fix(agent): stabilize sandboxed streaming workflows

2026-08-26 12:26:29 +08:00
parent 80cfc1f2ab
commit 4ec010455b
12 changed files with 916 additions and 34 deletions
@@ -0,0 +1,186 @@
+diff --git a/package.json b/package.json
+index 15725c8..4378766 100644
+--- a/package.json
+++ b/package.json
+@@ -4,7 +4,7 @@
+   "description": "AI-powered development tool",
+   "private": true,
+   "type": "module",
+-  "packageManager": "bun@1.3.14",
+  "packageManager": "bun@1.3.13",
+   "scripts": {
+     "dev": "bun run --cwd packages/opencode --conditions=browser src/index.ts",
+     "dev:desktop": "bun --cwd packages/desktop dev",
+diff --git a/packages/llm/src/protocols/openai-responses.ts b/packages/llm/src/protocols/openai-responses.ts
+index 4936d31..53138af 100644
+--- a/packages/llm/src/protocols/openai-responses.ts
+++ b/packages/llm/src/protocols/openai-responses.ts
+@@ -177,6 +177,7 @@ type OpenAIResponsesUsage = Schema.Schema.Type<typeof OpenAIResponsesUsage>
+ const OpenAIResponsesStreamItem = Schema.Struct({
+   type: Schema.String,
+   id: Schema.optional(Schema.String),
+  phase: optionalNull(Schema.String),
+   call_id: Schema.optional(Schema.String),
+   name: Schema.optional(Schema.String),
+   arguments: Schema.optional(Schema.String),
+@@ -238,6 +239,7 @@ interface ParserState {
+   readonly hasFunctionCall: boolean
+   readonly lifecycle: Lifecycle.State
+   readonly reasoningItems: Readonly<Record<string, ReasoningStreamItem>>
+  readonly textMetadata: Readonly<Record<string, ProviderMetadata>>
+   readonly store: boolean | undefined
+ }
+ 
+@@ -614,9 +616,19 @@ const TERMINAL_TYPES = new Set(["response.completed", "response.incomplete", "re
+ 
+ const onOutputTextDelta = (state: ParserState, event: OpenAIResponsesEvent): StepResult => {
+   if (!event.delta) return [state, NO_EVENTS]
+  const itemID = event.item_id ?? "text-0"
+   const events: LLMEvent[] = []
+   return [
+-    { ...state, lifecycle: Lifecycle.textDelta(state.lifecycle, events, event.item_id ?? "text-0", event.delta) },
+    {
+      ...state,
+      lifecycle: Lifecycle.textDelta(
+        state.lifecycle,
+        events,
+        itemID,
+        event.delta,
+        state.textMetadata[itemID],
+      ),
+    },
+     events,
+   ]
+ }
+@@ -655,6 +667,22 @@ const reasoningMetadata = (item: OpenAIResponsesStreamItem & { id: string }) =>
+ // best-effort, not guaranteed.
+ const onOutputItemAdded = (state: ParserState, event: OpenAIResponsesEvent): StepResult => {
+   const item = event.item
+  if (item?.type === "message" && item.id) {
+    const phase = item.phase === "commentary" || item.phase === "final_answer" ? item.phase : undefined
+    return [
+      {
+        ...state,
+        textMetadata: {
+          ...state.textMetadata,
+          [item.id]: openaiMetadata({
+            itemId: item.id,
+            ...(phase ? { phase } : {}),
+          }),
+        },
+      },
+      NO_EVENTS,
+    ]
+  }
+   if (item && isReasoningItem(item)) {
+     const events: LLMEvent[] = []
+     return [
+@@ -812,6 +840,20 @@ const onOutputItemDone = Effect.fn("OpenAIResponses.onOutputItemDone")(function*
+   const item = event.item
+   if (!item) return [state, NO_EVENTS] satisfies StepResult
+ 
+  if (item.type === "message" && item.id) {
+    const events: LLMEvent[] = []
+    const lifecycle = Lifecycle.textEnd(state.lifecycle, events, item.id, state.textMetadata[item.id])
+    const { [item.id]: _removed, ...textMetadata } = state.textMetadata
+    return [
+      {
+        ...state,
+        lifecycle,
+        textMetadata,
+      },
+      events,
+    ] satisfies StepResult
+  }
+
+   if (item.type === "function_call") {
+     if (!item.id || !item.call_id || !item.name) return [state, NO_EVENTS] satisfies StepResult
+     const tools = state.tools[item.id]
+@@ -969,6 +1011,7 @@ export const protocol = Protocol.make({
+       tools: ToolStream.empty<string>(),
+       lifecycle: Lifecycle.initial(),
+       reasoningItems: {},
+      textMetadata: {},
+       store: OpenAIOptions.store(request),
+     }),
+     step,
+diff --git a/packages/llm/src/protocols/utils/lifecycle.ts b/packages/llm/src/protocols/utils/lifecycle.ts
+index eb6c95d..64248be 100644
+--- a/packages/llm/src/protocols/utils/lifecycle.ts
+++ b/packages/llm/src/protocols/utils/lifecycle.ts
+@@ -14,13 +14,22 @@ export const stepStart = (state: State, events: LLMEvent[]): State => {
+   return { ...state, stepStarted: true }
+ }
+ 
+-export const textDelta = (state: State, events: LLMEvent[], id: string, text: string): State => {
+export const textDelta = (
+  state: State,
+  events: LLMEvent[],
+  id: string,
+  text: string,
+  providerMetadata?: ProviderMetadata,
+): State => {
+   const stepped = stepStart(state, events)
+   if (stepped.text.has(id)) {
+-    events.push(LLMEvent.textDelta({ id, text }))
+    events.push(LLMEvent.textDelta({ id, text, providerMetadata }))
+     return stepped
+   }
+-  events.push(LLMEvent.textStart({ id }), LLMEvent.textDelta({ id, text }))
+  events.push(
+    LLMEvent.textStart({ id, providerMetadata }),
+    LLMEvent.textDelta({ id, text, providerMetadata }),
+  )
+   return { ...stepped, text: new Set([...stepped.text, id]) }
+ }
+ 
+diff --git a/packages/llm/test/provider/openai-responses.test.ts b/packages/llm/test/provider/openai-responses.test.ts
+index cd8bad5..89ae576 100644
+--- a/packages/llm/test/provider/openai-responses.test.ts
+++ b/packages/llm/test/provider/openai-responses.test.ts
+@@ -754,6 +754,45 @@ describe("OpenAI Responses route", () => {
+     }),
+   )
+ 
+  it.effect("preserves assistant message phase in text provider metadata", () =>
+    Effect.gen(function* () {
+      const message = {
+        type: "message",
+        id: "msg_final",
+        phase: "final_answer",
+      }
+      const body = sseEvents(
+        { type: "response.output_item.added", item: message },
+        { type: "response.output_text.delta", item_id: "msg_final", delta: "Final" },
+        { type: "response.output_item.done", item: message },
+        { type: "response.completed", response: { id: "resp_1" } },
+      )
+
+      const response = yield* LLMClient.generate(request).pipe(Effect.provide(fixedResponse(body)))
+      const metadata = {
+        openai: {
+          itemId: "msg_final",
+          phase: "final_answer",
+        },
+      }
+
+      expect(response.text).toBe("Final")
+      expect(response.events).toMatchObject([
+        { type: "step-start", index: 0 },
+        { type: "text-start", id: "msg_final", providerMetadata: metadata },
+        {
+          type: "text-delta",
+          id: "msg_final",
+          text: "Final",
+          providerMetadata: metadata,
+        },
+        { type: "text-end", id: "msg_final", providerMetadata: metadata },
+        { type: "step-finish", index: 0, reason: "stop" },
+        { type: "finish", reason: "stop" },
+      ])
+    }),
+  )
+
+   it.effect("parses reasoning summary stream fixtures", () =>
+     Effect.gen(function* () {
+       const body = sseEvents(