fix(drivers): flush assistant text before tools

server/drivers/acp/acp.test.ts

@@ -225,10 +225,10 @@ describe("ACP turns (fake CLI)", () => {
       "turn.started",
       "session.started",
       "content.delta",
+      "item.completed", // assistant_text before the tool, not summed on settle
       "item.started", // tool tc-1
       "item.completed", // tool tc-1 done
       "thread.token-usage.updated",
-      "item.completed", // assistant_text (summed) on settle
       "turn.completed",
     ]);
     expect(recorder.events.every((e) => e.turnId === turnId && e.provider === "grokAgent")).toBe(true);
@@ -241,6 +241,34 @@ describe("ACP turns (fake CLI)", () => {
     expect(instance.adapter.hasSession("t-happy")).toBe(false);
   });
 
+  it("emits each assistant text block before the tool that follows it", async () => {
+    await create(GrokAgentDriver, "interleave");
+    await instance.adapter.sendTurn({ threadId: "t-interleave", text: "go", model: "grok-4.5" });
+    await recorder.until((e) => e.type === "turn.completed");
+
+    const types = recorder.events.map((e) => e.type);
+    expect(types).toEqual([
+      "turn.started",
+      "session.started",
+      "content.delta",
+      "item.completed", // before one
+      "item.started", // tc-1
+      "item.completed", // tc-1
+      "content.delta",
+      "item.completed", // before two
+      "item.started", // tc-2
+      "item.completed", // tc-2
+      "content.delta",
+      "thread.token-usage.updated",
+      "item.completed", // after — no following tool, so settle flushes
+      "turn.completed",
+    ]);
+    const texts = recorder.events
+      .filter((e) => e.type === "item.completed" && (e as { itemType?: string }).itemType === "assistant_text")
+      .map((e) => (e as { text: string }).text);
+    expect(texts).toEqual(["before one", "before two", "after"]);
+  });
+
   it("reads token usage from the root of the prompt result", async () => {
     process.env.FAKE_ACP_USAGE_ROOT = "1";
     await create();

server/drivers/acp/core.ts

@@ -317,6 +317,14 @@ export function createAcpDriver(support: AcpSupport): ProviderDriver<AcpConfig>
 
         const stop = () => killCliTree(child);
 
+        /** Emit buffered assistant text as its own item, then clear it. */
+        const flushAssistantText = () => {
+          const text = state.text;
+          if (!text.trim()) return;
+          state.text = "";
+          emit({ ...base(threadId, turnId), type: "item.completed", itemType: "assistant_text", text });
+        };
+
         const settle = (ok: boolean, stopReason: string | null) => {
           if (state.settled) return;
           state.settled = true;
@@ -328,9 +336,7 @@ export function createAcpDriver(support: AcpSupport): ProviderDriver<AcpConfig>
           }
           rpcPending.clear();
           active.delete(threadId);
-          if (state.text.trim()) {
-            emit({ ...base(threadId, turnId), type: "item.completed", itemType: "assistant_text", text: state.text });
-          }
+          flushAssistantText();
           emit({ ...base(threadId, turnId), type: "turn.completed", ok, stopReason, cost: null });
           stop(); // the agent process does not exit on its own
         };
@@ -342,6 +348,7 @@ export function createAcpDriver(support: AcpSupport): ProviderDriver<AcpConfig>
             return send({ jsonrpc: "2.0", id: msg.id, error: { code: -32601, message: "method not found" } });
           }
           const params = msg.params ?? {};
+          flushAssistantText();
           const options: Array<{ optionId?: string; kind?: string }> = Array.isArray(params.options) ? params.options : [];
           const optionFor = (want: "allow" | "reject") =>
             options.find((o) => String(o.kind ?? "").startsWith(want) && typeof o.optionId === "string")?.optionId ?? null;
@@ -428,6 +435,7 @@ export function createAcpDriver(support: AcpSupport): ProviderDriver<AcpConfig>
               break;
             }
             case "tool_call": {
+              flushAssistantText();
               emit({
                 ...base(threadId, turnId),
                 type: "item.started",

server/drivers/boxagent.test.ts

@@ -0,0 +1,200 @@
+// Box agent contract tests against a scripted fake of ascii.dev's box HTTP
+// API. The driver polls events + prompt status; the fake advances one poll
+// per GET so we can assert message → tool → message order without sleeping.
+import { afterEach, beforeEach, describe, expect, it } from "vitest";
+
+import { ensureDirs } from "../config.ts";
+import type { ProviderInstance } from "../contracts.ts";
+import { recordEvents, type EventRecorder } from "../testing/events.ts";
+import { BoxAgentDriver } from "./boxagent.ts";
+
+const BOX = "box-1";
+const PROMPT = "p1";
+
+/** JSON Response helper for the in-process Box HTTP fake. */
+function json(body: unknown, status = 200) {
+  return new Response(JSON.stringify(body), { status, headers: { "content-type": "application/json" } });
+}
+
+type Poll = { events: unknown[]; status?: { promptRun: { status: string; result?: string } } };
+
+/** Stub fetch so each GET /events + /prompts pair advances one poll in `script`. */
+function installFakeBox(script: Poll[]) {
+  let i = 0;
+  const previous = globalThis.fetch;
+  globalThis.fetch = (async (input: string | URL, init?: RequestInit) => {
+    const url = String(input);
+    const method = String(init?.method ?? "GET").toUpperCase();
+    if (url.endsWith("/me")) return json({ ok: true });
+    if (method === "POST" && /\/boxes\/[^/]+\/prompt$/.test(url)) return json({ promptRun: { id: PROMPT } });
+    if (method === "POST" && url.includes("/interrupt")) return json({ ok: true });
+    if (url.includes("/events")) {
+      const step = script[Math.min(i, script.length - 1)]!;
+      i += 1;
+      return json({ events: step.events });
+    }
+    if (url.includes(`/prompts/${PROMPT}`)) {
+      const step = script[Math.min(Math.max(i - 1, 0), script.length - 1)]!;
+      return json(step.status ?? { promptRun: { status: "running" } });
+    }
+    return json({ error: `unexpected ${method} ${url}` }, 404);
+  }) as typeof fetch;
+  return () => {
+    globalThis.fetch = previous;
+  };
+}
+
+const computer = { boxId: BOX, token: "box-test-token" };
+
+describe("BoxAgentDriver turns (fake API)", () => {
+  let instance: ProviderInstance;
+  let recorder: EventRecorder;
+  let restoreFetch: (() => void) | undefined;
+
+  const create = async () => {
+    instance = await BoxAgentDriver.create({
+      instanceId: "box-test",
+      displayName: "Box Test",
+      environment: { BOX_TOKEN: "box-test-token" },
+      enabled: true,
+      config: { pollMs: 0 },
+    });
+    recorder = recordEvents(instance.adapter);
+  };
+
+  beforeEach(() => {
+    ensureDirs();
+  });
+
+  afterEach(async () => {
+    recorder?.stop();
+    await instance?.dispose();
+    restoreFetch?.();
+    restoreFetch = undefined;
+  });
+
+  it("flushes prefix-grown text before a tool, then the tail at settle", async () => {
+    restoreFetch = installFakeBox([
+      {
+        events: [{ id: "e1", type: "response", text: "hel" }],
+        status: { promptRun: { status: "running" } },
+      },
+      {
+        events: [
+          { id: "e1", type: "response", text: "hel" },
+          { id: "e2", type: "tool", title: "run" },
+        ],
+        status: { promptRun: { status: "running" } },
+      },
+      {
+        events: [
+          { id: "e1", type: "response", text: "hel" },
+          { id: "e2", type: "tool", title: "run" },
+          { id: "e3", type: "response", text: "hello there" },
+        ],
+        status: { promptRun: { status: "finished", result: "hello there" } },
+      },
+    ]);
+    await create();
+    await instance.adapter.sendTurn({ threadId: "t-prefix", text: "go", integrations: { computer } });
+    await recorder.until((e) => e.type === "turn.completed");
+
+    const texts = recorder.events
+      .filter((e) => e.type === "item.completed" && (e as { itemType: string }).itemType === "assistant_text")
+      .map((e) => (e as { text: string }).text);
+    expect(texts).toEqual(["hel", "lo there"]);
+  });
+
+  it("keeps a non-prefix response after a flush instead of slicing it away", async () => {
+    restoreFetch = installFakeBox([
+      {
+        events: [{ id: "e1", type: "response", text: "before" }],
+        status: { promptRun: { status: "running" } },
+      },
+      {
+        events: [
+          { id: "e1", type: "response", text: "before" },
+          { id: "e2", type: "tool", title: "run" },
+        ],
+        status: { promptRun: { status: "running" } },
+      },
+      {
+        events: [
+          { id: "e1", type: "response", text: "before" },
+          { id: "e2", type: "tool", title: "run" },
+          { id: "e3", type: "response", text: "after" },
+        ],
+        status: { promptRun: { status: "finished", result: "after" } },
+      },
+    ]);
+    await create();
+    await instance.adapter.sendTurn({ threadId: "t-nonprefix", text: "go", integrations: { computer } });
+    await recorder.until((e) => e.type === "turn.completed");
+
+    const types = recorder.events.map((e) => e.type);
+    expect(types).toEqual([
+      "turn.started",
+      "session.started",
+      "content.delta",
+      "item.completed", // before
+      "item.started",
+      "content.delta",
+      "item.completed", // after — must not be sliced to ""
+      "turn.completed",
+    ]);
+    const texts = recorder.events
+      .filter((e) => e.type === "item.completed" && (e as { itemType: string }).itemType === "assistant_text")
+      .map((e) => (e as { text: string }).text);
+    expect(texts).toEqual(["before", "after"]);
+  });
+
+  it("ingests a non-prefix prompt result when events already set lastText", async () => {
+    restoreFetch = installFakeBox([
+      {
+        events: [{ id: "e1", type: "response", text: "before" }],
+        status: { promptRun: { status: "running" } },
+      },
+      {
+        events: [
+          { id: "e1", type: "response", text: "before" },
+          { id: "e2", type: "tool", title: "run" },
+        ],
+        status: { promptRun: { status: "running" } },
+      },
+      {
+        events: [
+          { id: "e1", type: "response", text: "before" },
+          { id: "e2", type: "tool", title: "run" },
+        ],
+        status: { promptRun: { status: "finished", result: "done" } },
+      },
+    ]);
+    await create();
+    await instance.adapter.sendTurn({ threadId: "t-status", text: "go", integrations: { computer } });
+    await recorder.until((e) => e.type === "turn.completed");
+
+    const texts = recorder.events
+      .filter((e) => e.type === "item.completed" && (e as { itemType: string }).itemType === "assistant_text")
+      .map((e) => (e as { text: string }).text);
+    expect(texts).toEqual(["before", "done"]);
+  });
+
+  it("flushes pending assistant text when the turn is interrupted", async () => {
+    restoreFetch = installFakeBox([
+      {
+        events: [{ id: "e1", type: "response", text: "half" }],
+        status: { promptRun: { status: "running" } },
+      },
+    ]);
+    await create();
+    await instance.adapter.sendTurn({ threadId: "t-cancel", text: "go", integrations: { computer } });
+    await recorder.until((e) => e.type === "content.delta");
+    await instance.adapter.interruptTurn("t-cancel");
+    const done = await recorder.until((e) => e.type === "turn.completed");
+    expect(done).toMatchObject({ ok: false, stopReason: "interrupted" });
+    const texts = recorder.events
+      .filter((e) => e.type === "item.completed" && (e as { itemType: string }).itemType === "assistant_text")
+      .map((e) => (e as { text: string }).text);
+    expect(texts).toEqual(["half"]);
+  });
+});

server/drivers/boxagent.ts

@@ -129,6 +129,22 @@ export const BoxAgentDriver: ProviderDriver<BoxAgentConfig> = {
         const seen = new Set<string>();
         const startedAt = Date.now();
         let lastText = "";
+        let pendingText = "";
+        /** Emit unflushed deltas as assistant_text and reset pendingText. */
+        const flushAssistantText = () => {
+          const text = pendingText;
+          pendingText = "";
+          if (!text.trim()) return;
+          emit({ ...base(threadId, turnId), type: "item.completed", itemType: "assistant_text", text });
+        };
+        /** Stream a full-text snapshot as a delta and accumulate it for flush. */
+        const ingest = (text: string) => {
+          const delta = text.startsWith(lastText) ? text.slice(lastText.length) : text;
+          lastText = text;
+          if (!delta) return;
+          pendingText += delta;
+          emit({ ...base(threadId, turnId), type: "content.delta", streamKind: "assistant_text", delta });
+        };
         try {
           for (;;) {
             if (cancelled) break;
@@ -148,12 +164,9 @@ export const BoxAgentDriver: ProviderDriver<BoxAgentConfig> = {
               // stream anyway.
               const text = ev.text ?? ev.message ?? ev.data?.text ?? ev.data?.content ?? null;
               if (/assistant|message|output|response/i.test(kind) && typeof text === "string" && text.trim()) {
-                const delta = text.startsWith(lastText) ? text.slice(lastText.length) : text;
-                lastText = text;
-                if (delta) {
-                  emit({ ...base(threadId, turnId), type: "content.delta", streamKind: "assistant_text", delta });
-                }
+                ingest(text);
               } else if (/tool|command|exec|browse/i.test(kind)) {
+                flushAssistantText();
                 emit({
                   ...base(threadId, turnId),
                   type: "item.started",
@@ -167,9 +180,7 @@ export const BoxAgentDriver: ProviderDriver<BoxAgentConfig> = {
               // events themselves instead of hanging to the 30-min ceiling
               if (!promptId && /complete|finish|done|success|fail|error/i.test(kind)) {
                 active.delete(threadId);
-                if (lastText) {
-                  emit({ ...base(threadId, turnId), type: "item.completed", itemType: "assistant_text", text: lastText });
-                }
+                flushAssistantText();
                 const failed = /fail|error/i.test(kind);
                 emit({ ...base(threadId, turnId), type: "turn.completed", ok: !failed, stopReason: failed ? kind : null, cost: null });
                 return;
@@ -184,30 +195,17 @@ export const BoxAgentDriver: ProviderDriver<BoxAgentConfig> = {
               const state = String(run?.status ?? "");
               if (/completed|succeeded|done|finished/i.test(state)) {
                 const result = run?.result ?? run?.output ?? lastText;
-                // stream only the growth past what events already sent —
-                // the settled message below carries the full text regardless
-                if (typeof result === "string" && result.trim() && result !== lastText && result.startsWith(lastText)) {
-                  emit({
-                    ...base(threadId, turnId),
-                    type: "content.delta",
-                    streamKind: "assistant_text",
-                    delta: result.slice(lastText.length),
-                  });
+                if (typeof result === "string" && result.trim() && result !== lastText) {
+                  ingest(result);
                 }
-                emit({
-                  ...base(threadId, turnId),
-                  type: "item.completed",
-                  itemType: "assistant_text",
-                  text: typeof result === "string" && result.trim() ? result : lastText || "(finished)",
-                });
+                if (!pendingText.trim() && !lastText.trim()) pendingText = "(finished)";
+                flushAssistantText();
                 active.delete(threadId);
                 emit({ ...base(threadId, turnId), type: "turn.completed", ok: true, stopReason: null, cost: null });
                 return;
               }
               if (/failed|error|cancelled|interrupted/i.test(state)) {
-                if (lastText) {
-                  emit({ ...base(threadId, turnId), type: "item.completed", itemType: "assistant_text", text: lastText });
-                }
+                flushAssistantText();
                 active.delete(threadId);
                 emit({ ...base(threadId, turnId), type: "turn.completed", ok: false, stopReason: state, cost: null });
                 return;
@@ -218,9 +216,11 @@ export const BoxAgentDriver: ProviderDriver<BoxAgentConfig> = {
             }
           }
           // cancelled
+          flushAssistantText();
           active.delete(threadId);
           emit({ ...base(threadId, turnId), type: "turn.completed", ok: false, stopReason: "interrupted", cost: null });
         } catch (e) {
+          flushAssistantText();
           active.delete(threadId);
           emit({ ...base(threadId, turnId), type: "runtime.error", message: (e as Error).message });
           emit({ ...base(threadId, turnId), type: "turn.completed", ok: false, stopReason: "error", cost: null });