macro-agent 0.2.5 → 0.2.6

package/src/__tests__/e2e/resume-continue.e2e.test.ts

@@ -1,312 +0,0 @@
-/**
- * Resume & Continue E2E Tests (V2)
- *
- * Tests that verify agent resume (session restoration) and continue
- * (new agent with same role) work with REAL Claude Code agents.
- *
- * REQUIRES: RUN_FULL_AGENT_TESTS=true
- *
- * Run with:
- *   RUN_FULL_AGENT_TESTS=true npx vitest run --config vitest.e2e.config.ts src/__tests__/e2e/resume-continue.e2e.test.ts
- */
-
-import {
-  describe,
-  it,
-  expect,
-  beforeEach,
-  afterEach,
-} from "vitest";
-import * as path from "path";
-import * as os from "os";
-import * as fs from "fs";
-import { execSync } from "child_process";
-import { bootV2, type MacroAgentSystemV2 } from "../../boot-v2.js";
-
-// ─────────────────────────────────────────────────────────────────
-// Configuration
-// ─────────────────────────────────────────────────────────────────
-
-const RUN_FULL_AGENT = !!process.env.RUN_FULL_AGENT_TESTS;
-const describeFn = RUN_FULL_AGENT ? describe : describe.skip;
-
-const TIMEOUT = {
-  SPAWN: 60_000,
-  PROMPT: 90_000,
-  MULTI: 180_000,
-};
-
-// ─────────────────────────────────────────────────────────────────
-// Helpers
-// ─────────────────────────────────────────────────────────────────
-
-function createTestRepo(prefix: string): { path: string; cleanup: () => void } {
-  const tmpDir = fs.mkdtempSync(
-    path.join(os.tmpdir(), `resume-continue-${prefix}-`)
-  );
-  const repoPath = path.join(tmpDir, "test-repo");
-  fs.mkdirSync(repoPath);
-  execSync("git init", { cwd: repoPath, stdio: "pipe" });
-  execSync('git config user.email "test@test.com"', { cwd: repoPath, stdio: "pipe" });
-  execSync('git config user.name "Test User"', { cwd: repoPath, stdio: "pipe" });
-  fs.writeFileSync(path.join(repoPath, "README.md"), "# Test Repo\n");
-  execSync("git add -A", { cwd: repoPath, stdio: "pipe" });
-  execSync('git commit -m "Initial commit"', { cwd: repoPath, stdio: "pipe" });
-
-  return {
-    path: repoPath,
-    cleanup: () => fs.rmSync(tmpDir, { recursive: true, force: true }),
-  };
-}
-
-function log(msg: string): void {
-  console.log(`[RESUME-E2E] ${msg}`);
-}
-
-function getTextContent(updates: any[]): string {
-  const parts: string[] = [];
-  for (const u of updates) {
-    const uAny = u as any;
-    if (uAny.sessionUpdate === "agent_message_chunk" && uAny.content?.text) {
-      parts.push(uAny.content.text);
-    }
-  }
-  return parts.join("");
-}
-
-// ─────────────────────────────────────────────────────────────────
-// Tests
-// ─────────────────────────────────────────────────────────────────
-
-describeFn("Resume & Continue E2E (V2)", () => {
-  let system: MacroAgentSystemV2;
-  let testRepo: { path: string; cleanup: () => void };
-  let baseDir: string;
-
-  beforeEach(async () => {
-    testRepo = createTestRepo("v2");
-    baseDir = path.join(testRepo.path, ".macro-agent");
-    fs.mkdirSync(baseDir, { recursive: true });
-
-    system = await bootV2({
-      cwd: testRepo.path,
-      baseDir,
-      defaultPermissionMode: "auto-approve",
-      inbox: {
-        socketPath: path.join(baseDir, "inbox.sock"),
-      },
-    });
-    log("System booted");
-  });
-
-  afterEach(async () => {
-    if (system) {
-      try {
-        const running = system.agentManager.list({ state: "running" });
-        for (const agent of running) {
-          try {
-            await system.agentManager.terminate(agent.id, "cancelled");
-          } catch {
-            // Best effort
-          }
-        }
-        await system.shutdown();
-      } catch {
-        // Best effort
-      }
-    }
-    testRepo?.cleanup();
-    log("Cleanup complete");
-  });
-
-  // ── Resume restores session ──────────────────────────────────
-
-  it(
-    "should resume an agent and restore its session memory",
-    async () => {
-      log("Spawning agent...");
-      const agent = await system.agentManager.spawn({
-        task: "You are a memory test agent. Remember any words you are told.",
-        role: "worker",
-        cwd: testRepo.path,
-      });
-
-      log(`Agent spawned: ${agent.id}`);
-      expect(agent.id).toBeDefined();
-
-      // Give the agent a fact to remember
-      log("Prompting agent to remember BANANA...");
-      const rememberUpdates: any[] = [];
-      for await (const update of system.agentManager.prompt(
-        agent.id,
-        'Remember the word BANANA. Reply with "I will remember BANANA." and nothing else.'
-      )) {
-        rememberUpdates.push(update);
-      }
-      log(`Got ${rememberUpdates.length} updates from remember prompt`);
-
-      const rememberText = getTextContent(rememberUpdates);
-      log(`Remember response: ${rememberText.slice(0, 200)}`);
-      expect(rememberText.toUpperCase()).toContain("BANANA");
-
-      // Terminate the agent
-      log("Terminating agent...");
-      await system.agentManager.terminate(agent.id, "completed");
-      expect(system.agentStore.getAgent(agent.id)!.state).toBe("stopped");
-      expect(system.agentManager.hasActiveSession(agent.id)).toBe(false);
-
-      // Resume the agent
-      log("Resuming agent...");
-      const resumed = await system.agentManager.resume(agent.id);
-      log(`Resumed agent: ${resumed.id}`);
-      expect(resumed.id).toBe(agent.id);
-      expect(system.agentManager.hasActiveSession(agent.id)).toBe(true);
-
-      // Ask the resumed agent what the word was
-      log("Prompting resumed agent for the remembered word...");
-      const recallUpdates: any[] = [];
-      for await (const update of system.agentManager.prompt(
-        agent.id,
-        "What was the word I asked you to remember? Reply with just the word."
-      )) {
-        recallUpdates.push(update);
-      }
-
-      const recallText = getTextContent(recallUpdates);
-      log(`Recall response: ${recallText.slice(0, 200)}`);
-      expect(recallText.toUpperCase()).toContain("BANANA");
-
-      log("Resume session memory verified!");
-    },
-    TIMEOUT.MULTI
-  );
-
-  // ── Resume preserves conversation history ─────────────────────
-
-  it(
-    "should preserve conversation history across resume",
-    async () => {
-      log("Spawning agent...");
-      const agent = await system.agentManager.spawn({
-        task: "You are a fact-tracking agent. Store and recall facts precisely.",
-        role: "worker",
-        cwd: testRepo.path,
-      });
-
-      log(`Agent spawned: ${agent.id}`);
-
-      // Give the agent a specific fact
-      log("Giving agent a fact...");
-      const factUpdates: any[] = [];
-      for await (const update of system.agentManager.prompt(
-        agent.id,
-        'The project code is ALPHA-7. Please confirm you have noted this by replying "Noted: ALPHA-7".'
-      )) {
-        factUpdates.push(update);
-      }
-
-      const factText = getTextContent(factUpdates);
-      log(`Fact response: ${factText.slice(0, 200)}`);
-      expect(factText.toUpperCase()).toContain("ALPHA-7");
-
-      // Terminate
-      log("Terminating agent...");
-      await system.agentManager.terminate(agent.id, "completed");
-      expect(system.agentStore.getAgent(agent.id)!.state).toBe("stopped");
-
-      // Resume
-      log("Resuming agent...");
-      const resumed = await system.agentManager.resume(agent.id);
-      expect(resumed.id).toBe(agent.id);
-
-      // Ask for the fact
-      log("Asking resumed agent for the project code...");
-      const recallUpdates: any[] = [];
-      for await (const update of system.agentManager.prompt(
-        agent.id,
-        "What is the project code I told you earlier? Reply with just the code."
-      )) {
-        recallUpdates.push(update);
-      }
-
-      const recallText = getTextContent(recallUpdates);
-      log(`Recall response: ${recallText.slice(0, 200)}`);
-      expect(recallText.toUpperCase()).toContain("ALPHA-7");
-
-      log("Conversation history preserved across resume!");
-    },
-    TIMEOUT.MULTI
-  );
-
-  // ── Continue creates new agent ────────────────────────────────
-
-  it(
-    "should continue a terminated agent as a new agent with same role",
-    async () => {
-      log("Spawning original worker...");
-      const original = await system.agentManager.spawn({
-        task: "You are a worker agent performing calculations.",
-        role: "worker",
-        cwd: testRepo.path,
-      });
-
-      log(`Original: ${original.id}`);
-
-      // Give it some context
-      log("Prompting original with context...");
-      const contextUpdates: any[] = [];
-      for await (const update of system.agentManager.prompt(
-        original.id,
-        "Compute 10 + 20. Reply with the answer."
-      )) {
-        contextUpdates.push(update);
-      }
-
-      const contextText = getTextContent(contextUpdates);
-      log(`Context response: ${contextText.slice(0, 200)}`);
-
-      // Terminate
-      log("Terminating original...");
-      await system.agentManager.terminate(original.id, "completed");
-      expect(system.agentStore.getAgent(original.id)!.state).toBe("stopped");
-
-      // Continue
-      log("Continuing agent...");
-      const continued = await system.agentManager.continueAgent(original.id, {
-        additionalContext: "Previous agent computed 10+20=30. Continue from there.",
-        task: "Continue previous calculations.",
-      });
-
-      log(`Continued: ${continued.id}`);
-      expect(continued.id).not.toBe(original.id);
-      expect(continued.agent.state).toBe("running");
-      expect(continued.agent.role).toBe("worker");
-
-      // Verify in store
-      const record = system.agentStore.getAgent(continued.id)!;
-      expect(record.state).toBe("running");
-      expect(record.role).toBe("worker");
-
-      // Verify in inbox
-      const inbox = (system.inboxAdapter as any).getInbox();
-      expect(inbox.storage.getAgent(continued.id)).toBeDefined();
-
-      // Prompt the continuation to verify it works
-      log("Prompting continuation...");
-      const contUpdates: any[] = [];
-      for await (const update of system.agentManager.prompt(
-        continued.id,
-        "What is 30 + 15? Reply briefly."
-      )) {
-        contUpdates.push(update);
-      }
-
-      const contText = getTextContent(contUpdates);
-      log(`Continuation response: ${contText.slice(0, 200)}`);
-      expect(contUpdates.length).toBeGreaterThan(0);
-
-      log("Continue created new agent with same role!");
-    },
-    TIMEOUT.MULTI
-  );
-});

package/src/__tests__/e2e/self-driving-v3.e2e.test.ts

@@ -1,197 +0,0 @@
-/**
- * Self-driving team V3 end-to-end test.
- *
- * Exercises the complete path from a real team YAML on disk:
- *   .multiagent/teams/self-driving/team.yaml
- *   → TeamManagerV2.startTeam("self-driving")
- *   → extracts macro_agent.workspace block
- *   → constructs YamlDrivenTopology, installs on AgentManager
- *   → onTeamStart creates the team root stream
- *   → agents spawn with the correct V3 WorkspaceDecision per role
- *
- * Uses mocked acp-factory to avoid spawning real Claude Code — the purpose
- * here is verifying the wiring, not agent behavior.
- *
- * REQUIRES: RUN_E2E_TESTS=true
- */
-
-import {
-  describe,
-  it,
-  expect,
-  beforeEach,
-  afterEach,
-  vi,
-} from 'vitest';
-import * as fs from 'fs';
-import * as os from 'os';
-import * as path from 'path';
-import { execSync } from 'child_process';
-import { bootV2, type MacroAgentSystemV2 } from '../../boot-v2.js';
-import { GitCascadeAdapter } from '../../workspace/git-cascade-adapter.js';
-import {
-  DefaultWorkspaceManager,
-  createWorkspaceManagerWithAdapter,
-} from '../../workspace/workspace-manager.js';
-import { TeamManagerV2 } from '../../teams/team-manager-v2.js';
-
-const RUN_E2E = !!process.env.RUN_E2E_TESTS;
-const describeFn = RUN_E2E ? describe : describe.skip;
-
-// Mock acp-factory so we don't spawn real Claude Code
-vi.mock('acp-factory', () => ({
-  AgentFactory: {
-    spawn: vi.fn().mockResolvedValue({
-      createSession: vi.fn().mockResolvedValue({
-        id: `session-${Date.now()}`,
-        prompt: vi.fn().mockReturnValue({
-          [Symbol.asyncIterator]: () => ({
-            next: () => Promise.resolve({ done: true, value: undefined }),
-          }),
-        }),
-        forkWithFlush: vi.fn().mockResolvedValue({ id: `forked-${Date.now()}` }),
-      }),
-      loadSession: vi.fn().mockResolvedValue({ id: `loaded-${Date.now()}` }),
-      close: vi.fn().mockResolvedValue(undefined),
-      isRunning: vi.fn().mockReturnValue(true),
-    }),
-  },
-}));
-
-vi.mock('opentasks', () => ({
-  OpenTasksClient: vi.fn().mockImplementation(() => ({
-    connect: vi.fn().mockRejectedValue(new Error('No daemon')),
-    disconnect: vi.fn(),
-    query: vi.fn().mockResolvedValue({ items: [] }),
-    link: vi.fn().mockResolvedValue({ success: true }),
-    task: vi.fn().mockResolvedValue({ id: 't-1' }),
-  })),
-}));
-
-describeFn('Self-driving team V3 auto-wire', () => {
-  let system: MacroAgentSystemV2;
-  let testDir: string;
-  let repoPath: string;
-  let adapter: GitCascadeAdapter;
-  let workspaceManager: DefaultWorkspaceManager;
-  let teamManager: TeamManagerV2;
-  const projectRoot = process.cwd();
-
-  beforeEach(async () => {
-    testDir = fs.mkdtempSync(path.join(os.tmpdir(), 'self-driving-e2e-'));
-    repoPath = path.join(testDir, 'repo');
-    fs.mkdirSync(repoPath);
-
-    execSync('git init -b main', { cwd: repoPath, stdio: 'pipe' });
-    execSync('git config user.email "t@t.com"', { cwd: repoPath, stdio: 'pipe' });
-    execSync('git config user.name "T"', { cwd: repoPath, stdio: 'pipe' });
-    fs.writeFileSync(path.join(repoPath, 'README.md'), '# test');
-    execSync('git add .', { cwd: repoPath, stdio: 'pipe' });
-    execSync('git commit -m "init"', { cwd: repoPath, stdio: 'pipe' });
-
-    adapter = new GitCascadeAdapter({
-      enabled: true,
-      repoPath,
-      dbPath: path.join(testDir, 'gc.db'),
-      skipRecovery: true,
-    });
-    workspaceManager = createWorkspaceManagerWithAdapter(adapter, {
-      worktreeBaseDir: path.join(repoPath, '.worktrees'),
-    }) as DefaultWorkspaceManager;
-
-    system = await bootV2({
-      cwd: repoPath,
-      baseDir: testDir,
-      inbox: { socketPath: path.join(testDir, 'inbox.sock') },
-      workspaceManager,
-    });
-
-    teamManager = new TeamManagerV2({
-      agentManager: system.agentManager,
-      inboxAdapter: system.inboxAdapter,
-      tasksAdapter: system.tasksAdapter,
-      workspaceManager,
-    });
-    teamManager.install();
-  });
-
-  afterEach(async () => {
-    if (system) await system.shutdown();
-    if (workspaceManager) workspaceManager.close();
-    if (adapter) adapter.close();
-    if (fs.existsSync(testDir)) fs.rmSync(testDir, { recursive: true, force: true });
-  });
-
-  it('startTeam auto-wires YamlDrivenTopology from macro_agent.workspace', async () => {
-    // Start the real self-driving team template from .multiagent/teams/
-    await teamManager.startTeam('self-driving', projectRoot);
-
-    // Team root stream should have been created by onTeamStart
-    const streams = workspaceManager.listStreams();
-    const teamRoot = streams.find((s) => s.agentId === 'team:self-driving');
-    expect(teamRoot).toBeDefined();
-    expect(teamRoot!.name).toBe('self-driving');
-  });
-
-  it('spawning a grinder via the V3 path forks a stream off team root', async () => {
-    await teamManager.startTeam('self-driving', projectRoot);
-
-    const teamStreams = workspaceManager.listStreams();
-    const teamRoot = teamStreams.find((s) => s.agentId === 'team:self-driving');
-    expect(teamRoot).toBeDefined();
-
-    const grinder = await system.agentManager.spawn({
-      role: 'grinder',
-      task: 'do a thing',
-    });
-
-    const record = system.agentStore.getAgent(grinder.id);
-    expect(record?.workspace_path).toBeDefined();
-    expect(fs.existsSync(record!.workspace_path!)).toBe(true);
-
-    const grinderStream = workspaceManager
-      .listStreams()
-      .find((s) => s.agentId === grinder.id);
-    expect(grinderStream).toBeDefined();
-    expect(grinderStream?.parentStream).toBe(teamRoot!.id);
-  });
-
-  it('judge spawn returns workspace: none (no worktree allocated)', async () => {
-    await teamManager.startTeam('self-driving', projectRoot);
-
-    const judge = await system.agentManager.spawn({
-      role: 'judge',
-      task: 'review',
-    });
-
-    const record = system.agentStore.getAgent(judge.id);
-    // workspace: none → no workspace_path assigned
-    expect(record?.workspace_path).toBeFalsy();
-  });
-
-  it('planner attaches to team root stream (not a new fork)', async () => {
-    // Note: self-driving's topology already bootstraps the planner as the
-    // root agent; we spawn explicitly to verify the decision path.
-    await teamManager.startTeam('self-driving', projectRoot);
-
-    const plannerBeforeBootstrap = system.agentStore.listAgents({
-      state: 'running',
-    });
-
-    // The bootstrap planner is the "root" — already spawned.
-    const rootPlanner = plannerBeforeBootstrap.find((a) => a.role === 'planner');
-    expect(rootPlanner).toBeDefined();
-
-    // Verify its workspace_path (if allocated) is on the team root branch
-    if (rootPlanner?.workspace_path) {
-      expect(fs.existsSync(rootPlanner.workspace_path)).toBe(true);
-    }
-
-    // The team_root stream should still be the only top-level stream
-    // (planner did not fork a new stream — it attached)
-    const topLevelStreams = workspaceManager
-      .listStreams()
-      .filter((s) => !s.parentStream);
-    expect(topLevelStreams.length).toBe(1);
-  });
-});