Files
maestro/test/orchestrator.test.ts
T
wangjia 740d2c2637 feat: PR 审核流程 + 自动合并 + macOS 通知 + 订阅额度透传 + 归档详情
PR 流程(开发→提PR→CodeReview→安全审计→审核→通过即合并):
- reviewer 拆分为 code review 与安全审计两个独立 CC run(kind=reviewer/security),
  TaskResult 四字段(summary/verdict + securitySummary/securityVerdict),闸上两节报告两枚结论徽章
- executor/merge.ts: 通过即合并——临时 worktree 内 merge --no-ff,绝不碰用户工作区/不 push;
  冲突安全拒绝(报冲突文件);重复合并幂等;合并后回收执行 worktree + 删分支
- decide 路由 merge:false 逃生口 + 看板「仅通过」按钮(目标分支被工作区检出时用)

通知(daemon/notify.ts):
- macOS 原生通知: 进审核闸(复审建议拒绝标⚠)/连续失败需人工/合并完成
- 同任务同类型 60s 抑制、osascript 转义截断、MAESTRO_NOTIFY=0 关闭

订阅额度透传(daemon/usage.ts):
- OAuth usage API(与 Claude Code/claude-hud 同源),凭证 keychain→内存零泄漏
- 60s 成败双缓存+并发去重+5s 超时,失败降级 null
- GET /api/agents 顶层 usage 字段;Agent 面板显示 5h/周用量条+重置倒计时(>80%琥珀/>95%红)

看板与生命周期:
- 归档区: 深度1整树完成沉底,时间倒序分页(尺寸 chip 10/20/50/100 置底)
- 归档详情对话框: 全属性/执行历史与时长/审批记录/状态流转时间线(GET /api/tasks/:id/events)
- 容器收口: 已拆解 Hard 子任务全 done 自动 done(afterDone 逐级向上)
- 同步按钮收进配置面板;执行白名单扩测试命令(npm/go/shellcheck/make/pytest)
- Agent 面板显示调度模式与各复杂度模型;被依赖阻塞→被阻塞

测试: 74/74(新增 merge 6/notify 10/usage 7/容器收口/双复审适配)

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-13 03:22:30 +08:00

336 lines
14 KiB
TypeScript
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
import { test } from 'node:test';
import assert from 'node:assert/strict';
import { Store } from '../src/store/index.js';
import { createOrchestrator, MAX_RETRIES, type OrchestratorDeps } from '../src/daemon/orchestrator.js';
import type { RunnerResult } from '../src/executor/runner.js';
import type { ReviewResult } from '../src/executor/reviewer.js';
import type { Autonomy } from '../src/model/types.js';
const noopLog = { info: (): void => undefined, error: (): void => undefined };
function deferred<T>(): { promise: Promise<T>; resolve: (v: T) => void } {
let resolve!: (v: T) => void;
const promise = new Promise<T>((r) => { resolve = r; });
return { promise, resolve };
}
const okRun: RunnerResult = { ok: true, transcriptRef: '/tmp/fake.jsonl', sessionId: 'sess-mock-1', finalText: '执行自述:改了 README' };
const okReview: ReviewResult = {
summary: '## 做了什么\nmock 复审通过',
verdict: 'approve',
transcriptRef: '/tmp/fake-review.jsonl',
sessionId: 'sess-review-1',
};
const okSecurity: ReviewResult = {
summary: '## 安全审计\nmock 审计通过',
verdict: 'approve',
transcriptRef: '/tmp/fake-security.jsonl',
sessionId: 'sess-security-1',
};
/** 全 mock 依赖(不真起 CC、不动 git):可按用例覆盖 */
function mockDeps(overrides: Partial<OrchestratorDeps> = {}): OrchestratorDeps {
return {
createWorktree: async (_repo, taskId) => ({ dir: `/tmp/fake-wt/${taskId}`, branch: `maestro/${taskId}` }),
worktreeDiff: async () => ({ diffSummary: ' README.md | 1 +', commits: ['abc1234 hello maestro'] }),
verify: async () => ({ ok: true, exitCode: 0, logRef: null }),
runner: async () => okRun,
reviewCode: async () => okReview,
reviewSecurity: async () => okSecurity,
...overrides,
};
}
function setup(autonomy: Autonomy, concurrency = 1): { store: Store; projectId: string } {
const store = new Store(':memory:');
const p = store.createProject({
name: 'orch', repoPath: '/tmp/orch-repo-' + Math.random(), autonomy, concurrency,
});
return { store, projectId: p.id };
}
async function settle(): Promise<void> {
await new Promise((r) => setImmediate(r));
}
test('autonomy=manual:编排器不领取任何任务', async () => {
const { store, projectId } = setup('manual');
const t = store.createTask({ projectId, title: 'easy task', complexity: 'easy' });
let calls = 0;
const orch = createOrchestrator(store, noopLog, mockDeps({
runner: async () => { calls++; return okRun; },
}));
orch.tick();
await orch.drain();
assert.equal(calls, 0);
assert.equal(store.getTask(t.id)!.status, 'ready');
store.close();
});
test('autonomy=auto-easy:只领 easymedium ready 不动', async () => {
const { store, projectId } = setup('auto-easy', 5);
const easy = store.createTask({ projectId, title: 'small', complexity: 'easy' });
const medium = store.createTask({ projectId, title: 'mid', complexity: 'medium' });
store.setSpec(medium.id, '方案');
store.transition(medium.id, 'spec_review');
store.decide(medium.id, 'accept', 'user'); // medium → ready
assert.equal(store.getTask(medium.id)!.status, 'ready');
const ran: string[] = [];
const orch = createOrchestrator(store, noopLog, mockDeps({
runner: async (task) => { ran.push(task.id); return okRun; },
}));
orch.tick();
await orch.drain();
assert.deepEqual(ran, [easy.id]);
assert.equal(store.getTask(easy.id)!.status, 'exec_review');
assert.equal(store.getTask(medium.id)!.status, 'ready'); // 不自动跑
store.close();
});
test('autonomy=auto-approved:领全部 ready(含 medium),依赖未满足/非叶子不领', async () => {
const { store, projectId } = setup('auto-approved', 5);
const medium = store.createTask({ projectId, title: 'mid', complexity: 'medium' });
store.setSpec(medium.id, '方案');
store.transition(medium.id, 'spec_review');
store.decide(medium.id, 'accept', 'user');
// 依赖未 done 的任务:建在 medium 上 → blocked,不可领
const dep = store.createTask({ projectId, title: 'after-mid', complexity: 'easy', deps: [medium.id] });
assert.equal(store.getTask(dep.id)!.status, 'blocked');
const ran: string[] = [];
const orch = createOrchestrator(store, noopLog, mockDeps({
runner: async (task) => { ran.push(task.id); return okRun; },
}));
orch.tick();
await orch.drain();
assert.deepEqual(ran, [medium.id]);
assert.equal(store.getTask(medium.id)!.status, 'exec_review');
assert.equal(store.getTask(dep.id)!.status, 'blocked');
store.close();
});
test('concurrency=1:同项目同轮只领 1 个,跑完下一轮再领', async () => {
const { store, projectId } = setup('auto-approved', 1);
const t1 = store.createTask({ projectId, title: 'a', complexity: 'easy' });
const t2 = store.createTask({ projectId, title: 'b', complexity: 'easy' });
const gate = deferred<void>();
const started: string[] = [];
const orch = createOrchestrator(store, noopLog, mockDeps({
runner: async (task) => { started.push(task.id); await gate.promise; return okRun; },
}));
orch.tick();
await settle();
assert.deepEqual(started, [t1.id], '并发=1 只应启动第一个任务');
assert.equal(store.getTask(t1.id)!.status, 'executing');
assert.equal(store.getTask(t2.id)!.status, 'ready');
assert.equal(orch.inflight.size, 1);
orch.tick(); // 在途占满 → 本轮不领
await settle();
assert.deepEqual(started, [t1.id]);
gate.resolve();
await orch.drain();
assert.equal(store.getTask(t1.id)!.status, 'exec_review');
orch.tick(); // 槽位释放 → 领第二个
gate.resolve();
await orch.drain();
assert.deepEqual(started, [t1.id, t2.id]);
assert.equal(store.getTask(t2.id)!.status, 'exec_review');
store.close();
});
test('成功路径:状态流转 + setResult(四字段) + executor/reviewer/security 三 run succeeded', async () => {
const { store, projectId } = setup('auto-easy');
const t = store.createTask({ projectId, title: 'tweak', complexity: 'easy' });
store.setOperations(t.id, '在 README.md 追加一行');
const seen: string[] = [];
store.subscribe((e) => { if (e.type === 'status.changed' && e.taskId === t.id) seen.push(String(e.payload.to)); });
let codeReportSeen: string | null = null;
let secReportSeen: string | null = null;
const orch = createOrchestrator(store, noopLog, mockDeps({
reviewCode: async (_task, _project, _wt, _runId, executorReport) => {
codeReportSeen = executorReport;
return okReview;
},
reviewSecurity: async (_task, _project, _wt, _runId, executorReport) => {
secReportSeen = executorReport;
return okSecurity;
},
}));
orch.tick();
await orch.drain();
const done = store.getTask(t.id)!;
assert.equal(done.status, 'exec_review');
assert.deepEqual(seen, ['queued', 'executing', 'exec_review']);
assert.deepEqual(done.result, {
branch: `maestro/${t.id}`,
worktree: `/tmp/fake-wt/${t.id}`,
diffSummary: ' README.md | 1 +',
commits: ['abc1234 hello maestro'],
prUrl: null,
summary: '## 做了什么\nmock 复审通过',
verdict: 'approve',
securitySummary: '## 安全审计\nmock 审计通过',
securityVerdict: 'approve',
});
assert.equal(codeReportSeen, '执行自述:改了 README'); // runner finalText 传给两个复审作执行者自述
assert.equal(secReportSeen, '执行自述:改了 README');
const runs = store.listRuns(t.id);
assert.equal(runs.length, 3);
const executor = runs.find((r) => r.kind === 'executor')!;
assert.equal(executor.status, 'succeeded');
assert.equal(executor.branch, `maestro/${t.id}`);
assert.equal(executor.transcriptRef, '/tmp/fake.jsonl');
assert.equal(executor.claudeSessionId, 'sess-mock-1');
const reviewer = runs.find((r) => r.kind === 'reviewer')!;
assert.equal(reviewer.status, 'succeeded');
assert.equal(reviewer.transcriptRef, '/tmp/fake-review.jsonl');
assert.equal(reviewer.claudeSessionId, 'sess-review-1');
const security = runs.find((r) => r.kind === 'security')!;
assert.equal(security.status, 'succeeded');
assert.equal(security.transcriptRef, '/tmp/fake-security.jsonl');
assert.equal(security.claudeSessionId, 'sess-security-1');
store.close();
});
test('任一 verdict=reject 也照常落进 result(最终裁决仍归用户)', async () => {
const { store, projectId } = setup('auto-easy');
const t = store.createTask({ projectId, title: 'risky', complexity: 'easy' });
const orch = createOrchestrator(store, noopLog, mockDeps({
reviewCode: async () => ({ ...okReview, summary: '发现问题', verdict: 'reject' as const }),
reviewSecurity: async () => ({ ...okSecurity, summary: '发现密钥泄露', verdict: 'reject' as const }),
}));
orch.tick();
await orch.drain();
const done = store.getTask(t.id)!;
assert.equal(done.status, 'exec_review');
assert.equal(done.result!.verdict, 'reject');
assert.equal(done.result!.summary, '发现问题');
assert.equal(done.result!.securityVerdict, 'reject');
assert.equal(done.result!.securitySummary, '发现密钥泄露');
store.close();
});
test('code review 失败不挡任务:summary 记失败原因、verdict=null,安全审计照常跑', async () => {
const { store, projectId } = setup('auto-easy');
const t = store.createTask({ projectId, title: 'review-broken', complexity: 'easy' });
const orch = createOrchestrator(store, noopLog, mockDeps({
reviewCode: async () => { throw new Error('复审 CC 崩了'); },
}));
orch.tick();
await orch.drain();
const done = store.getTask(t.id)!;
assert.equal(done.status, 'exec_review'); // 不挡结果闸
assert.equal(done.result!.verdict, null);
assert.equal(done.result!.summary, '自动复审失败:复审 CC 崩了');
assert.equal(done.result!.securityVerdict, 'approve'); // 另一个复审不受影响
assert.equal(done.result!.securitySummary, '## 安全审计\nmock 审计通过');
const runs = store.listRuns(t.id);
assert.equal(runs.find((r) => r.kind === 'executor')!.status, 'succeeded');
const reviewer = runs.find((r) => r.kind === 'reviewer')!;
assert.equal(reviewer.status, 'failed');
assert.match(reviewer.error ?? '', /复审 CC 崩了/);
assert.equal(runs.find((r) => r.kind === 'security')!.status, 'succeeded');
store.close();
});
test('安全审计失败不挡任务:securitySummary 记失败原因、securityVerdict=nullcode review 不受影响', async () => {
const { store, projectId } = setup('auto-easy');
const t = store.createTask({ projectId, title: 'security-broken', complexity: 'easy' });
const orch = createOrchestrator(store, noopLog, mockDeps({
reviewSecurity: async () => { throw new Error('审计 CC 崩了'); },
}));
orch.tick();
await orch.drain();
const done = store.getTask(t.id)!;
assert.equal(done.status, 'exec_review');
assert.equal(done.result!.verdict, 'approve');
assert.equal(done.result!.summary, '## 做了什么\nmock 复审通过');
assert.equal(done.result!.securityVerdict, null);
assert.equal(done.result!.securitySummary, '自动复审失败:审计 CC 崩了');
const runs = store.listRuns(t.id);
assert.equal(runs.find((r) => r.kind === 'executor')!.status, 'succeeded');
assert.equal(runs.find((r) => r.kind === 'reviewer')!.status, 'succeeded');
const security = runs.find((r) => r.kind === 'security')!;
assert.equal(security.status, 'failed');
assert.match(security.error ?? '', /审计 CC 崩了/);
store.close();
});
test('verify 不过:按失败处理(run failed + 重新入队)', async () => {
const { store, projectId } = setup('auto-easy');
const t = store.createTask({ projectId, title: 'v', complexity: 'easy' });
const orch = createOrchestrator(store, noopLog, mockDeps({
verify: async () => ({ ok: false, exitCode: 1, logRef: '/tmp/v.log', error: 'verify 失败(exit 1' }),
}));
orch.tick();
await orch.drain();
assert.equal(store.getTask(t.id)!.status, 'queued'); // 第一次失败 → 重新入队
const runs = store.listRuns(t.id);
assert.equal(runs.length, 1);
assert.equal(runs[0].status, 'failed');
assert.match(runs[0].error ?? '', /verify 失败/);
store.close();
});
test(`失败重试:重试 ${MAX_RETRIES} 次后 → needs_attention(共 ${MAX_RETRIES + 1} 次失败 run`, async () => {
const { store, projectId } = setup('auto-easy');
const t = store.createTask({ projectId, title: 'flaky', complexity: 'easy' });
let attempts = 0;
const orch = createOrchestrator(store, noopLog, mockDeps({
runner: async () => { attempts++; return { ok: false, transcriptRef: null, sessionId: null, error: `boom #${attempts}` }; },
}));
for (let i = 1; i <= MAX_RETRIES; i++) {
orch.tick();
await orch.drain();
assert.equal(store.getTask(t.id)!.status, 'queued', `${i} 次失败后应重新入队`);
}
orch.tick(); // 最后一次重试也失败
await orch.drain();
assert.equal(store.getTask(t.id)!.status, 'needs_attention');
assert.equal(attempts, MAX_RETRIES + 1);
const failed = store.listRuns(t.id).filter((r) => r.status === 'failed');
assert.equal(failed.length, MAX_RETRIES + 1);
orch.tick(); // needs_attention 不会再被领取
await orch.drain();
assert.equal(attempts, MAX_RETRIES + 1);
store.close();
});
test('project paused:不领取', async () => {
const { store, projectId } = setup('auto-approved');
const t = store.createTask({ projectId, title: 'x', complexity: 'easy' });
store.patchProject(projectId, { status: 'paused' });
let calls = 0;
const orch = createOrchestrator(store, noopLog, mockDeps({
runner: async () => { calls++; return okRun; },
}));
orch.tick();
await orch.drain();
assert.equal(calls, 0);
assert.equal(store.getTask(t.id)!.status, 'ready');
store.close();
});