((editor) =>
-        Effect.sync(() => {
-          editor.add(value)
-        }),
-      )
+      yield* state.transform((editor) => {
+        editor.add(value)
+      })
       expect(state.get().values).toEqual(["first"])
 
       value = "second"
-      yield* state.reload()
+      const reload = yield* state.reload().pipe(Effect.forkChild({ startImmediately: true }))
+      yield* TestClock.adjust("500 millis")
+      yield* Fiber.join(reload)
       expect(state.get().values).toEqual(["second"])
     }),
   )
@@ -112,4 +113,30 @@ describe("State", () => {
       expect(finalized).toBe(2)
     }),
   )
+
+  it.effect("debounces reload bursts", () =>
+    Effect.gen(function* () {
+      let finalized = 0
+      const state = State.create({
+        initial: () => ({ values: [] as string[] }),
+        draft: (draft) => ({ add: (item: string) => draft.values.push(item) }),
+        finalize: () => Effect.sync(() => finalized++),
+      })
+      yield* state.transform((draft) => {
+        draft.add("value")
+      })
+      finalized = 0
+
+      const first = yield* state.reload().pipe(Effect.forkChild({ startImmediately: true }))
+      yield* TestClock.adjust("250 millis")
+      const second = yield* state.reload().pipe(Effect.forkChild({ startImmediately: true }))
+      yield* TestClock.adjust("499 millis")
+      expect(finalized).toBe(0)
+      yield* TestClock.adjust("1 millis")
+      yield* Fiber.join(first)
+      yield* Fiber.join(second)
+
+      expect(finalized).toBe(1)
+    }),
+  )
 })

packages/core/test/tool-question.test.ts

@@ -8,7 +8,9 @@ import { SessionV2 } from "@opencode-ai/core/session"
 import { ToolRegistry } from "@opencode-ai/core/tool/registry"
 import { QuestionTool } from "@opencode-ai/core/tool/question"
 import { ToolOutputStore } from "@opencode-ai/core/tool-output-store"
+import { Image } from "@opencode-ai/core/image"
 import { testEffect } from "./lib/effect"
+import { imagePassthrough } from "./lib/image"
 import { makeLocationNode } from "@opencode-ai/core/effect/app-node"
 import { toolIdentity, executeTool, registerToolPlugin, settleTool, toolDefinitions } from "./lib/tool"
 
@@ -84,6 +86,7 @@ const it = testEffect(
     [PermissionV2.node, permission],
     [Form.node, form],
     [ToolOutputStore.node, ToolOutputStore.nodeWithoutConfig],
+    [Image.node, imagePassthrough],
   ]),
 )
 

packages/core/test/tool-read.test.ts

@@ -291,6 +291,9 @@ describe("ReadTool", () => {
         name: "pixel.png",
         mime: "image/png",
         encoding: "base64",
+        // Image base64 is carried by the content file item only; structured is slimmed
+        // so the original bytes are never persisted twice.
+        content: "",
       })
       expect(settled.output?.content).toMatchObject([
         { type: "text", text: "Image read successfully" },
@@ -364,7 +367,7 @@ describe("ReadTool", () => {
     }),
   )
 
-  it.effect("rejects invalid image data returned by the filesystem", () =>
+  it.effect("drops undecodable image data at settlement", () =>
     Effect.gen(function* () {
       readResult = {
         uri: "file:///truncated.png",
@@ -381,11 +384,17 @@ describe("ReadTool", () => {
           ...toolIdentity,
           call: { type: "tool-call", id: "call-truncated-image", name: "read", input: { path: "truncated.png" } },
         }),
-      ).toEqual({ type: "error", value: "Image could not be decoded: truncated.png" })
+      ).toEqual({
+        type: "content",
+        value: [
+          { type: "text", text: "Image read successfully" },
+          { type: "text", text: "[1 image omitted: could not be decoded.]" },
+        ],
+      })
     }),
   )
 
-  it.effect("rejects oversized images when resizing is disabled", () =>
+  it.effect("drops oversized images at settlement when resizing is disabled", () =>
     Effect.gen(function* () {
       const photon = yield* Effect.promise(() => import("@silvia-odwyer/photon-node"))
       const source = new photon.PhotonImage(new Uint8Array(Array.from({ length: 16 * 4 }, () => 255)), 16, 1)
@@ -409,14 +418,20 @@ describe("ReadTool", () => {
         }),
       ]
       const registry = yield* ToolRegistry.Service
-      const result = yield* executeTool(registry, {
-        sessionID,
-        ...toolIdentity,
-        call: { type: "tool-call", id: "call-wide-image", name: "read", input: { path: "wide.png" } },
-      })
 
-      expect(result.type).toBe("error")
-      if (result.type === "error") expect(result.value).toContain("exceeding configured limits 4x2000")
+      expect(
+        yield* executeTool(registry, {
+          sessionID,
+          ...toolIdentity,
+          call: { type: "tool-call", id: "call-wide-image", name: "read", input: { path: "wide.png" } },
+        }),
+      ).toEqual({
+        type: "content",
+        value: [
+          { type: "text", text: "Image read successfully" },
+          { type: "text", text: "[1 image omitted: could not be resized below the image size limit.]" },
+        ],
+      })
     }),
   )
 
@@ -460,7 +475,7 @@ describe("ReadTool", () => {
     }),
   )
 
-  it.effect("enforces max base64 bytes after resize attempts", () =>
+  it.effect("drops images that cannot fit max base64 bytes after resize attempts", () =>
     Effect.gen(function* () {
       const png = "iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR42mNk+A8AAQUBAScY42YAAAAASUVORK5CYII="
       readResult = {
@@ -481,14 +496,20 @@ describe("ReadTool", () => {
         }),
       ]
       const registry = yield* ToolRegistry.Service
-      const result = yield* executeTool(registry, {
-        sessionID,
-        ...toolIdentity,
-        call: { type: "tool-call", id: "call-max-bytes", name: "read", input: { path: "pixel.png" } },
-      })
 
-      expect(result.type).toBe("error")
-      if (result.type === "error") expect(result.value).toContain("/1 bytes")
+      expect(
+        yield* executeTool(registry, {
+          sessionID,
+          ...toolIdentity,
+          call: { type: "tool-call", id: "call-max-bytes", name: "read", input: { path: "pixel.png" } },
+        }),
+      ).toEqual({
+        type: "content",
+        value: [
+          { type: "text", text: "Image read successfully" },
+          { type: "text", text: "[1 image omitted: could not be resized below the image size limit.]" },
+        ],
+      })
     }),
   )
 

packages/core/test/tool-shell.test.ts

@@ -166,10 +166,10 @@ const overflowCommand = (bytes: number) =>
   isWindows
     ? `[Console]::Out.Write(('x' * ${bytes})); Start-Sleep -Milliseconds 100`
     : `head -c ${bytes} /dev/zero | tr '\\0' 'x'`
-const progressOverflowCommand = (bytes: number) =>
+const progressLinesCommand = (lines: number) =>
   isWindows
-    ? `[Console]::Out.Write(('x' * ${bytes})); Start-Sleep -Milliseconds 1500`
-    : `head -c ${bytes} /dev/zero | tr '\\0' 'x'; sleep 1.5`
+    ? `1..${lines} | ForEach-Object { [Console]::Out.WriteLine(('line {0:d2}' -f $_)) }; Start-Sleep -Milliseconds 1500`
+    : `i=1; while [ $i -le ${lines} ]; do printf 'line %02d\\n' "$i"; i=$((i+1)); done; sleep 1.5`
 
 const withSession = <A, E, R>(directory: string, body: (registry: ToolRegistry.Interface) => Effect.Effect<A, E, R>) =>
   Effect.gen(function* () {
@@ -417,17 +417,17 @@ describe("ShellTool", () => {
     ),
   )
 
-  it.live("reports bounded output progress for a running command", () =>
+  it.live("reports the latest output lines for a running command", () =>
     Effect.acquireUseRelease(
       Effect.promise(() => tmpdir()),
       (tmp) => {
         reset()
-        const bytes = ShellTool.MAX_CAPTURE_BYTES + 1024
+        const lines = ShellTool.PROGRESS_LINES + 10
         return withSession(tmp.path, (registry) =>
           Effect.gen(function* () {
             const progress: ToolRegistry.Progress[] = []
             yield* settleTool(registry, {
-              ...call({ command: progressOverflowCommand(bytes) }, "call-progress"),
+              ...call({ command: progressLinesCommand(lines) }, "call-progress"),
               progress: (update) => Effect.sync(() => progress.push(update)),
             })
 
@@ -436,9 +436,12 @@ describe("ShellTool", () => {
             const content = progress[0]?.content[0]
             expect(content?.type).toBe("text")
             if (content?.type !== "text") return
-            expect(content.text.indexOf("\n\n[output truncated; full output saved to:")).toBe(
-              ShellTool.MAX_CAPTURE_BYTES,
+            expect(content.text).toStartWith(
+              `[output truncated; showing last ${ShellTool.PROGRESS_LINES} lines. Full output saved to:`,
             )
+            expect(content.text).not.toContain("line 10\n")
+            expect(content.text).toContain("line 11\n")
+            expect(content.text).toContain(`line ${lines}\n`)
           }),
         )
       },

packages/core/test/tool-skill.test.ts

@@ -12,7 +12,9 @@ import { SkillTool } from "@opencode-ai/core/tool/skill"
 import { ToolRegistry } from "@opencode-ai/core/tool/registry"
 import { ToolOutputStore } from "@opencode-ai/core/tool-output-store"
 import { tmpdir } from "./fixture/tmpdir"
+import { Image } from "@opencode-ai/core/image"
 import { it } from "./lib/effect"
+import { imagePassthrough } from "./lib/image"
 import { makeLocationNode } from "@opencode-ai/core/effect/app-node"
 import { FSUtil } from "@opencode-ai/core/fs-util"
 import { toolIdentity, executeTool, registerToolPlugin, settleTool, toolDefinitions } from "./lib/tool"
@@ -90,6 +92,7 @@ describe("SkillTool", () => {
               [PermissionV2.node, permission],
               [SkillV2.node, skills],
               [ToolOutputStore.node, ToolOutputStore.nodeWithoutConfig],
+              [Image.node, imagePassthrough],
             ],
           )
 

packages/core/test/tool-subagent.test.ts

@@ -1,5 +1,6 @@
 import { describe, expect } from "bun:test"
 import { DateTime, Effect, Fiber, Layer, Schema, Stream } from "effect"
+import path from "path"
 import { Money } from "@opencode-ai/schema/money"
 import { AppNodeBuilder } from "@opencode-ai/core/effect/app-node-builder"
 import { LayerNode } from "@opencode-ai/core/effect/layer-node"
@@ -164,6 +165,77 @@ describe("SubagentTool", () => {
     ),
   )
 
+  it.live("prevents subagents from launching subagents by default", () =>
+    Effect.acquireRelease(
+      Effect.promise(() => tmpdir()),
+      (dir) => Effect.promise(() => dir[Symbol.asyncDispose]()),
+    ).pipe(
+      Effect.flatMap((dir) =>
+        Effect.gen(function* () {
+          const location = Location.Ref.make({ directory: AbsolutePath.make(dir.path) })
+          const sessions = yield* SessionV2.Service
+          const root = yield* sessions.create({ location })
+          const parent = yield* sessions.create({ parentID: root.id, title: "parent" })
+          yield* withSubagent(parent.location)
+          const locations = yield* LocationServiceMap.Service
+          const registry = yield* ToolRegistry.Service.pipe(Effect.provide(locations.get(parent.location)))
+          yield* waitForTool(registry, SubagentTool.name)
+
+          expect(
+            yield* executeTool(registry, {
+              sessionID: parent.id,
+              ...toolIdentity,
+              call: {
+                type: "tool-call",
+                id: "call-nested-subagent",
+                name: SubagentTool.name,
+                input: { agent: "reviewer", description: "nested", prompt: "should fail" },
+              },
+            }),
+          ).toEqual({ type: "error", value: expect.stringContaining("Subagent depth limit reached (1)") })
+          expect((yield* sessions.list({ parentID: parent.id })).data).toHaveLength(0)
+        }),
+      ),
+    ),
+  )
+
+  it.live("allows nested subagents up to the configured depth", () =>
+    Effect.acquireRelease(
+      Effect.promise(() => tmpdir()),
+      (dir) => Effect.promise(() => dir[Symbol.asyncDispose]()),
+    ).pipe(
+      Effect.flatMap((dir) =>
+        Effect.gen(function* () {
+          yield* Effect.promise(() =>
+            Bun.write(path.join(dir.path, "opencode.json"), JSON.stringify({ experimental: { subagent_depth: 2 } })),
+          )
+          const location = Location.Ref.make({ directory: AbsolutePath.make(dir.path) })
+          const sessions = yield* SessionV2.Service
+          const root = yield* sessions.create({ location })
+          const parent = yield* sessions.create({ parentID: root.id, title: "parent", model: parentModel })
+          yield* withSubagent(parent.location)
+          const locations = yield* LocationServiceMap.Service
+          const registry = yield* ToolRegistry.Service.pipe(Effect.provide(locations.get(parent.location)))
+          yield* waitForTool(registry, SubagentTool.name)
+
+          const settled = yield* settleTool(registry, {
+            sessionID: parent.id,
+            ...toolIdentity,
+            call: {
+              type: "tool-call",
+              id: "call-configured-nested-subagent",
+              name: SubagentTool.name,
+              input: { agent: "reviewer", description: "nested", prompt: "should run" },
+            },
+          })
+
+          expect(settled.output?.structured).toMatchObject({ status: "completed", output: childText })
+          expect((yield* sessions.get(outputSessionID(settled.output?.structured))).parentID).toBe(parent.id)
+        }),
+      ),
+    ),
+  )
+
   it.live("runs a foreground child session and returns the final assistant text", () =>
     Effect.acquireRelease(
       Effect.promise(() => tmpdir()),

packages/core/test/tool-webfetch.test.ts

@@ -11,7 +11,9 @@ import { ToolRegistry } from "@opencode-ai/core/tool/registry"
 import { WebFetchTool } from "@opencode-ai/core/tool/webfetch"
 import { ToolOutputStore } from "@opencode-ai/core/tool-output-store"
 import { makeLocationNode } from "@opencode-ai/core/effect/app-node"
+import { Image } from "@opencode-ai/core/image"
 import { testEffect } from "./lib/effect"
+import { imagePassthrough } from "./lib/image"
 import { toolIdentity, executeTool, registerToolPlugin, settleTool, toolDefinitions } from "./lib/tool"
 
 const webFetchToolNode = makeLocationNode({
@@ -50,6 +52,7 @@ const toolLayer = (replacements: LayerNode.Replacements = []) =>
   AppNodeBuilder.build(LayerNode.group([ToolRegistry.node, ToolRegistry.toolsNode, webFetchToolNode]), [
     [PermissionV2.node, permission],
     [ToolOutputStore.node, ToolOutputStore.nodeWithoutConfig],
+    [Image.node, imagePassthrough],
     ...replacements,
   ])
 const it = testEffect(toolLayer([[LayerNodePlatform.httpClient, http]]))

packages/core/test/tool-websearch.test.ts

@@ -10,7 +10,9 @@ import { ToolRegistry } from "@opencode-ai/core/tool/registry"
 import { WebSearchTool } from "@opencode-ai/core/tool/websearch"
 import { ToolOutputStore } from "@opencode-ai/core/tool-output-store"
 import { makeLocationNode } from "@opencode-ai/core/effect/app-node"
+import { Image } from "@opencode-ai/core/image"
 import { testEffect } from "./lib/effect"
+import { imagePassthrough } from "./lib/image"
 import { toolIdentity, executeTool, registerToolPlugin, settleTool, toolDefinitions } from "./lib/tool"
 
 const webSearchToolNode = makeLocationNode({
@@ -138,6 +140,7 @@ const it = testEffect(
       [LayerNodePlatform.httpClient, http],
       [WebSearchTool.configNode, websearchConfig],
       [ToolOutputStore.node, ToolOutputStore.nodeWithoutConfig],
+      [Image.node, imagePassthrough],
     ],
   ),
 )

packages/core/test/wellknown.test.ts

@@ -0,0 +1,119 @@
+import { expect } from "bun:test"
+import { Effect, Fiber, Stream } from "effect"
+import { FetchHttpClient } from "effect/unstable/http"
+import { KV } from "@opencode-ai/core/kv"
+import { LayerNode } from "@opencode-ai/core/effect/layer-node"
+import { EventV2 } from "@opencode-ai/core/event"
+import { WellKnown } from "@opencode-ai/core/wellknown"
+import { testEffect } from "./lib/effect"
+
+const it = testEffect(FetchHttpClient.layer)
+const serviceIt = testEffect(LayerNode.compile(LayerNode.group([WellKnown.node, KV.node, EventV2.node])))
+
+it.live("loads embedded and remote configuration", () =>
+  Effect.acquireUseRelease(
+    Effect.sync(() =>
+      Bun.serve({
+        port: 0,
+        fetch(request) {
+          const url = new URL(request.url)
+          if (url.pathname === "/.well-known/opencode") {
+            return Response.json({
+              auth: { command: ["login"], env: "TOKEN" },
+              config: { model: "embedded/model" },
+              remote_config: {
+                url: `${url.origin}/config/{env:TOKEN}`,
+                headers: { authorization: "Bearer {env:TOKEN}" },
+              },
+            })
+          }
+          if (url.pathname === "/config/secret" && request.headers.get("authorization") === "Bearer secret") {
+            return Response.json({ config: { model: "remote/model" } })
+          }
+          return new Response("Not found", { status: 404 })
+        },
+      }),
+    ),
+    (server) =>
+      Effect.gen(function* () {
+        const origin = server.url.origin
+        expect(yield* WellKnown.inspect(`${origin}/`)).toEqual({
+          auth: { command: ["login"], env: "TOKEN" },
+          config: { model: "embedded/model" },
+          remote_config: {
+            url: `${origin}/config/{env:TOKEN}`,
+            headers: { authorization: "Bearer {env:TOKEN}" },
+          },
+        })
+        expect(yield* WellKnown.resolve({ origin, variables: { TOKEN: "secret" } })).toEqual([
+          { model: "embedded/model" },
+          { model: "remote/model" },
+        ])
+      }),
+    (server) => Effect.promise(() => server.stop(true)),
+  ),
+)
+
+serviceIt.live("persists sources in one KV value", () =>
+  Effect.acquireUseRelease(
+    Effect.sync(() =>
+      Bun.serve({
+        port: 0,
+        fetch: () => Response.json({ auth: { command: ["login"], env: "TOKEN" } }),
+      }),
+    ),
+    (server) =>
+      Effect.gen(function* () {
+        const wellknown = yield* WellKnown.Service
+        const kv = yield* KV.Service
+        const events = yield* EventV2.Service
+        const changed = yield* events
+          .subscribe(WellKnown.Event.Updated)
+          .pipe(Stream.take(1), Stream.runCollect, Effect.forkScoped)
+        const entry = yield* wellknown.add(`${server.url.origin}/`)
+
+        expect(entry.origin).toBe(server.url.origin)
+        expect(yield* kv.get("wellknown:sources")).toEqual([server.url.origin])
+        expect(yield* wellknown.entries()).toEqual([entry])
+        expect(yield* Fiber.join(changed)).toHaveLength(1)
+
+        yield* wellknown.remove(server.url.origin)
+        expect(yield* kv.get("wellknown:sources")).toEqual([])
+        expect(yield* wellknown.entries()).toEqual([])
+      }),
+    (server) => Effect.promise(() => server.stop(true)),
+  ),
+)
+
+serviceIt.live("refreshes changed manifests", () =>
+  Effect.acquireUseRelease(
+    Effect.sync(() => {
+      let command = "first"
+      return {
+        server: Bun.serve({
+          port: 0,
+          fetch: () => Response.json({ auth: { command: [command], env: "TOKEN" } }),
+        }),
+        update: () => {
+          command = "second"
+        },
+      }
+    }),
+    ({ server, update }) =>
+      Effect.gen(function* () {
+        const wellknown = yield* WellKnown.Service
+        const events = yield* EventV2.Service
+        yield* wellknown.add(server.url.origin)
+        expect(yield* wellknown.refresh()).toBe(false)
+
+        const changed = yield* events
+          .subscribe(WellKnown.Event.Updated)
+          .pipe(Stream.take(1), Stream.runCollect, Effect.forkScoped)
+        update()
+        expect(yield* wellknown.refresh()).toBe(true)
+        expect(yield* Fiber.join(changed)).toHaveLength(1)
+        expect(wellknown.snapshot()[0]?.manifest.auth?.command).toEqual(["second"])
+      }),
+    ({ server }) => Effect.promise(() => server.stop(true)),
+  ),
+)

packages/docs/build/plugins.mdx

@@ -177,7 +177,7 @@ and plugin options.
 | `ctx.integration`      | `list`, `get`, `connect`, `attempt`, `transform`, `reload`, and connection lookup/resolution |
 | `ctx.plugin`           | `list` currently active plugin IDs                                                           |
 | `ctx.reference`        | `list`, `transform`, `reload`                                                                |
-| `ctx.session`          | `create`, `get`, `prompt`, `command`, `interrupt`, and `hook`                                |
+| `ctx.session`          | `create`, `get`, `prompt`, `command`, `synthetic`, `interrupt`, and `hook`                   |
 | `ctx.skill`            | `list`, `transform`, `reload`                                                                |
 | `ctx.tool`             | `transform` and `hook`                                                                       |
 | `ctx.aisdk`            | `hook`                                                                                       |

packages/opencode/test/session/message-v2.test.ts

@@ -1528,7 +1528,7 @@ describe("session.message-v2.fromError", () => {
 
   test("classifies ZlibError from fetch as retryable APIError", () => {
     const zlibError = new Error(
-      'ZlibError fetching "https://opencode.cloudflare.dev/anthropic/messages". For more information, pass `verbose: true` in the second argument to fetch()',
+      'ZlibError fetching "https://example.com/anthropic/messages". For more information, pass `verbose: true` in the second argument to fetch()',
     )
     ;(zlibError as any).code = "ZlibError"
     ;(zlibError as any).errno = 0
@@ -1543,7 +1543,7 @@ describe("session.message-v2.fromError", () => {
 
   test("classifies ZlibError as AbortedError when abort context is provided", () => {
     const zlibError = new Error(
-      'ZlibError fetching "https://opencode.cloudflare.dev/anthropic/messages". For more information, pass `verbose: true` in the second argument to fetch()',
+      'ZlibError fetching "https://example.com/anthropic/messages". For more information, pass `verbose: true` in the second argument to fetch()',
     )
     ;(zlibError as any).code = "ZlibError"
     ;(zlibError as any).errno = 0