Genarrative/apps/ai-game-creator-shell/scripts/agent-runtime-steer-real-e2e.mjs

import { spawn } from 'node:child_process';
import { randomUUID } from 'node:crypto';
import fs from 'node:fs/promises';
import os from 'node:os';
import path from 'node:path';
import { fileURLToPath } from 'node:url';

const appRoot = path.resolve(fileURLToPath(new URL('..', import.meta.url)));
const repoRoot = path.resolve(appRoot, '../..');
const manifestPath = path.join(appRoot, 'src-tauri/Cargo.toml');
const agentId = 'code-prototype';
const runId = `steer-real-${Date.now()}-${randomUUID().slice(0, 8)}`;
const projectRoot = path.join(os.tmpdir(), `genarrative-${runId}`);
const marker = `STEER_REAL_${randomUUID().replaceAll('-', '').toUpperCase()}`;
const instruction = `运行中补充：继续保持只读，不调用写入、命令、预览或生成工具；最终回复必须原样包含 ${marker}，并明确说明已按追加指令更新。`;
const timeoutMs = 8 * 60 * 1000;
const pollMs = 200;

const options = parseArguments(process.argv.slice(2));
const evidence = {
  providerInterrupted: false,
  steerAttempts: 0,
  taskRunIds: [],
  steerStatuses: [],
  userMessageCount: 0,
  assistantMessageCount: 0,
  publicInstructionLeakCount: 0,
  loadedKeyLeakCount: 0,
};
let status = 'FAIL';
let error = null;
let binary = null;

try {
  const config = await loadConfig(options.configDir);
  const secrets = collectSecrets(config);
  assert(isAgentConfigured(config, agentId), 'llm-not-configured');
  binary = await prepareCliBinary();
  await runCli(
    [
      '--agent-enqueue',
      '--init',
      projectRoot,
      agentId,
      runId,
      '这是只读验收。不要修改文件、不要执行命令、不要预览或生成资产；请仔细分析后给出一句简短结论。',
    ],
    null,
    120_000,
  );

  const initial = await waitForRuntime(
    (runtime) => runtime.state.runId === runId && isSteerable(runtime.state),
    60_000,
  );
  const sessionId = initial.state.sessionId;
  assert(sessionId, 'missing-session-id');

  let latestSteerId = null;
  for (let attempt = 1; attempt <= 3; attempt += 1) {
    const steerId = `steer-real-${attempt}-${randomUUID().slice(0, 8)}`;
    const result = await runCli(
      [
        '--agent-steer',
        projectRoot,
        agentId,
        sessionId,
        runId,
        steerId,
        '--stdin',
      ],
      instruction,
      120_000,
    );
    assert(!result.stdout.includes(instruction), 'steer-body-stdout-leak');
    const steer = parsePrefixedJson(result.stdout, 'steerJson=');
    evidence.steerAttempts = attempt;
    latestSteerId = steerId;
    evidence.providerInterrupted ||= steer.providerInterrupted === true;
    if (evidence.providerInterrupted) break;
    await waitForRuntime(
      (runtime) =>
        runtime.state.runId === runId &&
        Number(runtime.state.appliedSteerCursor ?? 0) >= attempt &&
        isSteerable(runtime.state),
      60_000,
    );
  }
  assert(latestSteerId, 'steer-not-accepted');
  assert(evidence.providerInterrupted, 'provider-not-interrupted');

  const terminal = await waitForRuntime(
    (runtime) =>
      runtime.recentTasks?.some(
        (task) => task.runId === runId && isTerminalTask(task),
      ) === true,
    timeoutMs,
  );
  const terminalTask = terminal.recentTasks.find(
    (task) => task.runId === runId && isTerminalTask(task),
  );
  assert(terminalTask?.status === 'completed', 'steered-run-not-completed');

  const taskRecords = await readJsonl(
    path.join(projectRoot, '.agent/runtime/tasks', `${agentId}.jsonl`),
  );
  evidence.taskRunIds = [...new Set(taskRecords.map((record) => record.runId))];
  assert(
    evidence.taskRunIds.length === 1 && evidence.taskRunIds[0] === runId,
    'steer-created-new-run',
  );
  assert(
    !taskRecords.some((record) => JSON.stringify(record).includes(instruction)),
    'steer-body-task-leak',
  );

  const steerRecords = await readJsonl(
    path.join(projectRoot, '.agent/runtime/steers', agentId, `${runId}.jsonl`),
  );
  evidence.steerStatuses = steerRecords.map((record) => record.status);
  assert(evidence.steerStatuses.includes('prepared'), 'steer-prepared-missing');
  assert(evidence.steerStatuses.includes('queued'), 'steer-queued-missing');
  assert(evidence.steerStatuses.includes('applied'), 'steer-applied-missing');
  assert(evidence.steerStatuses.at(-1) === 'closed', 'steer-ledger-not-closed');
  assert(
    steerRecords
      .filter((record) => record.status === 'prepared')
      .every((record) => typeof record.instruction === 'string'),
    'steer-prepared-body-missing',
  );
  assert(
    steerRecords
      .filter((record) => record.status !== 'prepared')
      .every((record) => record.instruction == null),
    'steer-body-transition-leak',
  );

  const conversationPath =
    sessionId === `agent-session-${agentId}`
      ? path.join(
          projectRoot,
          '.agent/conversations/agents',
          `${agentId}.jsonl`,
        )
      : path.join(
          projectRoot,
          '.agent/conversations/agents',
          agentId,
          'sessions',
          `${sessionId}.jsonl`,
        );
  const messages = await readJsonl(conversationPath);
  evidence.userMessageCount = messages.filter(
    (message) => message.role === 'user',
  ).length;
  evidence.assistantMessageCount = messages.filter(
    (message) => message.role === 'assistant',
  ).length;
  assert(
    evidence.userMessageCount === evidence.steerAttempts + 1,
    'steer-user-message-count-invalid',
  );
  assert(evidence.assistantMessageCount === 1, 'assistant-count-invalid');
  const assistant = messages.find((message) => message.role === 'assistant');
  assert(assistant?.content.includes(marker), 'final-assistant-missing-marker');

  const publicPaths = [
    path.join(projectRoot, '.agent/runtime/tasks', `${agentId}.jsonl`),
    path.join(projectRoot, '.agent/runtime/events', `${agentId}.jsonl`),
    path.join(projectRoot, '.agent/runtime/agents', `${agentId}.json`),
    path.join(projectRoot, '.agent/agent.db'),
  ];
  evidence.publicInstructionLeakCount = await countNeedleInFiles(
    publicPaths,
    instruction,
  );
  assert(evidence.publicInstructionLeakCount === 0, 'steer-body-public-leak');
  evidence.loadedKeyLeakCount = await countNeedlesInTree(projectRoot, secrets);
  assert(evidence.loadedKeyLeakCount === 0, 'loaded-key-project-leak');
  status = 'PASS';
} catch (caught) {
  error = caught instanceof Error ? caught.message : String(caught);
  if (error === 'llm-not-configured') status = 'BLOCKED';
} finally {
  if (!options.keepProject) {
    await fs.rm(projectRoot, { recursive: true, force: true });
  }
  process.stdout.write(
    `${JSON.stringify(
      {
        status,
        suite: 'steer',
        runId,
        projectKept: options.keepProject,
        evidence,
        error,
      },
      null,
      2,
    )}\n`,
  );
  process.exitCode = status === 'PASS' ? 0 : status === 'BLOCKED' ? 2 : 1;
}

function parseArguments(args) {
  let configDir = null;
  let keepProject = false;
  for (let index = 0; index < args.length; index += 1) {
    if (args[index] === '--config-dir') {
      configDir = args[++index];
    } else if (args[index] === '--keep-project') {
      keepProject = true;
    } else {
      throw new Error(`unknown-argument:${args[index]}`);
    }
  }
  assert(configDir && path.isAbsolute(configDir), 'config-dir-not-absolute');
  const resolved = path.resolve(configDir);
  assert(!isInside(repoRoot, resolved), 'config-dir-inside-repository');
  return { configDir: resolved, keepProject };
}

async function loadConfig(configDir) {
  const content = await fs.readFile(
    path.join(configDir, 'game-creator.config.json'),
    'utf8',
  );
  return JSON.parse(content);
}

function isAgentConfigured(config, targetAgentId) {
  const effective = {
    apiKey: config.llm?.apiKey,
    baseUrl: config.llm?.baseUrl,
    model: config.llm?.model,
    ...(config.agentLlm?.[targetAgentId] ?? {}),
  };
  return ['apiKey', 'baseUrl', 'model'].every(
    (key) => typeof effective[key] === 'string' && effective[key].trim(),
  );
}

function collectSecrets(config) {
  const secrets = [];
  for (const candidate of [
    config.llm?.apiKey,
    config.editorApi?.apiKey,
    ...Object.values(config.agentLlm ?? {}).map((agent) => agent?.apiKey),
  ]) {
    if (typeof candidate === 'string' && candidate.trim().length >= 8) {
      secrets.push(candidate.trim());
    }
  }
  return [...new Set(secrets)];
}

async function prepareCliBinary() {
  const cargo = process.platform === 'win32' ? 'cargo.exe' : 'cargo';
  await runProcess(
    cargo,
    ['build', '--quiet', '--manifest-path', manifestPath],
    null,
    15 * 60 * 1000,
  );
  const metadata = await runProcess(
    cargo,
    [
      'metadata',
      '--format-version',
      '1',
      '--no-deps',
      '--manifest-path',
      manifestPath,
    ],
    null,
    120_000,
  );
  const parsed = JSON.parse(metadata.stdout);
  return path.join(
    parsed.target_directory,
    'debug',
    `genarrative-ai-game-creator-shell${process.platform === 'win32' ? '.exe' : ''}`,
  );
}

async function runCli(args, input = null, commandTimeoutMs = 60_000) {
  return runProcess(
    binary,
    [...args, '--config-dir', options.configDir],
    input,
    commandTimeoutMs,
  );
}

async function waitForRuntime(predicate, waitMs) {
  const deadline = Date.now() + waitMs;
  let latest = null;
  while (Date.now() < deadline) {
    const result = await runCli([
      '--agent-runtime-status',
      projectRoot,
      agentId,
    ]);
    latest = parsePrefixedJson(result.stdout, 'runtimeJson=');
    if (predicate(latest)) return latest;
    await new Promise((resolve) => setTimeout(resolve, pollMs));
  }
  throw new Error(
    `runtime-timeout:${latest?.state?.status ?? 'missing'}/${latest?.state?.phase ?? 'missing'}`,
  );
}

function isSteerable(runtime) {
  return (
    ['running', 'waiting-for-confirmation'].includes(runtime.status) &&
    !['cancelling', 'finalizing', 'needs-reconciliation'].includes(
      runtime.phase,
    )
  );
}

function isTerminalTask(task) {
  return ['completed', 'failed', 'cancelled'].includes(task.status);
}

function parsePrefixedJson(stdout, prefix) {
  const line = stdout
    .split(/\r?\n/u)
    .find((candidate) => candidate.startsWith(prefix));
  assert(line, `missing-output:${prefix}`);
  return JSON.parse(line.slice(prefix.length));
}

async function readJsonl(file) {
  const content = await fs.readFile(file, 'utf8');
  return content
    .split(/\r?\n/u)
    .filter((line) => line.trim())
    .map((line) => JSON.parse(line));
}

async function countNeedleInFiles(files, needle) {
  let count = 0;
  for (const file of files) {
    const content = await fs.readFile(file).catch(() => Buffer.alloc(0));
    count += countNeedle(content, Buffer.from(needle));
  }
  return count;
}

async function countNeedlesInTree(root, needles) {
  let count = 0;
  const entries = await fs.readdir(root, { withFileTypes: true });
  for (const entry of entries) {
    const target = path.join(root, entry.name);
    if (entry.isDirectory()) {
      count += await countNeedlesInTree(target, needles);
    } else if (entry.isFile()) {
      const content = await fs.readFile(target);
      for (const needle of needles)
        count += countNeedle(content, Buffer.from(needle));
    }
  }
  return count;
}

function countNeedle(haystack, needle) {
  if (needle.length === 0) return 0;
  let count = 0;
  let offset = 0;
  while (offset <= haystack.length - needle.length) {
    const index = haystack.indexOf(needle, offset);
    if (index < 0) break;
    count += 1;
    offset = index + needle.length;
  }
  return count;
}

function runProcess(program, args, input, commandTimeoutMs) {
  return new Promise((resolve, reject) => {
    const child = spawn(program, args, {
      cwd: appRoot,
      env: { ...process.env, NO_COLOR: '1', RUST_BACKTRACE: '0' },
      stdio: ['pipe', 'pipe', 'pipe'],
    });
    let stdout = '';
    let stderr = '';
    const timer = setTimeout(() => {
      child.kill('SIGKILL');
    }, commandTimeoutMs);
    child.stdout.on('data', (chunk) => {
      stdout += chunk.toString('utf8');
    });
    child.stderr.on('data', (chunk) => {
      stderr += chunk.toString('utf8');
    });
    child.on('error', reject);
    child.on('close', (code, signal) => {
      clearTimeout(timer);
      if (code === 0) {
        resolve({ stdout, stderr });
      } else {
        reject(new Error(`command-failed:${code ?? signal}:${stderr.trim()}`));
      }
    });
    if (input === null) child.stdin.end();
    else child.stdin.end(input);
  });
}

function isInside(parent, target) {
  const relative = path.relative(parent, target);
  return (
    relative === '' ||
    (!relative.startsWith('..') && !path.isAbsolute(relative))
  );
}

function assert(condition, code) {
  if (!condition) throw new Error(code);
}