@canonmsg/codex-plugin 0.29.4 → 0.31.0

package/scripts/smoke-work-sessions.mjs

@@ -0,0 +1,218 @@
+#!/usr/bin/env node
+// Real shared Codex protocol with a localhost fake model and Canon publisher.
+// No real model/Canon calls; no HOME/CODEX_HOME or existing-session changes.
+import assert from 'node:assert/strict';
+import { spawn } from 'node:child_process';
+import { mkdtemp, rm } from 'node:fs/promises';
+import { createServer as httpServer } from 'node:http';
+import { createServer as netServer } from 'node:net';
+import { tmpdir } from 'node:os';
+import { join } from 'node:path';
+import WebSocket from 'ws';
+import { createAttachedNativeSession, createFileAttachedSessionStore } from '@canonmsg/core';
+import { createCanonAttachedSessionPublisher } from '@canonmsg/agent-sdk';
+import { CodexWorkSessionProvider } from '../dist/work-session-provider.js';
+
+const directory = await mkdtemp(join(tmpdir(), 'canon-work-session-smoke-'));
+const sleep = (ms) => new Promise((resolve) => setTimeout(resolve, ms));
+async function until(predicate, label) {
+  for (let attempt = 0; attempt < 300; attempt += 1) {
+    if (await predicate()) return;
+    await sleep(50);
+  }
+  throw new Error(`Timed out: ${label}`);
+}
+async function listen(server) {
+  await new Promise((resolve) => server.listen(0, '127.0.0.1', resolve));
+  return server.address().port;
+}
+
+let modelCalls = 0;
+let catalogCalls = 0;
+let askApproval = false;
+const model = httpServer((request, response) => {
+  request.resume();
+  if (request.method === 'GET' && request.url?.startsWith('/v1/models')) {
+    catalogCalls += 1;
+    // Exercise the native server's built-in model catalog fallback.
+    response.writeHead(404, { 'content-type': 'application/json' });
+    response.end(JSON.stringify({ error: 'This disposable provider does not implement model discovery.' }));
+    return;
+  }
+  assert.equal(request.method, 'POST');
+  assert.equal(request.url, '/v1/responses');
+  modelCalls += 1;
+  const item = askApproval ? {
+    id: `function_${modelCalls}`, type: 'function_call', call_id: `approval_${modelCalls}`, name: 'exec_command',
+    arguments: JSON.stringify({ cmd: 'printf canon-work-session-smoke', sandbox_permissions: 'require_escalated', justification: 'Disposable test; the command will be declined.' }),
+  } : { id: `message_${modelCalls}`, type: 'message', role: 'assistant', status: 'completed', content: [{ type: 'output_text', text: `Mock answer ${modelCalls}.`, annotations: [] }] };
+  askApproval = false;
+  const result = { id: `response_${modelCalls}`, object: 'response', model: 'probe', status: 'completed', output: [item], usage: { input_tokens: 1, output_tokens: 1, total_tokens: 2 } };
+  response.writeHead(200, { 'content-type': 'text/event-stream' });
+  for (const event of [
+    { type: 'response.created', response: { ...result, status: 'in_progress', output: [] } },
+    { type: 'response.output_item.added', output_index: 0, item },
+    { type: 'response.output_item.done', output_index: 0, item },
+    { type: 'response.completed', response: result },
+  ]) response.write(`data: ${JSON.stringify(event)}\n\n`);
+  response.end();
+});
+const modelPort = await listen(model);
+const reservation = netServer();
+const port = await listen(reservation);
+await new Promise((resolve) => reservation.close(resolve));
+const endpoint = `ws://127.0.0.1:${port}`;
+const { CODEX_INTERNAL_ORIGINATOR_OVERRIDE: _originatorOverride, ...serverEnv } = process.env;
+const server = spawn(process.argv[2] ?? 'codex', [
+  'app-server', '--disable', 'plugins', '--disable', 'apps', '--listen', endpoint,
+  '-c', 'model_provider="canon_smoke"',
+  '-c', `model_providers.canon_smoke={name="Canon local mock provider",base_url="http://127.0.0.1:${modelPort}/v1",wire_api="responses",requires_openai_auth=false}`,
+], { cwd: directory, env: serverEnv, stdio: ['ignore', 'pipe', 'pipe'] });
+server.stdout.resume();
+let serverError = '';
+server.stderr.on('data', (data) => { serverError = `${serverError}${data}`.slice(-1_000); });
+
+let provider;
+let controller;
+let native;
+let threadId;
+const bridgeRequests = [];
+const nativeResponses = [];
+const published = new Map();
+const errors = [];
+const bridge = { request(request, { signal }) {
+  return new Promise((resolve) => { bridgeRequests.push({ request, signal, resolve }); });
+} };
+function makeProvider() {
+  return new CodexWorkSessionProvider({ endpoint, cwd: directory, defaultExecutionMode: 'locked',
+    createSocket(url) {
+      const socket = new WebSocket(url);
+      const send = socket.send.bind(socket);
+      socket.send = (data, ...args) => {
+        const message = JSON.parse(data.toString());
+        if ('id' in message && !message.method) nativeResponses.push(message);
+        return send(data, ...args);
+      };
+      return socket;
+    },
+    onError: (error) => errors.push(error.message),
+  });
+}
+async function connectNative() {
+  const socket = new WebSocket(endpoint);
+  await new Promise((resolve, reject) => { socket.once('open', resolve); socket.once('error', reject); });
+  let sequence = 0;
+  const pending = new Map();
+  const events = [];
+  const requests = [];
+  socket.on('message', (data) => {
+    const message = JSON.parse(data.toString());
+    if (message.method) { ('id' in message ? requests : events).push(message); return; }
+    const waiter = pending.get(message.id);
+    if (!waiter) return;
+    pending.delete(message.id);
+    clearTimeout(waiter.timer);
+    message.error ? waiter.reject(new Error(JSON.stringify(message.error))) : waiter.resolve(message.result);
+  });
+  const rpc = (method, params = {}) => new Promise((resolve, reject) => {
+    const id = ++sequence;
+    const timer = setTimeout(() => { pending.delete(id); reject(new Error(`Timeout: ${method}`)); }, 10_000);
+    pending.set(id, { resolve, reject, timer });
+    socket.send(JSON.stringify({ id, method, params }));
+  });
+  await rpc('initialize', { clientInfo: { name: 'canon_work_session_smoke', version: '0.0.0' }, capabilities: { experimentalApi: true } });
+  socket.send(JSON.stringify({ method: 'initialized' }));
+  return { socket, rpc, events, requests };
+}
+
+try {
+  await until(async () => {
+    if (server.exitCode !== null) throw new Error(serverError);
+    return fetch(`http://127.0.0.1:${port}/readyz`).then((result) => result.ok).catch(() => false);
+  }, 'native server readiness');
+  provider = makeProvider();
+  const catalog = await provider.catalog();
+  assert.ok(catalog.models.length > 0);
+  assert.ok(!JSON.stringify(catalog).includes(directory), 'Private catalog must not expose raw cwd');
+  const created = await provider.create({ mode: 'create', projectId: catalog.projects[0].id,
+    modelId: catalog.defaults.modelId, executionMode: 'locked', permissionMode: 'workspace' },
+  { requestId: 'create-smoke', conversationId: 'room', agentId: 'agent' });
+  threadId = created.nativeSessionId;
+  const binding = { environmentId: 'smoke', agentId: 'agent', conversationId: 'room', provider: 'codex', nativeSessionId: threadId };
+  const publisher = createCanonAttachedSessionPublisher({ async sendProactiveMessage(conversationId, text, options) {
+    const payload = JSON.stringify({ text, options });
+    if (published.has(options.messageId)) assert.equal(published.get(options.messageId), payload);
+    published.set(options.messageId, payload);
+    return { messageId: options.messageId, conversationId };
+  } }, binding);
+  async function attach() {
+    const value = createAttachedNativeSession({ binding, adapter: await provider.open(threadId, bridge),
+      store: createFileAttachedSessionStore(join(directory, 'journal.json')), publisher,
+      onError: (error, operation) => errors.push(`${operation}: ${error.message}`),
+    });
+    await value.start();
+    return value;
+  }
+  controller = await attach();
+  assert.equal(modelCalls, 0, 'Creating/attaching an empty session must not inject a warmup turn');
+  assert.equal(published.size, 0);
+  askApproval = true;
+  const first = await controller.submit({ messageId: 'canon-first', text: 'First real input from Canon.' });
+  assert.equal(first.status, 'accepted');
+  await until(() => bridgeRequests.length === 1, 'Canon receives approval without any native client');
+  assert.equal(bridgeRequests[0].request.kind, 'approval');
+  bridgeRequests[0].resolve({ kind: 'approval', decision: 'deny' });
+  await until(async () => { await controller.reconcile(); return controller.getState().activeTurnId === null && published.size === 1; }, 'first Canon turn completion');
+  assert.equal(nativeResponses.length, 1);
+  assert.deepEqual(nativeResponses[0].result, { decision: 'decline' });
+
+  native = await connectNative();
+  const resumed = await native.rpc('thread/resume', { threadId, excludeTurns: true });
+  assert.equal(resumed.thread.id, threadId, 'Native client resumes the exact created session');
+  const local = await native.rpc('turn/start', { threadId, input: [{ type: 'text', text: 'Input from the native surface.' }] });
+  await until(() => native.events.some((event) => event.method === 'turn/completed' && event.params.turn.id === local.turn.id), 'native turn');
+  await controller.reconcile();
+  assert.equal(published.size, 3, 'Native user and assistant text should mirror once');
+
+  askApproval = true;
+  const second = await controller.submit({ messageId: 'canon-second', text: 'Native client will answer this approval.' });
+  await until(() => bridgeRequests.length === 2 && native.requests.some((request) => request.params.turnId === second.turnId), 'shared approval delivery');
+  const approval = native.requests.find((request) => request.params.turnId === second.turnId);
+  native.socket.send(JSON.stringify({ id: approval.id, result: { decision: 'decline' } }));
+  await until(() => bridgeRequests[1].signal.aborted, 'native answer cancels Canon interaction');
+  bridgeRequests[1].resolve({ kind: 'approval', decision: 'allow' });
+  await until(async () => { await controller.reconcile(); return controller.getState().activeTurnId === null && published.size === 4; }, 'native approval resolution');
+  assert.equal(nativeResponses.length, 1, 'A late Canon answer must not override the native decision');
+
+  askApproval = true;
+  await controller.submit({ messageId: 'canon-third', text: 'Recover this pending approval after the Canon host restarts.' });
+  await until(() => bridgeRequests.length === 3, 'approval before host restart');
+  await controller.stop();
+  await provider.close();
+  assert.equal(bridgeRequests[2].signal.aborted, true);
+  provider = makeProvider();
+  controller = await attach();
+  await until(() => bridgeRequests.length === 4, 'pending native approval replay after reconnect');
+  assert.equal(bridgeRequests[3].request.nativeRequestId, bridgeRequests[2].request.nativeRequestId);
+  bridgeRequests[3].resolve({ kind: 'approval', decision: 'deny' });
+  await until(async () => { await controller.reconcile(); return controller.getState().activeTurnId === null && published.size === 5; }, 'recovered approval turn');
+  assert.equal(errors.length, 0, errors.join('\n'));
+  await controller.stop();
+  await provider.close();
+  assert.equal((await native.rpc('thread/read', { threadId, includeTurns: false })).thread.id, threadId);
+  console.log(JSON.stringify({ passed: true, emptySessionWithoutWarmup: true, canonApprovalWithoutNativeUi: true, nativeAndCanonInput: true, nativeAnswerCancelsCanon: true, pendingApprovalRecovery: true, publishedMessages: published.size, localMockModelCalls: modelCalls, localCatalogCalls: catalogCalls }));
+} finally {
+  await controller?.stop().catch(() => {});
+  await provider?.close().catch(() => {});
+  if (threadId && server.exitCode === null) {
+    native ??= await connectNative().catch(() => null);
+    if (native) await native.rpc('thread/delete', { threadId }).catch((error) => console.error(`Could not delete disposable native thread: ${error.message}`));
+  }
+  native?.socket.terminate();
+  server.kill('SIGTERM');
+  await sleep(400);
+  if (server.exitCode === null) server.kill('SIGKILL');
+  model.closeAllConnections();
+  await new Promise((resolve) => model.close(resolve));
+  await rm(directory, { recursive: true, force: true });
+}