Merge remote-tracking branch 'origin/v2' into responses-item-replay

fix(ai): harden Responses item replay
fix(ai): preserve Responses item IDs
2026-08-09 10:59:49 -04:00 · 2026-08-06 20:41:22 -05:00 · 2026-08-06 17:45:37 -05:00 · 2026-08-06 17:41:39 -05:00
18 changed files with 450 additions and 361 deletions
@@ -70,10 +70,14 @@ const OpenResponsesReasoningItem = Schema.Struct({
  encrypted_content: optionalNull(Schema.String),
 })

-const OpenResponsesItemReference = Schema.Struct({
-  type: Schema.tag("item_reference"),
-  id: Schema.String,
-})
+export const ProviderItem = Schema.StructWithRest(
+  Schema.Struct({
+    type: Schema.String,
+    id: Schema.String,
+  }),
+  [Schema.Record(Schema.String, Schema.Unknown)],
+)
+type OpenResponsesProviderItem = Schema.Schema.Type<typeof ProviderItem>

 // `function_call_output.output` accepts either a plain string or an ordered
 // array of content items so tools can return images and files in addition to text.
@@ -91,29 +95,42 @@ const OpenResponsesFunctionCallOutput = Schema.Union([

 export const InputItem = Schema.Union([
  Schema.Struct({ role: Schema.tag("system"), content: Schema.String }),
-  Schema.Struct({ role: Schema.tag("user"), content: Schema.Array(OpenResponsesInputContent) }),
  Schema.Struct({
+    type: Schema.optionalKey(Schema.tag("message")),
+    id: Schema.optionalKey(Schema.String),
+    role: Schema.tag("user"),
+    content: Schema.Array(OpenResponsesInputContent),
+    status: Schema.optionalKey(Schema.Literals(["in_progress", "completed", "incomplete"])),
+  }),
+  Schema.Struct({
+    type: Schema.optionalKey(Schema.tag("message")),
+    id: Schema.optionalKey(Schema.String),
    role: Schema.tag("assistant"),
    content: Schema.Array(OpenResponsesOutputText),
+    status: Schema.optionalKey(Schema.Literals(["in_progress", "completed", "incomplete"])),
    phase: Schema.optionalKey(MessagePhase),
  }),
  OpenResponsesReasoningItem,
-  OpenResponsesItemReference,
  Schema.Struct({
    type: Schema.tag("function_call"),
+    id: Schema.optionalKey(Schema.String),
    call_id: Schema.String,
    name: Schema.String,
    arguments: Schema.String,
+    status: Schema.optionalKey(Schema.Literals(["in_progress", "completed", "incomplete"])),
  }),
  Schema.Struct({
    type: Schema.tag("function_call_output"),
+    id: Schema.optionalKey(Schema.String),
    call_id: Schema.String,
    output: OpenResponsesFunctionCallOutput,
+    status: Schema.optionalKey(Schema.Literals(["in_progress", "completed", "incomplete"])),
  }),
 ])
 type OpenResponsesInputItem = Schema.Schema.Type<typeof InputItem>
 type LoweredInputItem =
  | OpenResponsesInputItem
+  | OpenResponsesProviderItem
  | {
      readonly role: "assistant"
      readonly content: ReadonlyArray<{ readonly type: "output_text"; readonly text: string }>
@@ -128,7 +145,7 @@ type OpenResponsesReasoningInput = {
  summary: Array<{ type: "summary_text"; text: string }>
  encrypted_content?: string | null
 }
-type OpenResponsesReasoningReplay = Omit<OpenResponsesReasoningInput, "id">
+type OpenResponsesReasoningReplay = OpenResponsesReasoningInput

 export const Tool = Schema.Struct({
  type: Schema.tag("function"),
@@ -269,10 +286,9 @@ export interface ParserState {
  readonly messagePhase: (value: unknown) => MessagePhase | null | undefined
  readonly messagePhases: Readonly<Record<string, MessagePhase | null>>
  readonly reasoningItems: Readonly<Record<string, ReasoningStreamItem>>
-  readonly store: boolean | undefined
 }

-type ReasoningSummaryStatus = "active" | "can-conclude" | "concluded"
+type ReasoningSummaryStatus = "active" | "can-conclude"

 interface ReasoningStreamItem {
  readonly encryptedContent: string | null | undefined
@@ -310,34 +326,82 @@ export const lowerToolChoice = (protocolName: string, toolChoice: NonNullable<LL
    tool: (toolName) => ({ type: "function" as const, name: toolName }),
  })

-const lowerToolCall = (part: ToolCallPart): OpenResponsesInputItem => ({
-  type: "function_call",
-  call_id: part.id,
-  name: part.name,
-  arguments: ProviderShared.encodeJson(part.input),
-})
+const responseItemID = (prefix: string, id: string) => {
+  const value = id.startsWith("call_") ? id.slice(5) : id
+  const sanitized = value.replace(/[^a-zA-Z0-9_-]/g, "_").replace(/^_+|_+$/g, "") || "item"
+  const direct = `${prefix}_${sanitized}`
+  if (value === sanitized && direct.length <= 64) return direct
+  const hash = Array.from(id).reduce(
+    (hash, character) => BigInt.asUintN(64, (hash ^ BigInt(character.codePointAt(0) ?? 0)) * 1099511628211n),
+    14695981039346656037n,
+  )
+  const suffix = hash.toString(36)
+  return `${prefix}_${sanitized.slice(0, 62 - prefix.length - suffix.length)}_${suffix}`
+}
+
+const validResponseItemID = (value: unknown): value is string =>
+  typeof value === "string" && value.length <= 64 && /^[a-zA-Z0-9]+_.+$/.test(value)
+
+const responseItemMetadata = (part: { readonly providerMetadata?: ProviderMetadata }, providerMetadataKey: string) => {
+  const metadata = part.providerMetadata?.[providerMetadataKey]
+  return ProviderShared.isRecord(metadata) ? metadata : undefined
+}
+
+const lowerToolCall = (part: ToolCallPart, providerMetadataKey: string): OpenResponsesInputItem => {
+  const metadata = responseItemMetadata(part, providerMetadataKey)
+  return {
+    type: "function_call",
+    id: validResponseItemID(metadata?.itemId) ? metadata.itemId : responseItemID("fc", part.id),
+    call_id: part.id,
+    name: part.name,
+    arguments: ProviderShared.encodeJson(part.input),
+    status: "completed",
+  }
+}

 const lowerReasoning = (part: ReasoningPart, providerMetadataKey: string): OpenResponsesReasoningInput | undefined => {
  const metadata = part.providerMetadata?.[providerMetadataKey]
-  if (!ProviderShared.isRecord(metadata) || typeof metadata.itemId !== "string" || metadata.itemId.length === 0)
-    return undefined
+  if (!ProviderShared.isRecord(metadata) || typeof metadata.itemId !== "string") return undefined
  const encryptedContent =
    typeof metadata.reasoningEncryptedContent === "string" || metadata.reasoningEncryptedContent === null
      ? metadata.reasoningEncryptedContent
      : undefined
  return {
    type: "reasoning",
-    id: metadata.itemId,
+    id: validResponseItemID(metadata.itemId) ? metadata.itemId : responseItemID("rs", metadata.itemId),
    summary: part.text.length > 0 ? [{ type: "summary_text", text: part.text }] : [],
    encrypted_content: encryptedContent,
  }
 }

-const hostedToolItemID = (part: ToolResultPart, providerMetadataKey: string) => {
-  const metadata = part.providerMetadata?.[providerMetadataKey]
-  return ProviderShared.isRecord(metadata) && typeof metadata.itemId === "string" && metadata.itemId.length > 0
-    ? metadata.itemId
-    : undefined
+const hostedToolItem = (part: ToolResultPart, providerMetadataKey: string): OpenResponsesProviderItem | undefined => {
+  const metadata = responseItemMetadata(part, providerMetadataKey)
+  if (
+    ProviderShared.isRecord(metadata?.responseItem) &&
+    typeof metadata.responseItem.id === "string" &&
+    typeof metadata.responseItem.type === "string"
+  )
+    return {
+      ...metadata.responseItem,
+      type: metadata.responseItem.type,
+      id: validResponseItemID(metadata.responseItem.id)
+        ? metadata.responseItem.id
+        : responseItemID("item", metadata.responseItem.id),
+    }
+  if (
+    part.result.type === "json" &&
+    ProviderShared.isRecord(part.result.value) &&
+    typeof part.result.value.id === "string" &&
+    typeof part.result.value.type === "string"
+  )
+    return {
+      ...part.result.value,
+      type: part.result.value.type,
+      id: validResponseItemID(part.result.value.id)
+        ? part.result.value.id
+        : responseItemID("item", part.result.value.id),
+    }
+  return undefined
 }

 const lowerMedia = Effect.fn("OpenResponses.lowerMedia")(function* (
@@ -400,14 +464,15 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
  const system: LoweredInputItem[] =
    request.system.length === 0 ? [] : [{ role: "system", content: ProviderShared.joinText(request.system) }]
  const input: LoweredInputItem[] = [...system]
-  const store = OpenResponsesOptions.resolve(request).store
  const providerMetadataKey = request.model.route.providerMetadataKey ?? "openresponses"

+  // `store` controls server persistence, not client-managed history. Replay the
+  // same stable item identities for stored and stateless requests.
  for (const message of request.messages) {
    if (message.role === "system") {
      const part = yield* ProviderShared.wrappedSystemUpdate(extension.name, message)
      const previous = input.at(-1)
-      if (previous && "role" in previous && previous.role === "user")
+      if (previous && "role" in previous && previous.role === "user" && Array.isArray(previous.content))
        input[input.length - 1] = {
          role: "user",
          content: [...previous.content, { type: "input_text", text: part.text }],
@@ -427,8 +492,8 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
    if (message.role === "assistant") {
      const content: TextPart[] = []
      const reasoningItems: Record<string, OpenResponsesReasoningReplay> = {}
-      const reasoningReferences = new Set<string>()
-      const hostedToolReferences = new Set<string>()
+      const hostedToolItems = new Set<string>()
+      let textItemIndex = 0
      const flushText = () => {
        if (content.length === 0) return
        const groups = content.reduce<Array<{ phase: MessagePhase | null | undefined; parts: TextPart[] }>>(
@@ -443,11 +508,26 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
          [],
        )
        input.push(
-          ...groups.map((group) => ({
-            role: "assistant" as const,
-            content: group.parts.map((part) => ({ type: "output_text" as const, text: part.text })),
-            ...(group.phase === undefined ? {} : { phase: group.phase }),
-          })),
+          ...groups.map((group) => {
+            const index = textItemIndex++
+            const first = group.parts[0]
+            const metadata = first ? responseItemMetadata(first, providerMetadataKey) : undefined
+            const id = validResponseItemID(metadata?.itemId)
+              ? metadata.itemId
+              : message.id === undefined && typeof metadata?.itemId !== "string"
+                ? undefined
+                : index === 0 && validResponseItemID(message.id)
+                  ? message.id
+                  : responseItemID("msg", `${message.id ?? metadata?.itemId}_${index}`)
+            return {
+              type: "message" as const,
+              ...(id === undefined ? {} : { id }),
+              role: "assistant" as const,
+              content: group.parts.map((part) => ({ type: "output_text" as const, text: part.text })),
+              status: "completed" as const,
+              ...(group.phase === undefined ? {} : { phase: group.phase }),
+            }
+          }),
        )
        content.splice(0, content.length)
      }
@@ -460,11 +540,6 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
          flushText()
          const reasoning = lowerReasoning(part, providerMetadataKey)
          if (!reasoning) continue
-          if (store !== false) {
-            if (!reasoningReferences.has(reasoning.id)) input.push({ type: "item_reference", id: reasoning.id })
-            reasoningReferences.add(reasoning.id)
-            continue
-          }
          const existing = reasoningItems[reasoning.id]
          if (existing) {
            existing.summary.push(...reasoning.summary)
@@ -472,11 +547,7 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
              existing.encrypted_content = reasoning.encrypted_content
            continue
          }
-          const replay = {
-            type: reasoning.type,
-            summary: reasoning.summary,
-            encrypted_content: reasoning.encrypted_content,
-          }
+          const replay = { ...reasoning }
          reasoningItems[reasoning.id] = replay
          input.push(replay)
          continue
@@ -484,22 +555,21 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
        if (part.type === "tool-call") {
          flushText()
          if (part.providerExecuted === true) continue
-          input.push(lowerToolCall(part))
+          input.push(lowerToolCall(part, providerMetadataKey))
          continue
        }
        if (part.type === "tool-result" && part.providerExecuted === true) {
          flushText()
-          const itemID = hostedToolItemID(part, providerMetadataKey)
-          if (store !== false && itemID && !hostedToolReferences.has(itemID))
-            input.push({ type: "item_reference", id: itemID })
-          if (store === false && part.result.type === "content") {
+          const item = hostedToolItem(part, providerMetadataKey)
+          if (item && !hostedToolItems.has(item.id)) input.push(item)
+          if (!item && part.result.type === "content") {
            const content: ReadonlyArray<Content> = part.result.value
            input.push({
              role: "user",
              content: yield* Effect.forEach(content, (item) => lowerToolResultContentItem(item, request, extension)),
            })
          }
-          if (itemID) hostedToolReferences.add(itemID)
+          if (item) hostedToolItems.add(item.id)
          continue
        }
        return yield* ProviderShared.unsupportedContent(extension.name, "assistant", [
@@ -518,20 +588,15 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
        return yield* ProviderShared.unsupportedContent(extension.name, "tool", ["tool-result"])
      input.push({
        type: "function_call_output",
+        id: responseItemID("fco", part.id),
        call_id: part.id,
        output: yield* lowerToolResultOutput(part, request, extension),
+        status: "completed",
      })
    }
  }

-  // With store:false, Responses APIs only accept previous reasoning items when the
-  // complete item has encrypted state. Summary blocks for one item may carry
-  // that state only on the last block, so filter after they have been joined.
-  return store === false
-    ? input.filter(
-        (item) => !("type" in item) || item.type !== "reasoning" || typeof item.encrypted_content === "string",
-      )
-    : input
+  return input
 })

 const lowerOptions = (request: LLMRequest) => {
@@ -641,7 +706,7 @@ const onOutputTextDelta = (state: ParserState, event: Event, id: string): StepRe
  if (!event.delta) return [state, NO_EVENTS]
  const events: LLMEvent[] = []
  const phase = state.messagePhases[id]
-  const metadata = phase === undefined ? undefined : providerMetadata(state, { phase })
+  const metadata = providerMetadata(state, { itemId: id, ...(phase === undefined ? {} : { phase }) })
  const lifecycle = Lifecycle.textStart(state.lifecycle, events, id, metadata)
  return [{ ...state, lifecycle: Lifecycle.textDelta(lifecycle, events, id, event.delta) }, events]
 }
@@ -652,7 +717,13 @@ const onOutputTextDone = (state: ParserState, event: Event, id: string): StepRes
    return onOutputTextDelta(state, { ...event, delta: event.text }, id)
  }
  const events: LLMEvent[] = []
-  return [{ ...state, lifecycle: Lifecycle.textEnd(state.lifecycle, events, id) }, events]
+  return [
+    {
+      ...state,
+      lifecycle: Lifecycle.textEnd(state.lifecycle, events, id, providerMetadata(state, { itemId: id })),
+    },
+    events,
+  ]
 }

 export const onReasoningDelta = (state: ParserState, event: Event, itemID: string): StepResult => {
@@ -761,23 +832,11 @@ const onReasoningSummaryPartAdded = (state: ParserState, event: Event): StepResu
  }

  const events: LLMEvent[] = []
-  const closed = Object.entries(item.summaryParts)
-    .filter((entry) => entry[1] === "can-conclude")
-    .reduce(
-      (lifecycle, entry) =>
-        Lifecycle.reasoningEnd(
-          lifecycle,
-          events,
-          `${event.item_id}:${entry[0]}`,
-          providerMetadata(state, { itemId: event.item_id }),
-        ),
-      state.lifecycle,
-    )
  return [
    {
      ...state,
      lifecycle: Lifecycle.reasoningStart(
-        closed,
+        state.lifecycle,
        events,
        `${event.item_id}:${event.summary_index}`,
        providerMetadata(state, { itemId: event.item_id, reasoningEncryptedContent: item.encryptedContent ?? null }),
@@ -787,11 +846,7 @@ const onReasoningSummaryPartAdded = (state: ParserState, event: Event): StepResu
        [event.item_id]: {
          ...item,
          summaryParts: {
-            ...Object.fromEntries(
-              Object.entries(item.summaryParts).map((entry) =>
-                entry[1] === "can-conclude" ? [entry[0], "concluded" as const] : entry,
-              ),
-            ),
+            ...item.summaryParts,
            [event.summary_index]: "active",
          },
        },
@@ -809,22 +864,13 @@ const onReasoningSummaryPartDone = (state: ParserState, event: Event): StepResul
  return [
    {
      ...state,
-      lifecycle:
-        state.store !== false
-          ? Lifecycle.reasoningEnd(
-              state.lifecycle,
-              events,
-              `${event.item_id}:${event.summary_index}`,
-              providerMetadata(state, { itemId: event.item_id }),
-            )
-          : state.lifecycle,
      reasoningItems: {
        ...state.reasoningItems,
        [event.item_id]: {
          ...item,
          summaryParts: {
            ...item.summaryParts,
-            [event.summary_index]: state.store !== false ? "concluded" : "can-conclude",
+            [event.summary_index]: "can-conclude",
          },
        },
      },
@@ -870,7 +916,10 @@ const onOutputItemDone = Effect.fn("OpenResponses.onOutputItemDone")(function* (
          state.lifecycle,
          events,
          item.id,
-          phase === undefined ? undefined : providerMetadata(state, { phase }),
+          providerMetadata(state, {
+            itemId: item.id,
+            ...(phase === undefined ? {} : { phase }),
+          }),
        ),
        messageItems,
        messagePhases,
@@ -1037,7 +1086,6 @@ export const initial = (request: LLMRequest, extension: Extension = BASE): Parse
  messagePhase: (value) => messagePhase(value, extension),
  messagePhases: {},
  reasoningItems: {},
-  store: OpenResponsesOptions.resolve(request).store,
 })

 const messagePhase = (value: unknown, extension: Extension): MessagePhase | null | undefined => {
@@ -241,8 +241,6 @@ export interface ParserState {
  readonly reasoningEmitted: boolean
  readonly latestToolIndex?: number
  readonly nextToolIndex: number
-  readonly outputStarted: boolean
-  readonly requireFinishReason: boolean
 }

 // =============================================================================
@@ -709,7 +707,9 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
      Boolean(delta?.content) ||
      reasoning !== undefined ||
      (Array.isArray(delta?.reasoning_details) && delta.reasoning_details.length > 0) ||
-      toolDeltas.some((tool) => Boolean(tool.id) || Boolean(tool.function?.name) || Boolean(tool.function?.arguments))
+      toolDeltas.some(
+        (tool) => Boolean(tool.id) || Boolean(tool.function?.name) || Boolean(tool.function?.arguments),
+      )
    if (state.finishReason !== undefined) {
      if (hasLateContent)
        return yield* ProviderShared.eventError(ADAPTER, "OpenAI Chat received content after the finish reason")
@@ -749,7 +749,8 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
      const fallback = toolDeltas.length > 1 ? position : (latestToolIndex ?? position)
      const fallbackTool = tools[fallback] ?? pendingTools[fallback]
      const index =
-        tool.index ?? matched ?? (tool.id && fallbackTool?.id && fallbackTool.id !== tool.id ? nextToolIndex : fallback)
+        tool.index ?? matched ??
+        (tool.id && fallbackTool?.id && fallbackTool.id !== tool.id ? nextToolIndex : fallback)
      const current = tools[index]
      const pending = pendingTools[index]
      const id = current?.id ?? pending?.id ?? (tool.id || undefined)
@@ -805,8 +806,6 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
        reasoningEmitted,
        latestToolIndex,
        nextToolIndex,
-        outputStarted: state.outputStarted || hasLateContent,
-        requireFinishReason: state.requireFinishReason,
      },
      events,
    ] as const
@@ -837,23 +836,6 @@ const finishEvents = (state: ParserState): ReadonlyArray<LLMEvent> => {
  return events
 }

-const onHalt = (state: ParserState) =>
-  Effect.gen(function* () {
-    if (state.finishReason !== undefined || state.requireFinishReason) return finishEvents(state)
-    if (!state.outputStarted) return []
-    if (Object.keys(state.pendingTools).length > 0)
-      return yield* ProviderShared.eventError(ADAPTER, "OpenAI Chat tool call delta is missing id or name")
-    // Chat has no per-call stop event, so an accepted EOF must finalize every
-    // accumulated tool input before publishing the synthetic terminal reason.
-    const finished = yield* ToolStream.finishAll(ADAPTER, state.tools)
-    return finishEvents({
-      ...state,
-      tools: finished.tools,
-      toolCallEvents: finished.events,
-      finishReason: { normalized: "unknown" },
-    })
-  })
-
 // =============================================================================
 // Protocol And OpenAI Route
 // =============================================================================
@@ -881,11 +863,9 @@ export const protocol = Protocol.make({
      reasoningDetailsObserved: false,
      reasoningEmitted: false,
      nextToolIndex: 0,
-      outputStarted: false,
-      requireFinishReason: request.model.compatibility?.requireFinishReason ?? true,
    }),
    step,
-    onHalt,
+    onHalt: finishEvents,
  },
 })

@@ -17,6 +17,7 @@ export const route = Route.make({
  protocol: OpenResponses.protocol,
  endpoint: Endpoint.path(OpenResponses.PATH),
  transport: OpenResponses.httpTransport,
+  defaults: { providerOptions: { openresponses: { store: false } } },
 })

 export * as OpenAICompatibleResponses from "./openai-compatible-responses"
@@ -37,10 +37,14 @@ const OpenAIResponsesToolChoice = Schema.Union([

 const OpenAIResponsesInputItem = Schema.Union([
  Schema.Struct({
+    type: Schema.optionalKey(Schema.tag("message")),
+    id: Schema.optionalKey(Schema.String),
    role: Schema.tag("assistant"),
    content: Schema.Array(Schema.Struct({ type: Schema.tag("output_text"), text: Schema.String })),
+    status: Schema.optionalKey(Schema.Literals(["in_progress", "completed", "incomplete"])),
    phase: Schema.optionalKey(Schema.NullOr(OpenResponses.MessagePhase)),
  }),
+  OpenResponses.ProviderItem,
  OpenResponses.InputItem,
 ])

@@ -195,7 +199,8 @@ const onHostedToolDone = Effect.fn("OpenAIResponses.onHostedToolDone")(function*
  item: HostedToolItem,
 ) {
  const tool = HOSTED_TOOLS[item.type]
-  const providerMetadata = OpenResponses.providerMetadata(state, { itemId: item.id })
+  const callMetadata = OpenResponses.providerMetadata(state, { itemId: item.id })
+  const resultMetadata = OpenResponses.providerMetadata(state, { itemId: item.id, responseItem: item })
  const events: LLMEvent[] = []
  const lifecycle = Lifecycle.stepStart(state.lifecycle, events)
  events.push(
@@ -204,14 +209,14 @@ const onHostedToolDone = Effect.fn("OpenAIResponses.onHostedToolDone")(function*
      name: tool.name,
      input: tool.input(item),
      providerExecuted: true,
-      providerMetadata,
+      providerMetadata: callMetadata,
    }),
    LLMEvent.toolResult({
      id: item.id,
      name: tool.name,
      result: yield* hostedToolResult(item),
      providerExecuted: true,
-      providerMetadata,
+      providerMetadata: resultMetadata,
    }),
  )
  return [{ ...state, lifecycle }, events] satisfies OpenResponses.StepResult
@@ -2,11 +2,11 @@ import { LLMEvent, type FinishReasonDetails, type ProviderMetadata, type Usage }

 export interface State {
  readonly stepStarted: boolean
-  readonly text: ReadonlySet<string>
-  readonly reasoning: ReadonlySet<string>
+  readonly text: ReadonlyMap<string, ProviderMetadata | undefined>
+  readonly reasoning: ReadonlyMap<string, ProviderMetadata | undefined>
 }

-export const initial = (): State => ({ stepStarted: false, text: new Set(), reasoning: new Set() })
+export const initial = (): State => ({ stepStarted: false, text: new Map(), reasoning: new Map() })

 export const stepStart = (state: State, events: LLMEvent[]): State => {
  if (state.stepStarted) return state
@@ -18,7 +18,7 @@ export const textStart = (state: State, events: LLMEvent[], id: string, provider
  if (state.text.has(id)) return state
  const stepped = stepStart(state, events)
  events.push(LLMEvent.textStart({ id, providerMetadata }))
-  return { ...stepped, text: new Set([...stepped.text, id]) }
+  return { ...stepped, text: new Map([...stepped.text, [id, providerMetadata]]) }
 }

 export const textDelta = (state: State, events: LLMEvent[], id: string, text: string): State => {
@@ -36,7 +36,7 @@ export const reasoningStart = (
  if (state.reasoning.has(id)) return state
  const stepped = stepStart(state, events)
  events.push(LLMEvent.reasoningStart({ id, providerMetadata }))
-  return { ...stepped, reasoning: new Set([...stepped.reasoning, id]) }
+  return { ...stepped, reasoning: new Map([...stepped.reasoning, [id, providerMetadata]]) }
 }

 export const reasoningDelta = (
@@ -59,8 +59,10 @@ export const reasoningEnd = (
 ): State => {
  if (!state.reasoning.has(id)) return state
  const stepped = stepStart(state, events)
-  events.push(LLMEvent.reasoningEnd({ id, providerMetadata }))
-  const reasoning = new Set(stepped.reasoning)
+  events.push(
+    LLMEvent.reasoningEnd({ id, providerMetadata: mergeMetadata(stepped.reasoning.get(id), providerMetadata) }),
+  )
+  const reasoning = new Map(stepped.reasoning)
  reasoning.delete(id)
  return { ...stepped, reasoning }
 }
@@ -68,16 +70,24 @@ export const reasoningEnd = (
 export const textEnd = (state: State, events: LLMEvent[], id: string, providerMetadata?: ProviderMetadata): State => {
  if (!state.text.has(id)) return state
  const stepped = stepStart(state, events)
-  events.push(LLMEvent.textEnd({ id, providerMetadata }))
-  const text = new Set(stepped.text)
+  events.push(LLMEvent.textEnd({ id, providerMetadata: mergeMetadata(stepped.text.get(id), providerMetadata) }))
+  const text = new Map(stepped.text)
  text.delete(id)
  return { ...stepped, text }
 }

+const mergeMetadata = (left: ProviderMetadata | undefined, right: ProviderMetadata | undefined) => {
+  if (left === undefined) return right
+  if (right === undefined) return left
+  return Object.fromEntries(
+    Array.from(new Set([...Object.keys(left), ...Object.keys(right)]), (key) => [key, { ...left[key], ...right[key] }]),
+  )
+}
+
 const closeOpenBlocks = (state: State, events: LLMEvent[]): State => {
-  for (const id of state.reasoning) events.push(LLMEvent.reasoningEnd({ id }))
-  for (const id of state.text) events.push(LLMEvent.textEnd({ id }))
-  return { ...state, text: new Set(), reasoning: new Set() }
+  for (const [id, providerMetadata] of state.reasoning) events.push(LLMEvent.reasoningEnd({ id, providerMetadata }))
+  for (const [id, providerMetadata] of state.text) events.push(LLMEvent.textEnd({ id, providerMetadata }))
+  return { ...state, text: new Map(), reasoning: new Map() }
 }

 export const finish = (
@@ -7,7 +7,7 @@ import { Framing } from "./framing"
 import { HttpTransport } from "./transport"
 import type { HttpMiddleware, Transport, TransportRuntime } from "./transport"
 import { WebSocketExecutor } from "./transport"
-import type { Protocol, ProtocolStream } from "./protocol"
+import type { Protocol } from "./protocol"
 import { applyCachePolicy } from "../cache-policy"
 import * as ProviderShared from "../protocols/shared"
 import type { ProtocolID, ProviderOptions } from "../schema"
@@ -243,56 +243,28 @@ const incompleteStreamError = (route: string) =>
    }),
  })

-const ensureTerminalEvent = (route: string, required: boolean) => (events: Stream.Stream<LLMEvent, AIError>) =>
+const requireTerminalEvent = (route: string) => (events: Stream.Stream<LLMEvent, AIError>) =>
  Stream.suspend(() => {
    let terminal = false
-    let output = false
-    const fallback = Stream.suspend(() => {
-      if (terminal) return Stream.empty
-      if (required || !output) return Stream.fail(incompleteStreamError(route))
-      // The compatibility override trusts a clean stream end, but it cannot
-      // recover the provider's omitted reason.
-      const reason = { normalized: "unknown" as const }
-      return Stream.make(LLMEvent.stepFinish({ index: 0, reason }), LLMEvent.finish({ reason }))
-    })
    return events.pipe(
      Stream.mapEffect((event) => {
        if (terminal)
          return Effect.fail(
            ProviderShared.eventError(route, `Provider emitted ${event.type} after the terminal event`),
          )
-        output = true
        if (LLMEvent.is.finish(event) || LLMEvent.is.providerError(event)) terminal = true
        return Effect.succeed(event)
      }),
-      Stream.concat(fallback),
+      Stream.onEnd(
+        Effect.suspend(() =>
+          terminal
+            ? Effect.void
+            : Effect.fail(incompleteStreamError(route)),
+        ),
+      ),
    )
  })

-type ProtocolEvent<Event> = { readonly type: "event"; readonly event: Event } | { readonly type: "halt" }
-
-const parseProtocolEvents = <Event, State>(
-  events: Stream.Stream<Event, AIError>,
-  request: LLMRequest,
-  protocol: { readonly stream: ProtocolStream<unknown, Event, State> },
-) =>
-  events.pipe(
-    Stream.map((event): ProtocolEvent<Event> => ({ type: "event", event })),
-    // A normal halt becomes an in-band parser input so finalization may fail.
-    Stream.concat(Stream.succeed({ type: "halt" } as const)),
-    Stream.mapAccumEffect(
-      () => protocol.stream.initial(request),
-      (state, event) => {
-        if (event.type === "event") return protocol.stream.step(state, event.event)
-        if (!protocol.stream.onHalt) return Effect.succeed([state, []] as const)
-        const events = protocol.stream.onHalt(state)
-        return Effect.isEffect(events)
-          ? events.pipe(Effect.map((events) => [state, events] as const))
-          : Effect.succeed([state, events] as const)
-      },
-    ),
-  )
-
 function makeFromTransport<Body, Prepared, Frame, Event, State>(
  input: MakeTransportInput<Body, Prepared, Frame, Event, State>,
 ): Route<Body, Prepared> {
@@ -357,9 +329,14 @@ function makeFromTransport<Body, Prepared, Frame, Event, State>(
            Stream.mapEffect(decodeEvent(route)),
            protocol.stream.terminal ? Stream.takeUntil(protocol.stream.terminal) : (stream) => stream,
          )
-        return parseProtocolEvents(events, request, protocol).pipe(
+        return events.pipe(
+          Stream.mapAccumEffect(
+            () => protocol.stream.initial(request),
+            protocol.stream.step,
+            protocol.stream.onHalt ? { onHalt: protocol.stream.onHalt } : undefined,
+          ),
          Stream.catchCause((cause) => Stream.fail(streamError(route, `Failed to read ${route} stream`, cause))),
-          ensureTerminalEvent(route, request.model.compatibility?.requireFinishReason ?? true),
+          requireTerminalEvent(route),
        )
      },
    } satisfies Route<Body, Prepared>
@@ -59,8 +59,8 @@ export interface ProtocolStream<Frame, Event, State> {
  readonly step: (state: State, event: Event) => Effect.Effect<readonly [State, ReadonlyArray<LLMEvent>], AIError>
  /** Optional request-completion signal for transports that do not end naturally. */
  readonly terminal?: (event: Event) => boolean
-  /** Optional flush emitted when the framed stream ends successfully. */
-  readonly onHalt?: (state: State) => ReadonlyArray<LLMEvent> | Effect.Effect<ReadonlyArray<LLMEvent>, AIError>
+  /** Optional flush emitted when the framed stream ends. */
+  readonly onHalt?: (state: State) => ReadonlyArray<LLMEvent>
 }

 /**
@@ -138,29 +138,6 @@ describe("llm route", () => {
    }),
  )

-  unterminated.effect("synthesizes an unknown finish when a terminal event is not required", () =>
-    Effect.gen(function* () {
-      const response = yield* (yield* LLMClient.Service).generate(
-        LLMRequest.update(request, {
-          model: updateModel(request.model, { compatibility: { requireFinishReason: false } }),
-        }),
-      )
-
-      expect(response.text).toBe("partial")
-      expect(response.finishReason).toEqual({ normalized: "unknown" })
-      expect(response.events.slice(-2)).toEqual([
-        {
-          type: "step-finish",
-          index: 0,
-          reason: { normalized: "unknown" },
-          usage: undefined,
-          providerMetadata: undefined,
-        },
-        { type: "finish", reason: { normalized: "unknown" }, usage: undefined },
-      ])
-    }),
-  )
-
  it.effect("selects routes by model route value", () =>
    Effect.gen(function* () {
      const prepared = yield* compileRequest(
@@ -44,7 +44,7 @@
        "headers": {
          "content-type": "application/json"
        },
-        "body": "{\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Think briefly, then reply exactly with: Hello!\"}]},{\"type\":\"reasoning\",\"summary\":[],\"encrypted_content\":\"gAAAAABqEjXoGMCw3WDXpoD9151PEr2Lt8raW7KBKefQhZJGWx5f8jy152bApO6oE-Mr1BhUtfZNq3OPBVfSL4ioQ9bHREfujIBXgk9LUDBAz2Sle7KjOr9HaUV16A4HBiaFIRFjsHPS9G8yEySp1m6F1CD_WR6apyUGgugRh_y39EcOJmxPOzmiac5DVM6fraA1VpcGbqrZ1x2ANHFDOfnYTycPtPNTgzE7LjkYjDDWbT03uN1YxfP4pqjDVRzY14pA8bSZ8ys-pDv5kUFCAsw-OlU4jYKUXp-M8_6KTaRQP71LPwppt__zG_NJPfy-qUil4pOU8_NoxtxerHgLLXbfExZdzfpoGinoEjn7nj7BJDEtl-LNeNEb5c-1ZymNfVMp-Cs3fLEPkAV8rtHFtZ0MhE_07GKbGo7hTrOmkM4DydxmHsdWGNbXAG35cprslEA5P7p3GHFKnRs5hGs2eq-XcZ3yki64ZBOU_Tv6UR7nUH09gF1rdrJo3dpre6M00COwwdZ02zUP5KxCuI8FKu2jsZu9zgMVXDALsdtM5orTCVLXsn4rddWd111zE-vMjNmMMmktW2cHMjH7j1ooA-9P083koNVYiLi4UhMA64gTqgyl8MxkZekl7eFSMa7qk295NaHOKtFxzYYcZ9jdioCwSPSZ0ZZWLoNgrK7SWfRh0uaTHNcMZ3wq8ae6CguktIeVTCPTQAqJLQqd7AU0oOCKCJ7BWnC-L8UC6m7Pm9ZS958uUVeWBhgKHzMAGq9UeQB7IEeAcbMn3EDgOSfd8qCb8iwU9iG9dcu9axQwWU7pd7kd-T-He61W7z5wWgpx1KehWCxrN6kuKSo6p-uUfwVnJukreOn8BJNAzADQgz68bhmN9VGih7YcKVnLgwDwKditrjSd6-tfE0Baarj3jWENvT6ohY17R9FDrKS-2v8IIX6tGjoKJw8SRhaWLNv4vWlmxRgR0gdac3qumd0GKqsWSveNz01naA==\"},{\"role\":\"assistant\",\"content\":[{\"type\":\"output_text\",\"text\":\"Hello!\"}]},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Now reply exactly with: Done.\"}]}],\"store\":false,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"low\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"max_output_tokens\":40,\"stream\":true}"
+        "body": "{\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Think briefly, then reply exactly with: Hello!\"}]},{\"type\":\"reasoning\",\"summary\":[],\"encrypted_content\":\"gAAAAABqEjXoGMCw3WDXpoD9151PEr2Lt8raW7KBKefQhZJGWx5f8jy152bApO6oE-Mr1BhUtfZNq3OPBVfSL4ioQ9bHREfujIBXgk9LUDBAz2Sle7KjOr9HaUV16A4HBiaFIRFjsHPS9G8yEySp1m6F1CD_WR6apyUGgugRh_y39EcOJmxPOzmiac5DVM6fraA1VpcGbqrZ1x2ANHFDOfnYTycPtPNTgzE7LjkYjDDWbT03uN1YxfP4pqjDVRzY14pA8bSZ8ys-pDv5kUFCAsw-OlU4jYKUXp-M8_6KTaRQP71LPwppt__zG_NJPfy-qUil4pOU8_NoxtxerHgLLXbfExZdzfpoGinoEjn7nj7BJDEtl-LNeNEb5c-1ZymNfVMp-Cs3fLEPkAV8rtHFtZ0MhE_07GKbGo7hTrOmkM4DydxmHsdWGNbXAG35cprslEA5P7p3GHFKnRs5hGs2eq-XcZ3yki64ZBOU_Tv6UR7nUH09gF1rdrJo3dpre6M00COwwdZ02zUP5KxCuI8FKu2jsZu9zgMVXDALsdtM5orTCVLXsn4rddWd111zE-vMjNmMMmktW2cHMjH7j1ooA-9P083koNVYiLi4UhMA64gTqgyl8MxkZekl7eFSMa7qk295NaHOKtFxzYYcZ9jdioCwSPSZ0ZZWLoNgrK7SWfRh0uaTHNcMZ3wq8ae6CguktIeVTCPTQAqJLQqd7AU0oOCKCJ7BWnC-L8UC6m7Pm9ZS958uUVeWBhgKHzMAGq9UeQB7IEeAcbMn3EDgOSfd8qCb8iwU9iG9dcu9axQwWU7pd7kd-T-He61W7z5wWgpx1KehWCxrN6kuKSo6p-uUfwVnJukreOn8BJNAzADQgz68bhmN9VGih7YcKVnLgwDwKditrjSd6-tfE0Baarj3jWENvT6ohY17R9FDrKS-2v8IIX6tGjoKJw8SRhaWLNv4vWlmxRgR0gdac3qumd0GKqsWSveNz01naA==\",\"id\":\"rs_0a0794dab3b8ec7d016a1235e7ce3881958a5eca32a36a14c5\"},{\"role\":\"assistant\",\"content\":[{\"type\":\"output_text\",\"text\":\"Hello!\"}],\"type\":\"message\",\"status\":\"completed\",\"id\":\"msg_0a0794dab3b8ec7d016a1235e8d64c81959a41f8db3ea7b66c\",\"phase\":\"final_answer\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Now reply exactly with: Done.\"}]}],\"store\":false,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"low\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"max_output_tokens\":40,\"stream\":true}"
      },
      "response": {
        "status": 200,
@@ -21,7 +21,7 @@
        "headers": {
          "content-type": "application/json"
        },
-        "body": "{\"model\":\"gpt-4o-mini\",\"input\":[{\"role\":\"system\",\"content\":\"Read the PDF returned by the tool and follow the user's response format exactly.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Return only the verification code from the PDF.\"}]},{\"type\":\"function_call\",\"call_id\":\"call_pdf_1\",\"name\":\"read_pdf\",\"arguments\":\"{}\"},{\"type\":\"function_call_output\",\"call_id\":\"call_pdf_1\",\"output\":[{\"type\":\"input_text\",\"text\":\"PDF read successfully\"},{\"type\":\"input_file\",\"filename\":\"verification.pdf\",\"file_data\":\"data:application/pdf;base64,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\"}]}],\"tools\":[{\"type\":\"function\",\"name\":\"read_pdf\",\"description\":\"Read the attached PDF.\",\"parameters\":{\"type\":\"object\",\"properties\":{},\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"max_output_tokens\":40,\"temperature\":0,\"stream\":true}"
+        "body": "{\"model\":\"gpt-4o-mini\",\"input\":[{\"role\":\"system\",\"content\":\"Read the PDF returned by the tool and follow the user's response format exactly.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Return only the verification code from the PDF.\"}]},{\"type\":\"function_call\",\"call_id\":\"call_pdf_1\",\"name\":\"read_pdf\",\"arguments\":\"{}\",\"id\":\"fc_pdf_1\",\"status\":\"completed\"},{\"type\":\"function_call_output\",\"call_id\":\"call_pdf_1\",\"output\":[{\"type\":\"input_text\",\"text\":\"PDF read successfully\"},{\"type\":\"input_file\",\"filename\":\"verification.pdf\",\"file_data\":\"data:application/pdf;base64,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\"}],\"id\":\"fco_pdf_1\",\"status\":\"completed\"}],\"tools\":[{\"type\":\"function\",\"name\":\"read_pdf\",\"description\":\"Read the attached PDF.\",\"parameters\":{\"type\":\"object\",\"properties\":{},\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"max_output_tokens\":40,\"temperature\":0,\"stream\":true}"
      },
      "response": {
        "status": 200,
@@ -21,7 +21,7 @@
        "headers": {
          "content-type": "application/json"
        },
-        "body": "{\"model\":\"grok-4.5\",\"input\":[{\"role\":\"system\",\"content\":\"Read the PDF returned by the tool and follow the user's response format exactly.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Return only the verification code from the PDF.\"}]},{\"type\":\"function_call\",\"call_id\":\"call_pdf_1\",\"name\":\"read_pdf\",\"arguments\":\"{}\"},{\"type\":\"function_call_output\",\"call_id\":\"call_pdf_1\",\"output\":[{\"type\":\"input_text\",\"text\":\"PDF read successfully\"},{\"type\":\"input_file\",\"filename\":\"verification.pdf\",\"file_data\":\"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\",\"mime_type\":\"application/pdf\"}]}],\"tools\":[{\"type\":\"function\",\"name\":\"read_pdf\",\"description\":\"Read the attached PDF.\",\"parameters\":{\"type\":\"object\",\"properties\":{},\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"max_output_tokens\":40,\"temperature\":0,\"stream\":true}"
+        "body": "{\"model\":\"grok-4.5\",\"input\":[{\"role\":\"system\",\"content\":\"Read the PDF returned by the tool and follow the user's response format exactly.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Return only the verification code from the PDF.\"}]},{\"type\":\"function_call\",\"call_id\":\"call_pdf_1\",\"name\":\"read_pdf\",\"arguments\":\"{}\",\"id\":\"fc_pdf_1\",\"status\":\"completed\"},{\"type\":\"function_call_output\",\"call_id\":\"call_pdf_1\",\"output\":[{\"type\":\"input_text\",\"text\":\"PDF read successfully\"},{\"type\":\"input_file\",\"filename\":\"verification.pdf\",\"file_data\":\"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\",\"mime_type\":\"application/pdf\"}],\"id\":\"fco_pdf_1\",\"status\":\"completed\"}],\"tools\":[{\"type\":\"function\",\"name\":\"read_pdf\",\"description\":\"Read the attached PDF.\",\"parameters\":{\"type\":\"object\",\"properties\":{},\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"max_output_tokens\":40,\"temperature\":0,\"stream\":true}"
      },
      "response": {
        "status": 200,
@@ -40,10 +40,6 @@ const request = LLM.request({
  generation: { maxTokens: 20, temperature: 0 },
 })

-const optionalFinishRequest = LLMRequest.update(request, {
-  model: LanguageModel.update(model, { compatibility: { requireFinishReason: false } }),
-})
-
 describe("OpenAI Chat route", () => {
  it.effect("prepares OpenAI Chat payload", () =>
    Effect.gen(function* () {
@@ -600,20 +596,6 @@ describe("OpenAI Chat route", () => {
    }),
  )

-  it.effect("accepts text and usage without a finish reason when configured", () =>
-    Effect.gen(function* () {
-      const body = sseEvents(
-        deltaChunk({ role: "assistant", content: "Hello" }),
-        usageChunk({ prompt_tokens: 5, completion_tokens: 1, total_tokens: 6 }),
-      )
-      const response = yield* LLMClient.generate(optionalFinishRequest).pipe(Effect.provide(fixedResponse(body)))
-
-      expect(response.text).toBe("Hello")
-      expect(response.finishReason).toEqual({ normalized: "unknown" })
-      expect(response.usage).toMatchObject({ inputTokens: 5, outputTokens: 1, totalTokens: 6 })
-    }),
-  )
-
  it.effect("parses and replays OpenAI-compatible reasoning fields", () =>
    Effect.gen(function* () {
      const fields = ["reasoning_content", "reasoning", "reasoning_text"] as const
@@ -1163,89 +1145,21 @@ describe("OpenAI Chat route", () => {
    }),
  )

-  it.effect("finalizes a streamed tool call without a finish reason when configured", () =>
-    Effect.gen(function* () {
-      const body = sseEvents(
-        deltaChunk({
-          role: "assistant",
-          tool_calls: [{ index: 0, id: "call_1", function: { name: "lookup", arguments: '{"query"' } }],
-        }),
-        deltaChunk({ tool_calls: [{ index: 0, function: { arguments: ':"weather"}' } }] }),
-      )
-      const response = yield* LLMClient.generate(
-        LLMRequest.update(optionalFinishRequest, {
-          tools: [ToolDefinition.make({ name: "lookup", description: "Lookup data", inputSchema: { type: "object" } })],
-        }),
-      ).pipe(Effect.provide(fixedResponse(body)))
-
-      expect(response.toolCalls).toMatchObject([{ id: "call_1", name: "lookup", input: { query: "weather" } }])
-      expect(response.finishReason).toEqual({ normalized: "unknown" })
-    }),
-  )
-
-  it.effect("settles malformed tool input without a finish reason when configured", () =>
-    Effect.gen(function* () {
-      const body = sseEvents(
-        deltaChunk({
-          role: "assistant",
-          tool_calls: [{ index: 0, id: "call_1", function: { name: "lookup", arguments: '{"query"' } }],
-        }),
-      )
-      const response = yield* LLMClient.generate(
-        LLMRequest.update(optionalFinishRequest, {
-          tools: [ToolDefinition.make({ name: "lookup", description: "Lookup data", inputSchema: { type: "object" } })],
-        }),
-      ).pipe(Effect.provide(fixedResponse(body)))
-
-      expect(response.events.filter(LLMEvent.is.toolInputError)).toMatchObject([
-        { id: "call_1", name: "lookup", raw: '{"query"' },
-      ])
-      expect(response.toolCalls).toEqual([])
-      expect(response.finishReason).toEqual({ normalized: "unknown" })
-    }),
-  )
-
-  it.effect("rejects incomplete tool identity without a finish reason when configured", () =>
-    Effect.gen(function* () {
-      const body = sseEvents(deltaChunk({ tool_calls: [{ index: 0, id: "call_1", function: { arguments: "{}" } }] }))
-      const error = yield* LLMClient.generate(optionalFinishRequest).pipe(
-        Effect.provide(fixedResponse(body)),
-        Effect.flip,
-      )
-
-      expect(error.message).toContain("OpenAI Chat tool call delta is missing id or name")
-    }),
-  )
-
-  it.effect("rejects an empty stream when a finish reason is not required", () =>
-    Effect.gen(function* () {
-      const error = yield* LLMClient.generate(optionalFinishRequest).pipe(
-        Effect.provide(fixedResponse(sseEvents())),
-        Effect.flip,
-      )
-
-      expect(error.reason).toMatchObject({ _tag: "InvalidProviderOutput", classification: "incomplete-stream" })
-    }),
-  )
-
-  it.effect("fails on malformed stream events when a finish reason is not required", () =>
+  it.effect("fails on malformed stream events", () =>
    Effect.gen(function* () {
      const body = sseEvents(deltaChunk({ content: 123 }))
-      const error = yield* LLMClient.generate(optionalFinishRequest).pipe(
-        Effect.provide(fixedResponse(body)),
-        Effect.flip,
-      )
+      const error = yield* LLMClient.generate(request).pipe(Effect.provide(fixedResponse(body)), Effect.flip)

      expect(error.message).toContain("Invalid openai/openai-chat stream event")
    }),
  )

-  it.effect("surfaces transport errors when a finish reason is not required", () =>
+  it.effect("surfaces transport errors that occur mid-stream", () =>
    Effect.gen(function* () {
      const layer = truncatedStream([
        `data: ${JSON.stringify(deltaChunk({ role: "assistant", content: "Hello" }))}\n\n`,
      ])
-      const error = yield* LLMClient.generate(optionalFinishRequest).pipe(Effect.provide(layer), Effect.flip)
+      const error = yield* LLMClient.generate(request).pipe(Effect.provide(layer), Effect.flip)

      expect(error.message).toContain("Failed to read openai/openai-chat stream")
    }),
@@ -51,7 +51,13 @@ describe("Open Responses-compatible route", () => {
          { role: "system", content: "You are concise." },
          { role: "user", content: [{ type: "input_text", text: "Say hello." }] },
        ],
+        store: false,
        stream: true,
+        max_output_tokens: undefined,
+        temperature: undefined,
+        tool_choice: undefined,
+        tools: undefined,
+        top_p: undefined,
      })
    }),
  )
@@ -112,6 +118,40 @@ describe("Open Responses-compatible route", () => {
    }),
  )

+  it.effect("keeps response item replay independent of store", () =>
+    Effect.gen(function* () {
+      const model = configure({
+        apiKey: "test-key",
+        baseURL: "https://responses.example.test/v1",
+      }).model("example-model")
+      const messages = [
+        Message.assistant({
+          type: "reasoning",
+          text: "Checked the previous diff.",
+          providerMetadata: {
+            openresponses: { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" },
+          },
+        }),
+      ]
+      const stored = yield* compileRequest(
+        LLM.request({ model, messages, providerOptions: { openresponses: { store: true } } }),
+      )
+      const stateless = yield* compileRequest(
+        LLM.request({ model, messages, providerOptions: { openresponses: { store: false } } }),
+      )
+
+      expect(stored.body.input).toEqual(stateless.body.input)
+      expect(stored.body.input).toEqual([
+        {
+          type: "reasoning",
+          id: "rs_1",
+          summary: [{ type: "summary_text", text: "Checked the previous diff." }],
+          encrypted_content: "encrypted-state",
+        },
+      ])
+    }),
+  )
+
  it.effect("does not interpret OpenAI hosted-tool items", () =>
    Effect.gen(function* () {
      const model = configure({
@@ -211,7 +211,12 @@ describe("OpenAI Responses route", () => {
            { type: "input_text", text: "<system-update>\nTreat &lt;/system-update&gt; literally.\n</system-update>" },
          ],
        },
-        { role: "assistant", content: [{ type: "output_text", text: "After." }] },
+        {
+          type: "message",
+          role: "assistant",
+          content: [{ type: "output_text", text: "After." }],
+          status: "completed",
+        },
      ])
    }),
  )
@@ -329,7 +334,7 @@ describe("OpenAI Responses route", () => {
      yield* LLMClient.generate(
        LLMRequest.update(request, {
          model: Azure.configure({
-            baseURL: "https://opencode-test.openai.azure.com/openai/v1/",
+            resourceName: "opencode-test",
            apiKey: "azure-key",
            headers: { authorization: "Bearer stale" },
          }).responses("gpt-4.1-mini"),
@@ -414,8 +419,21 @@ describe("OpenAI Responses route", () => {
        model: "gpt-4.1-mini",
        input: [
          { role: "user", content: [{ type: "input_text", text: "What is the weather?" }] },
-          { type: "function_call", call_id: "call_1", name: "lookup", arguments: '{"query":"weather"}' },
-          { type: "function_call_output", call_id: "call_1", output: '{"forecast":"sunny"}' },
+          {
+            type: "function_call",
+            id: "fc_1",
+            call_id: "call_1",
+            name: "lookup",
+            arguments: '{"query":"weather"}',
+            status: "completed",
+          },
+          {
+            type: "function_call_output",
+            id: "fco_1",
+            call_id: "call_1",
+            output: '{"forecast":"sunny"}',
+            status: "completed",
+          },
        ],
        store: false,
        stream: true,
@@ -864,10 +882,10 @@ describe("OpenAI Responses route", () => {
      expect(response.text).toBe("Hello!")
      expect(response.events).toEqual([
        { type: "step-start", index: 0 },
-        { type: "text-start", id: "msg_1" },
+        { type: "text-start", id: "msg_1", providerMetadata: { openai: { itemId: "msg_1" } } },
        { type: "text-delta", id: "msg_1", text: "Hello" },
        { type: "text-delta", id: "msg_1", text: "!" },
-        { type: "text-end", id: "msg_1" },
+        { type: "text-end", id: "msg_1", providerMetadata: { openai: { itemId: "msg_1" } } },
        {
          type: "step-finish",
          index: 0,
@@ -923,35 +941,44 @@ describe("OpenAI Responses route", () => {
        {
          type: "text",
          text: "Checking.",
-          providerMetadata: { openai: { phase: "commentary" } },
+          providerMetadata: { openai: { itemId: "msg_commentary", phase: "commentary" } },
        },
        {
          type: "text",
          text: "Finished.",
-          providerMetadata: { openai: { phase: "final_answer" } },
+          providerMetadata: { openai: { itemId: "msg_final", phase: "final_answer" } },
        },
        {
          type: "text",
          text: "Unclassified.",
-          providerMetadata: { openai: { phase: null } },
+          providerMetadata: { openai: { itemId: "msg_null", phase: null } },
        },
      ])

      const prepared = yield* compileRequest(LLM.request({ model, messages: [response.message] }))
      expect(prepared.body.input).toEqual([
        {
+          type: "message",
+          id: "msg_commentary",
          role: "assistant",
          content: [{ type: "output_text", text: "Checking." }],
+          status: "completed",
          phase: "commentary",
        },
        {
+          type: "message",
+          id: "msg_final",
          role: "assistant",
          content: [{ type: "output_text", text: "Finished." }],
+          status: "completed",
          phase: "final_answer",
        },
        {
+          type: "message",
+          id: "msg_null",
          role: "assistant",
          content: [{ type: "output_text", text: "Unclassified." }],
+          status: "completed",
          phase: null,
        },
      ])
@@ -1043,12 +1070,12 @@ describe("OpenAI Responses route", () => {
      )

      expect(response.events.filter((event) => event.type.startsWith("text-"))).toEqual([
-        { type: "text-start", id: "msg_1" },
+        { type: "text-start", id: "msg_1", providerMetadata: { openai: { itemId: "msg_1" } } },
        { type: "text-delta", id: "msg_1", text: "First" },
-        { type: "text-end", id: "msg_1" },
-        { type: "text-start", id: "msg_2" },
+        { type: "text-end", id: "msg_1", providerMetadata: { openai: { itemId: "msg_1" } } },
+        { type: "text-start", id: "msg_2", providerMetadata: { openai: { itemId: "msg_2" } } },
        { type: "text-delta", id: "msg_2", text: "Second" },
-        { type: "text-end", id: "msg_2" },
+        { type: "text-end", id: "msg_2", providerMetadata: { openai: { itemId: "msg_2" } } },
      ])
    }),
  )
@@ -1070,7 +1097,7 @@ describe("OpenAI Responses route", () => {
        { type: "step-start", index: 0 },
        { type: "reasoning-start", id: "rs_1" },
        { type: "reasoning-delta", id: "rs_1", text: "thinking" },
-        { type: "text-start", id: "msg_1" },
+        { type: "text-start", id: "msg_1", providerMetadata: { openai: { itemId: "msg_1" } } },
        { type: "text-delta", id: "msg_1", text: "Hello" },
        { type: "reasoning-end", id: "rs_1" },
        { type: "text-end", id: "msg_1" },
@@ -1080,7 +1107,7 @@ describe("OpenAI Responses route", () => {
      expect(response.events.filter((event) => event.type === "finish")).toHaveLength(1)
      expect(response.message.content).toEqual([
        { type: "reasoning", text: "thinking" },
-        { type: "text", text: "Hello" },
+        { type: "text", text: "Hello", providerMetadata: { openai: { itemId: "msg_1" } } },
      ])
    }),
  )
@@ -1146,33 +1173,34 @@ describe("OpenAI Responses route", () => {
      )

      expect(response.reasoning).toBe("FirstSecond")
-      expect(response.events).toMatchObject([
-        { type: "step-start", index: 0 },
+      expect(response.events.filter((event) => event.type.startsWith("reasoning-"))).toEqual([
        {
          type: "reasoning-start",
          id: "rs_1:0",
          providerMetadata: { openai: { itemId: "rs_1", reasoningEncryptedContent: null } },
        },
-        { type: "reasoning-delta", id: "rs_1:0", text: "First" },
-        { type: "reasoning-end", id: "rs_1:0", providerMetadata: { openai: { itemId: "rs_1" } } },
+        { type: "reasoning-delta", id: "rs_1:0", text: "First", providerMetadata: undefined },
        {
          type: "reasoning-start",
          id: "rs_1:1",
          providerMetadata: { openai: { itemId: "rs_1", reasoningEncryptedContent: null } },
        },
-        { type: "reasoning-delta", id: "rs_1:1", text: "Second" },
+        { type: "reasoning-delta", id: "rs_1:1", text: "Second", providerMetadata: undefined },
+        {
+          type: "reasoning-end",
+          id: "rs_1:0",
+          providerMetadata: { openai: { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" } },
+        },
        {
          type: "reasoning-end",
          id: "rs_1:1",
          providerMetadata: { openai: { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" } },
        },
-        { type: "step-finish", index: 0, reason: { normalized: "stop", raw: undefined } },
-        { type: "finish", reason: { normalized: "stop", raw: undefined } },
      ])
    }),
  )

-  it.effect("closes reasoning summary parts when storage is not disabled", () =>
+  it.effect("preserves complete reasoning metadata when storage is enabled", () =>
    Effect.gen(function* () {
      const response = yield* LLMClient.generate(
        LLMRequest.update(request, { providerOptions: { openai: { store: true } } }),
@@ -1192,7 +1220,7 @@ describe("OpenAI Responses route", () => {
              { type: "response.reasoning_summary_part.done", item_id: "rs_1", summary_index: 1 },
              {
                type: "response.output_item.done",
-                item: { type: "reasoning", id: "rs_1", encrypted_content: null },
+                item: { type: "reasoning", id: "rs_1", encrypted_content: "encrypted-state" },
              },
              { type: "response.completed", response: { id: "resp_1" } },
            ),
@@ -1201,8 +1229,16 @@ describe("OpenAI Responses route", () => {
      )

      expect(response.events.filter((event) => event.type === "reasoning-end")).toEqual([
-        { type: "reasoning-end", id: "rs_1:0", providerMetadata: { openai: { itemId: "rs_1" } } },
-        { type: "reasoning-end", id: "rs_1:1", providerMetadata: { openai: { itemId: "rs_1" } } },
+        {
+          type: "reasoning-end",
+          id: "rs_1:0",
+          providerMetadata: { openai: { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" } },
+        },
+        {
+          type: "reasoning-end",
+          id: "rs_1:1",
+          providerMetadata: { openai: { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" } },
+        },
      ])
    }),
  )
@@ -1250,7 +1286,7 @@ describe("OpenAI Responses route", () => {
                  { role: "user", content: [{ type: "input_text", text: "Summarize it." }] },
                ],
              })
-              expect(body.input[1]).not.toHaveProperty("id")
+              expect(body.input[1]).toHaveProperty("id", "rs_1")
              return input.respond(
                sseEvents(
                  { type: "response.output_text.delta", item_id: "msg_1", delta: "Parser now round-trips reasoning." },
@@ -1267,6 +1303,98 @@ describe("OpenAI Responses route", () => {
    }),
  )

+  it.effect("keeps OpenAI and Azure response item replay independent of store", () =>
+    Effect.gen(function* () {
+      const models = [
+        model,
+        Azure.configure({ resourceName: "opencode-test", apiKey: "azure-key" }).responses("gpt-4.1-mini"),
+      ]
+      const messages = [
+        Message.assistant([
+          {
+            type: "reasoning" as const,
+            text: "Checked the previous diff.",
+            providerMetadata: {
+              openai: { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" },
+            },
+          },
+          ToolCallPart.make({
+            id: "call_1",
+            name: "lookup",
+            input: { query: "weather" },
+            providerMetadata: { openai: { itemId: "fc_1" } },
+          }),
+        ]),
+        Message.tool({ id: "call_1", name: "lookup", result: { forecast: "sunny" } }),
+      ]
+
+      for (const current of models) {
+        const stored = yield* compileRequest(
+          LLM.request({ model: current, messages, providerOptions: { openai: { store: true } } }),
+        )
+        const stateless = yield* compileRequest(
+          LLM.request({ model: current, messages, providerOptions: { openai: { store: false } } }),
+        )
+
+        expect(stored.body.input).toEqual(stateless.body.input)
+        expect(stored.body.input).toEqual([
+          {
+            type: "reasoning",
+            id: "rs_1",
+            summary: [{ type: "summary_text", text: "Checked the previous diff." }],
+            encrypted_content: "encrypted-state",
+          },
+          {
+            type: "function_call",
+            id: "fc_1",
+            call_id: "call_1",
+            name: "lookup",
+            arguments: '{"query":"weather"}',
+            status: "completed",
+          },
+          {
+            type: "function_call_output",
+            id: "fco_1",
+            call_id: "call_1",
+            output: '{"forecast":"sunny"}',
+            status: "completed",
+          },
+        ])
+      }
+    }),
+  )
+
+  it.effect("replaces invalid item ids without creating collisions", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model,
+          messages: [
+            Message.make({
+              id: "msg_text",
+              role: "assistant",
+              content: [
+                { type: "text", text: "Ready.", providerMetadata: { openai: { itemId: "" } } },
+                {
+                  type: "reasoning",
+                  text: "Think.",
+                  providerMetadata: { openai: { itemId: "", reasoningEncryptedContent: "encrypted" } },
+                },
+                ToolCallPart.make({ id: "call_a/b", name: "one", input: {} }),
+                ToolCallPart.make({ id: "call_a?b", name: "two", input: {} }),
+              ],
+            }),
+          ],
+        }),
+      )
+      const ids = prepared.body.input.flatMap((item) => ("id" in item && typeof item.id === "string" ? [item.id] : []))
+
+      expect(ids).toHaveLength(4)
+      expect(new Set(ids).size).toBe(ids.length)
+      expect(ids.every((id) => /^[a-zA-Z0-9]+_.+$/.test(id) && id.length <= 64)).toBe(true)
+    }),
+  )
+
  it.effect("preserves assistant content order around reasoning items", () =>
    Effect.gen(function* () {
      const prepared = yield* compileRequest(
@@ -1274,38 +1402,55 @@ describe("OpenAI Responses route", () => {
          id: "req_reasoning_order",
          model,
          messages: [
-            Message.assistant([
-              { type: "text", text: "Before." },
-              {
-                type: "reasoning",
-                text: "Checked order.",
-                providerMetadata: {
-                  openai: {
-                    itemId: "rs_1",
-                    reasoningEncryptedContent: "encrypted-state",
+            Message.make({
+              id: "msg_assistant",
+              role: "assistant",
+              content: [
+                { type: "text", text: "Before." },
+                {
+                  type: "reasoning",
+                  text: "Checked order.",
+                  providerMetadata: {
+                    openai: {
+                      itemId: "rs_1",
+                      reasoningEncryptedContent: "encrypted-state",
+                    },
                  },
                },
-              },
-              { type: "text", text: "After." },
-            ]),
+                { type: "text", text: "After." },
+              ],
+            }),
          ],
          providerOptions: { openai: { store: false } },
        }),
      )

      expect(prepared.body.input).toEqual([
-        { role: "assistant", content: [{ type: "output_text", text: "Before." }] },
+        {
+          type: "message",
+          id: "msg_assistant",
+          role: "assistant",
+          content: [{ type: "output_text", text: "Before." }],
+          status: "completed",
+        },
        {
          type: "reasoning",
+          id: "rs_1",
          encrypted_content: "encrypted-state",
          summary: [{ type: "summary_text", text: "Checked order." }],
        },
-        { role: "assistant", content: [{ type: "output_text", text: "After." }] },
+        {
+          type: "message",
+          id: "msg_msg_assistant_1",
+          role: "assistant",
+          content: [{ type: "output_text", text: "After." }],
+          status: "completed",
+        },
      ])
    }),
  )

-  it.effect("references stored reasoning items by id", () =>
+  it.effect("replays stored reasoning items with their id", () =>
    Effect.gen(function* () {
      const prepared = yield* compileRequest(
        LLM.request({
@@ -1323,11 +1468,18 @@ describe("OpenAI Responses route", () => {
        }),
      )

-      expect(prepared.body.input).toEqual([{ type: "item_reference", id: "rs_1" }])
+      expect(prepared.body.input).toEqual([
+        {
+          type: "reasoning",
+          id: "rs_1",
+          summary: [{ type: "summary_text", text: "Checked the previous diff." }],
+          encrypted_content: undefined,
+        },
+      ])
    }),
  )

-  it.effect("references stored provider-executed hosted tool results by id", () =>
+  it.effect("replays stored provider-executed hosted tool results", () =>
    Effect.gen(function* () {
      const prepared = yield* compileRequest(
        LLM.request({
@@ -1357,7 +1509,7 @@ describe("OpenAI Responses route", () => {
      )

      expect(prepared.body.input).toEqual([
-        { type: "item_reference", id: "ws_1" },
+        { type: "web_search_call", id: "ws_1", status: "completed" },
        { role: "user", content: [{ type: "input_text", text: "Continue." }] },
      ])
    }),
@@ -1432,6 +1584,7 @@ describe("OpenAI Responses route", () => {
      expect(prepared.body.input).toEqual([
        {
          type: "reasoning",
+          id: "rs_1",
          encrypted_content: "encrypted-state",
          summary: [
            { type: "summary_text", text: "First" },
@@ -1442,7 +1595,7 @@ describe("OpenAI Responses route", () => {
    }),
  )

-  it.effect("skips non-persisted reasoning ids without encrypted state", () =>
+  it.effect("replays reasoning ids without encrypted state", () =>
    Effect.gen(function* () {
      const prepared = yield* compileRequest(
        LLM.request({
@@ -1472,6 +1625,12 @@ describe("OpenAI Responses route", () => {
      expect(prepared.body).toMatchObject({
        input: [
          { role: "user", content: [{ type: "input_text", text: "What changed?" }] },
+          {
+            type: "reasoning",
+            id: "rs_1",
+            summary: [{ type: "summary_text", text: "Checked the previous diff." }],
+            encrypted_content: null,
+          },
          { role: "assistant", content: [{ type: "output_text", text: "The parser changed." }] },
          { role: "user", content: [{ type: "input_text", text: "Summarize it." }] },
        ],
@@ -1663,7 +1822,8 @@ describe("OpenAI Responses route", () => {
          name: "web_search",
          result: { type: "json", value: item },
          providerExecuted: true,
-          providerMetadata: { openai: { itemId: "ws_1" } },
+          output: undefined,
+          providerMetadata: { openai: { itemId: "ws_1", responseItem: item } },
        },
      ])
    }),
@@ -1754,7 +1914,8 @@ describe("OpenAI Responses route", () => {
        name: "code_interpreter",
        result: { type: "json", value: item },
        providerExecuted: true,
-        providerMetadata: { openai: { itemId: "ci_1" } },
+        output: undefined,
+        providerMetadata: { openai: { itemId: "ci_1", responseItem: item } },
      })
    }),
  )
@@ -261,7 +261,7 @@ const assistantMessageFromResponse = (response: LLMResponse, step: AssistantStep
    content.push({ type: "reasoning", text: response.reasoning, providerMetadata: reasoning.providerMetadata })
  }

-  if (response.text.length > 0) content.push({ type: "text", text: response.text })
+  content.push(...response.message.content.filter((part) => part.type === "text"))
  content.push(...response.toolCalls)
  return Message.assistant(content)
 }
@@ -2440,30 +2440,6 @@ describe("SessionRunnerLLM", () => {
    }),
  )

-  it.effect("continues after an unknown finish containing a local tool call", () =>
-    Effect.gen(function* () {
-      const session = yield* setup
-      yield* admit(session, "Echo this")
-      yield* TestLLM.push(
-        TestLLM.complete(
-          { reason: { normalized: "unknown" } },
-          LLMEvent.toolCall({ id: "call-echo", name: "echo", input: { text: "hello" } }),
-        ),
-        TestLLM.text("Done", "text-final"),
-      )
-
-      yield* session.resume(sessionID)
-
-      expect(requests).toHaveLength(2)
-      expect(executions).toEqual(["hello"])
-      expect(yield* session.context(sessionID)).toMatchObject([
-        { type: "user", text: "Echo this" },
-        { type: "assistant", finish: "unknown", content: [{ type: "tool", state: { status: "completed" } }] },
-        { type: "assistant", finish: "stop", content: [{ type: "text", text: "Done" }] },
-      ])
-    }),
-  )
-
  it.effect("reloads a model switch before a tool-driven continuation step", () =>
    Effect.gen(function* () {
      const session = yield* setup
Author	SHA1	Message	Date
Aiden Cline	5483cd6e2b	Merge remote-tracking branch 'origin/v2' into responses-item-replay	2026-08-06 20:41:22 -05:00
Aiden Cline	50f76827bf	fix(ai): harden Responses item replay	2026-08-06 17:45:37 -05:00
Aiden Cline	a619814c79	fix(ai): preserve Responses item IDs	2026-08-06 17:41:39 -05:00