fix(ai): settle pending Anthropic tool calls

packages/ai/src/protocols/anthropic-messages.ts

@@ -957,9 +957,12 @@ const onMessageDelta = (state: ParserState, event: AnthropicEvent): StepResult =
   ]
 }
 
-const onMessageStop = (state: ParserState): StepResult => {
+const onMessageStop = Effect.fn("AnthropicMessages.onMessageStop")(function* (state: ParserState) {
+  const result = yield* ToolStream.finishAll(ADAPTER, state.tools)
   const events: LLMEvent[] = []
-  const lifecycle = Lifecycle.finish(state.lifecycle, events, {
+  const lifecycle = result.events.length ? Lifecycle.stepStart(state.lifecycle, events) : state.lifecycle
+  events.push(...result.events)
+  const finished = Lifecycle.finish(lifecycle, events, {
     reason: state.pendingFinish?.reason ?? {
       normalized: "unknown",
       raw: undefined,
@@ -967,8 +970,8 @@ const onMessageStop = (state: ParserState): StepResult => {
     usage: state.usage,
     providerMetadata: state.pendingFinish?.providerMetadata,
   })
-  return [{ ...state, lifecycle }, events]
-}
+  return [{ ...state, lifecycle: finished, tools: result.tools }, events] satisfies StepResult
+})
 
 // Prefix `error.type` so overloads, rate limits, and quota errors are visible
 // even when the provider message is generic or empty.
@@ -992,7 +995,7 @@ const step = (state: ParserState, event: AnthropicEvent) => {
   if (event.type === "content_block_delta") return onContentBlockDelta(state, event)
   if (event.type === "content_block_stop") return onContentBlockStop(state, event)
   if (event.type === "message_delta") return Effect.succeed(onMessageDelta(state, event))
-  if (event.type === "message_stop") return Effect.succeed(onMessageStop(state))
+  if (event.type === "message_stop") return onMessageStop(state)
   if (event.type === "error") return onError(event)
   return Effect.succeed<StepResult>([state, NO_EVENTS])
 }

packages/ai/test/provider/anthropic-messages.test.ts

@@ -955,6 +955,37 @@ describe("Anthropic Messages route", () => {
     }),
   )
 
+  it.effect("settles pending tool calls at message_stop", () =>
+    Effect.gen(function* () {
+      const response = yield* LLMClient.generate(request).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              { type: "message_start", message: { usage: { input_tokens: 5 } } },
+              {
+                type: "content_block_start",
+                index: 0,
+                content_block: { type: "tool_use", id: "call_1", name: "lookup" },
+              },
+              {
+                type: "content_block_delta",
+                index: 0,
+                delta: { type: "input_json_delta", partial_json: '{"query":"weather"}' },
+              },
+              { type: "message_delta", delta: { stop_reason: "tool_use" }, usage: { output_tokens: 1 } },
+              { type: "message_stop" },
+            ),
+          ),
+        ),
+      )
+
+      expect(response.toolCalls).toMatchObject([
+        { id: "call_1", name: "lookup", input: { query: "weather" } },
+      ])
+      expect(response.finishReason).toEqual({ normalized: "tool-calls", raw: "tool_use" })
+    }),
+  )
+
   it.effect("assembles and persists multiple tool calls from one Anthropic response", () =>
     Effect.gen(function* () {
       const response = yield* LLMClient.generate(request).pipe(