boss/tests/master-agent-task-reliability.test.ts

import test from "node:test";
import assert from "node:assert/strict";
import os from "node:os";
import path from "node:path";
import { mkdtemp, rm } from "node:fs/promises";

let runtimeRoot = "";
let data: typeof import("../src/lib/boss-data");

async function setup() {
  if (runtimeRoot) return;
  runtimeRoot = await mkdtemp(path.join(os.tmpdir(), "boss-master-task-reliability-"));
  process.env.BOSS_RUNTIME_ROOT = runtimeRoot;
  process.env.BOSS_STATE_FILE = path.join(runtimeRoot, "boss-state.json");
  data = await import("../src/lib/boss-data.ts");
}

test.after(async () => {
  if (runtimeRoot) await rm(runtimeRoot, { recursive: true, force: true });
});

test.beforeEach(async () => {
  await setup();
  await rm(runtimeRoot, { recursive: true, force: true });
});

async function queueDesktopTask(taskId: string) {
  return data.queueMasterAgentTask({
    taskId,
    projectId: "master-agent",
    taskType: "desktop_control",
    requestMessageId: `${taskId}-message`,
    requestText: "打开 Chrome",
    executionPrompt: "打开 Chrome",
    requestedBy: "krisolo",
    requestedByAccount: "krisolo",
    deviceId: "mac-studio",
    runtimeKind: "computer-use-runtime",
    controlPlatform: "macos",
  });
}

test("claiming a desktop control task records attempt count and a server lease", async () => {
  await queueDesktopTask("lease-task");

  const claimed = await data.claimNextMasterAgentTask("mac-studio");

  assert.equal(claimed?.taskId, "lease-task");
  assert.equal(claimed?.status, "running");
  assert.equal(claimed?.attemptCount, 1);
  assert.ok(claimed?.leaseExpiresAt);
  assert.ok(Date.parse(claimed.leaseExpiresAt) > Date.parse(claimed.claimedAt ?? ""));
});

test("an expired running desktop control task can be reclaimed with a new attempt", async () => {
  await queueDesktopTask("reclaim-task");
  const firstClaim = await data.claimNextMasterAgentTask("mac-studio");
  assert.equal(firstClaim?.attemptCount, 1);

  const state = await data.readState();
  const task = state.masterAgentTasks.find((item) => item.taskId === "reclaim-task");
  assert.ok(task);
  task.claimedAt = "2000-01-01T00:00:00.000Z";
  task.leaseExpiresAt = "2000-01-01T00:01:00.000Z";
  await data.writeState(state);

  const reclaimed = await data.claimNextMasterAgentTask("mac-studio");

  assert.equal(reclaimed?.taskId, "reclaim-task");
  assert.equal(reclaimed?.status, "running");
  assert.equal(reclaimed?.attemptCount, 2);
  assert.notEqual(reclaimed?.claimedAt, firstClaim?.claimedAt);
});

test("an expired running task is timed out after max attempts instead of being claimed forever", async () => {
  await queueDesktopTask("timeout-task");
  await data.claimNextMasterAgentTask("mac-studio");

  const state = await data.readState();
  const task = state.masterAgentTasks.find((item) => item.taskId === "timeout-task");
  assert.ok(task);
  task.status = "running";
  task.attemptCount = 3;
  task.maxAttempts = 3;
  task.leaseExpiresAt = "2000-01-01T00:01:00.000Z";
  await data.writeState(state);

  const claimed = await data.claimNextMasterAgentTask("mac-studio");

  assert.equal(claimed, null);
  const nextTask = (await data.readState()).masterAgentTasks.find((item) => item.taskId === "timeout-task");
  assert.equal(nextTask?.status, "timed_out");
  assert.match(nextTask?.errorMessage ?? "", /timed out/i);
});

test("canceling a running task prevents late success completion from overwriting the terminal state", async () => {
  await queueDesktopTask("cancel-task");
  const claimed = await data.claimNextMasterAgentTask("mac-studio");
  assert.equal(claimed?.status, "running");

  const canceled = await data.cancelMasterAgentTask({
    taskId: "cancel-task",
    actorAccount: "krisolo",
    reason: "用户取消演示任务",
  });
  assert.equal(canceled.status, "canceled");
  assert.ok(canceled.canceledAt);

  const late = await data.completeMasterAgentTask({
    taskId: "cancel-task",
    deviceId: "mac-studio",
    status: "completed",
    replyBody: "迟到的成功结果",
  });

  assert.equal(late.status, "canceled");
  assert.equal(late.replyBody, undefined);
  const finalTask = (await data.readState()).masterAgentTasks.find((item) => item.taskId === "cancel-task");
  assert.equal(finalTask?.status, "canceled");
  assert.equal(finalTask?.replyBody, undefined);
});

test("streaming task progress updates mutate the progress card without completing the task", async () => {
  await queueDesktopTask("live-progress-task");
  const claimed = await data.claimNextMasterAgentTask("mac-studio");
  assert.equal(claimed?.status, "running");

  const updated = await data.updateMasterAgentTaskProgress({
    taskId: "live-progress-task",
    deviceId: "mac-studio",
    status: "running",
    executionProgress: {
      steps: [
        { text: "读取 app-server 事件流", status: "done" },
        { text: "等待目标线程回复", status: "running" },
      ],
      artifacts: [{ label: "codex_app_server_protocol.schemas.json", kind: "file" }],
    },
  });

  assert.equal(updated.status, "running");
  assert.equal(updated.completedAt, undefined);

  const state = await data.readState();
  const progressMessage = state.projects
    .find((project) => project.id === "master-agent")
    ?.messages.find((message) => message.executionProgress?.taskId === "live-progress-task");
  assert.equal(progressMessage?.executionProgress?.status, "running");
  assert.equal(progressMessage?.executionProgress?.steps[0]?.text, "读取 app-server 事件流");
  assert.equal(progressMessage?.executionProgress?.steps[1]?.status, "running");
  assert.equal(progressMessage?.executionProgress?.artifacts?.[0]?.label, "codex_app_server_protocol.schemas.json");
});

test("queued thread collaboration tasks retain source and target thread references", async () => {
  const task = await data.queueMasterAgentTask({
    taskId: "thread-collaboration-task",
    projectId: "master-agent",
    taskType: "conversation_reply",
    requestMessageId: "msg-thread-collaboration",
    requestText: "让源线程和目标线程对一下方案",
    executionPrompt: "让源线程和目标线程对一下方案",
    requestedBy: "krisolo",
    requestedByAccount: "krisolo",
    deviceId: "mac-studio",
    intentCategory: "thread_collaboration",
    sourceThreadId: "source-thread-id",
    sourceThreadDisplayName: "源线程",
    sourceCodexThreadRef: "019d-source-codex",
    targetThreadId: "target-thread-id",
    targetThreadDisplayName: "目标线程",
    targetCodexThreadRef: "019d-target-codex",
  });

  assert.equal(task.intentCategory, "thread_collaboration");
  assert.equal(task.sourceThreadId, "source-thread-id");
  assert.equal(task.sourceThreadDisplayName, "源线程");
  assert.equal(task.sourceCodexThreadRef, "019d-source-codex");

  const claimed = await data.claimNextMasterAgentTask("mac-studio");
  assert.equal(claimed?.sourceCodexThreadRef, "019d-source-codex");
  assert.equal(claimed?.targetCodexThreadRef, "019d-target-codex");
});