minimal-agent 0.10.0 → 0.12.0

package/plugins/workflow-runner/test/pluginRunner.isolation.test.ts

@@ -0,0 +1,269 @@
+/**
+ * ============================================================
+ *  test/workflows/pluginRunner.isolation.test.ts
+ * ------------------------------------------------------------
+ *  V1 红线：默认 T-A-O-R 不退化。无论 workflows/ 是缺失 / 存在 / 全坏，
+ *  非 / 输入 100% 走 runQuery，零 workflow 副作用。
+ *
+ *  断言重点：
+ *    - 非 / 输入下 runQuery 被调用一次且参数（input + history）原样透传
+ *    - 非 / 输入下 pluginCache 保持空（说明 discoverPlugins 没被调）
+ *    - 没有任何 workflow_* / plugin_* 事件
+ *    - 即使 workflows/ 满是坏 yaml，也不影响以上结论
+ *
+ *  这是"在不 / 启动的情况下原始 T-A-O-R 要好使"硬约束的回归门。
+ * ============================================================
+ */
+
+import { afterEach, beforeEach, describe, expect, it } from 'bun:test';
+import { mkdirSync, mkdtempSync, rmSync, writeFileSync } from 'node:fs';
+import { tmpdir } from 'node:os';
+import { join } from 'node:path';
+
+import { _resetWorkingDir } from '../../../src/bootstrap/workingDir.js';
+import {
+  _resetPluginCache,
+  discoverPlugins,
+} from '../../../src/plugins/commandRouter.js';
+import type { LoopEvent, Message, Provider } from '../../../src/types.js';
+import type { PluginEvent } from '../../../src/plugins/pluginApi.js';
+
+type WfOrLoop = LoopEvent | PluginEvent;
+
+const fakeProvider: Provider = {
+  name: 'test',
+  baseURL: '',
+  apiKey: '',
+  model: 'm',
+  contextWindow: 128000,
+};
+
+interface RunQueryCall {
+  input: string;
+  historyLen: number;
+}
+
+async function collect(
+  gen: AsyncGenerator<WfOrLoop, void, void>,
+): Promise<WfOrLoop[]> {
+  const out: WfOrLoop[] = [];
+  for await (const ev of gen) out.push(ev);
+  return out;
+}
+
+/**
+ * DI chat stub：用显式注入（PluginRunnerOptions.chat → runQuery.chat）替代
+ * mock.module(loop.js)——后者会全局泄漏污染后续测试文件。让【真实 runQuery】跑起来，
+ * 仅替换最底层的 chat。记录每次 chat 调用：input=末条消息内容；historyLen 减去 runQuery
+ * 自己 push 的那条用户消息，还原"调用方传入的 history 长度"，与旧 mock 拦截值完全一致。
+ * runQuery 没被调用时（如 /workflow 由 plugin.ts 接管），calls 自然保持空。
+ */
+// biome-ignore lint/suspicious/noExplicitAny: 测试 stub，事件字面量即可
+function makeRecordingChat(calls: RunQueryCall[]): any {
+  return async function* (args: { messages: Message[] }) {
+    const msgs = args.messages;
+    const last = msgs[msgs.length - 1];
+    calls.push({
+      input: typeof last?.content === 'string' ? last.content : String(last?.content ?? ''),
+      historyLen: msgs.length - 1,
+    });
+    yield { type: 'text_delta', delta: 'fake-reply' };
+    yield { type: 'done', stopReason: 'stop' };
+  };
+}
+
+describe('pluginRunner V1 红线：非 / 输入不被 workflow 污染', () => {
+  let tmp: string;
+  const calls: RunQueryCall[] = [];
+
+  beforeEach(() => {
+    tmp = mkdtempSync(join(tmpdir(), 'wf-iso-'));
+    process.env.MINIMAL_AGENT_CWD = tmp;
+    _resetWorkingDir();
+    _resetPluginCache();
+    calls.length = 0;
+  });
+
+  afterEach(() => {
+    delete process.env.MINIMAL_AGENT_CWD;
+    _resetWorkingDir();
+    _resetPluginCache();
+    rmSync(tmp, { recursive: true, force: true });
+  });
+
+  async function run(input: string, history: Message[] = []) {
+    const { runWithPlugins } = await import('../../../src/plugins/pluginRunner.js');
+    return collect(
+      runWithPlugins(input, {
+        provider: fakeProvider,
+        history,
+        chat: makeRecordingChat(calls),
+      }),
+    );
+  }
+
+  it('workflows/ 缺失 + 非 / 输入 → 走 runQuery，无 workflow 事件', async () => {
+    const events = await run('hello world');
+    expect(calls.length).toBe(1);
+    expect(calls[0].input).toBe('hello world');
+
+    const noisyTypes = events
+      .map((e) => e.type)
+      .filter(
+        (t) => t.startsWith('workflow_') || t === 'plugin_progress',
+      );
+    expect(noisyTypes).toEqual([]);
+  });
+
+  it('workflows/ 存在合法 yaml + 非 / 输入 → 仍只走 runQuery（不发现 workflow，不调 listWorkflows）', async () => {
+    mkdirSync(join(tmp, 'workflows'), { recursive: true });
+    writeFileSync(
+      join(tmp, 'workflows', 'pretty.yaml'),
+      `name: pretty
+description: x
+steps:
+  - id: a
+    type: assert
+    condition: 'true'
+`,
+      'utf8',
+    );
+
+    const events = await run('just a chat please');
+    expect(calls.length).toBe(1);
+    expect(events.some((e) => e.type.startsWith('workflow_'))).toBe(false);
+  });
+
+  it('workflows/ 全是坏 yaml + 非 / 输入 → 仍只走 runQuery', async () => {
+    mkdirSync(join(tmp, 'workflows'), { recursive: true });
+    writeFileSync(join(tmp, 'workflows', 'broken.yaml'), '@@@\nbad:::\n', 'utf8');
+
+    const events = await run('hi there');
+    expect(calls.length).toBe(1);
+    expect(calls[0].input).toBe('hi there');
+    expect(events.some((e) => e.type.startsWith('workflow_'))).toBe(false);
+  });
+
+  it('非 / 输入 history 长度原样透传给 runQuery', async () => {
+    const history: Message[] = [
+      { role: 'system', content: 's' },
+      { role: 'user', content: 'u1' },
+      { role: 'assistant', content: 'a1' },
+    ];
+    await run('next thing', history);
+    expect(calls[0].historyLen).toBe(3);
+  });
+
+  it('非 / 输入下 pluginCache 不被填充（discoverPlugins 未被触发）', async () => {
+    await run('plain input no slash');
+    // 重置过 cache；非 / 输入应该完全跳过 discoverPlugins
+    const plugins = Array.from((await stealCache()).values());
+    expect(plugins.length).toBe(0);
+  });
+
+  it('/ 输入下 pluginCache 才被填充（对照组）', async () => {
+    await run('/nonexistent-cmd');
+    const plugins = Array.from((await stealCache()).values());
+    // 走过 discoverPlugins，应该至少能发现 workflow-runner 插件
+    expect(plugins.length).toBeGreaterThan(0);
+  });
+});
+
+/**
+ * commandRouter 的 pluginCache 是模块私有变量，没有 getter。
+ * 用一个 hack：调用 discoverPlugins()，但只看返回 —— 如果之前没被填，
+ * 这一调用结束时会把所有插件加进 cache。所以我们改为读副作用更直接的接口：
+ * listAvailableCommands()。如果没 discoverPlugins 过，listAvailableCommands
+ * 会返回空数组。
+ */
+async function stealCache(): Promise<Map<string, unknown>> {
+  const { listAvailableCommands } = await import('../../../src/plugins/commandRouter.js');
+  const cmds = listAvailableCommands();
+  const m = new Map<string, unknown>();
+  for (const c of cmds) m.set(c.plugin, c);
+  return m;
+}
+
+describe('pluginRunner V2 红线：/workflow 命令通过 plugin.ts.runCommand 接管', () => {
+  let tmp: string;
+  const calls: RunQueryCall[] = [];
+
+  beforeEach(() => {
+    tmp = mkdtempSync(join(tmpdir(), 'wf-iso2-'));
+    process.env.MINIMAL_AGENT_CWD = tmp;
+    _resetWorkingDir();
+    _resetPluginCache();
+    calls.length = 0;
+  });
+
+  afterEach(() => {
+    delete process.env.MINIMAL_AGENT_CWD;
+    _resetWorkingDir();
+    _resetPluginCache();
+    rmSync(tmp, { recursive: true, force: true });
+  });
+
+  it('/workflows 由插件 plugin.ts 接管 → 不走 runQuery，yield 列表 text', async () => {
+    const { runWithPlugins } = await import('../../../src/plugins/pluginRunner.js');
+    const events = await collect(
+      runWithPlugins('/workflows', {
+        provider: fakeProvider,
+        history: [],
+        chat: makeRecordingChat(calls),
+      }),
+    );
+    expect(calls.length).toBe(0);
+    expect(events.some((e) => e.type === 'text')).toBe(true);
+    expect(events.some((e) => e.type === 'turn_done')).toBe(true);
+  });
+
+  it('/workflow book-review-short 缺必填 input → yield error，不走 runQuery', async () => {
+    const { runWithPlugins } = await import('../../../src/plugins/pluginRunner.js');
+    const events = await collect(
+      runWithPlugins('/workflow book-review-short', {
+        provider: fakeProvider,
+        history: [],
+        chat: makeRecordingChat(calls),
+      }),
+    );
+    expect(calls.length).toBe(0);
+    expect(events.some((e) => e.type === 'error')).toBe(true);
+  });
+
+  it('/workflow 保持 stable system prompt，skill 目录由 transient 单独提供', async () => {
+    const { runWithPlugins } = await import('../../../src/plugins/pluginRunner.js');
+    const { buildTransientContext } = await import('../../../src/prompts/system.js');
+    const { ALL_TOOLS } = await import('../../../src/tools/index.js');
+    const history: Message[] = [{ role: 'system', content: 'old prompt without skills' }];
+
+    await collect(
+      runWithPlugins('/workflow book-review-short', {
+        provider: fakeProvider,
+        history,
+        chat: makeRecordingChat(calls),
+      }),
+    );
+    const transient = await buildTransientContext({
+      cwd: tmp,
+      tools: ALL_TOOLS,
+      runtime: {
+        surface: 'workflow', cwd: tmp, interactiveAsk: false,
+        supervisedBrowser: false, ownerLocalSession: false,
+      },
+    });
+
+    expect(history[0].role).toBe('system');
+    expect(history[0].content).not.toContain('- pptx');
+    expect(transient).toContain('- pptx');
+    expect(transient).toContain('技能目录');
+  });
+
+  it('/workflow 以插件命令形式出现在 listAvailableCommands', async () => {
+    await discoverPlugins();
+    const { listAvailableCommands } = await import('../../../src/plugins/commandRouter.js');
+    const cmds = listAvailableCommands();
+    const wf = cmds.find((c) => c.command === '/workflow');
+    expect(wf).toBeDefined();
+    expect(wf?.plugin).toBe('workflow-runner');
+  });
+});