codex.test.ts 6.7 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119
  1. import fs from 'node:fs'
  2. import path from 'node:path'
  3. import { describe, expect, it } from 'vitest'
  4. import type { AgentEventType, EventPayloads } from '@agent-server/shared'
  5. import { config } from '../src/config'
  6. import { CodexProvider } from '../src/providers/codex/adapter'
  7. import type { RunContext } from '../src/providers/types'
  8. import { resolveAccess } from '../src/services/access-policy'
  9. import type { InteractionOutcome, NewApproval, NewQuestion } from '../src/services/interactions'
  10. import { defaultSettings } from '../src/services/settings'
  11. const fake = path.resolve(__dirname, 'fixtures-fake-codex.mjs')
  12. const settings = defaultSettings()
  13. settings.providers.codex = { ...settings.providers.codex, enabled: true, cliPath: fake }
  14. const silent = { info: () => {}, warn: () => {}, error: () => {} }
  15. function makeCtx(text: string, opts: { providerSessionId?: string; mode?: 'ask' | 'auto' | 'readonly'; answer?: (r: NewApproval | NewQuestion) => InteractionOutcome; signal?: AbortSignal } = {}) {
  16. const ws = path.join(config.dataDir, 'workspaces', `ses_codex_${Math.random().toString(36).slice(2)}`)
  17. fs.mkdirSync(ws, { recursive: true })
  18. const events: { type: AgentEventType; payload: unknown }[] = []
  19. const interactions: (NewApproval | NewQuestion)[] = []
  20. const audits: string[] = []
  21. let sessionId: string | null = null
  22. const ctx: RunContext = {
  23. sessionId: 's', runId: 'r', model: 'gpt-test', permissionMode: opts.mode ?? 'ask', providerSessionId: opts.providerSessionId ?? null,
  24. workspaceDir: ws, systemPromptAppend: '说明', allowedTools: [], disallowedTools: [], skills: 'all', plugins: [],
  25. input: { text, images: [], attachments: [] },
  26. access: resolveAccess({ policy: { mode: 'restricted', read: [], write: [], deny: ['**/*.env'] }, knowledgeRoots: [], selectedKnowledgeRootIds: [], workspaceDir: ws, settings }),
  27. settings, signal: opts.signal ?? new AbortController().signal,
  28. emit: <T extends AgentEventType>(type: T, payload: EventPayloads[T]) => { events.push({ type, payload }) },
  29. interact: async r => { interactions.push(r); return opts.answer?.(r) ?? { resolution: 'allowed', scope: 'once' } },
  30. setProviderSessionId: id => { sessionId = id },
  31. audit: (kind, summary) => audits.push(`${kind}:${summary}`),
  32. log: silent,
  33. }
  34. return { ctx, events, interactions, audits, sessionId: () => sessionId }
  35. }
  36. describe('CodexProvider(模拟 app-server)', () => {
  37. const provider = new CodexProvider(silent)
  38. it('映射文本、命令与审批、用量;传入按 Profile 生成的权限配置', async () => {
  39. const t = makeCtx('请执行命令')
  40. const r = await provider.run(t.ctx)
  41. expect(t.sessionId()).toMatch(/^t-/)
  42. const types = t.events.map(e => e.type)
  43. expect(types).toEqual(expect.arrayContaining(['tool.started', 'tool.completed', 'message.delta', 'message.completed']))
  44. expect(t.interactions[0]).toMatchObject({ kind: 'approval', tool: 'Bash', detail: { command: 'ls -la' } })
  45. expect(t.events.find(e => e.type === 'tool.completed')!.payload).toMatchObject({ status: 'ok', output: 'a.csv' })
  46. // 本次用量 = 最新累计 - (首次累计 - 首次增量) = 220-100, 45-20
  47. expect(r.usage).toEqual({ inputTokens: 120, outputTokens: 25, cacheReadTokens: 0, cacheWriteTokens: 0 })
  48. const cfg = JSON.parse((t.events.find(e => e.type === 'message.completed' && (e.payload as { messageId: string }).messageId === 'cfg')!.payload as { text: string }).text)
  49. const fsTable = cfg.permissions.agent_server.filesystem
  50. expect(cfg.default_permissions).toBe('agent_server')
  51. expect(fsTable[':root']).toBe('deny')
  52. expect(fsTable[':workspace_roots']).toEqual({ '.': 'write', '**/*.env': 'deny' })
  53. expect(fsTable[path.join(config.home, '.ssh')]).toBe('deny')
  54. expect(cfg.permissions.agent_server.network.enabled).toBe(false)
  55. expect(cfg.shell_environment_policy.inherit).toBe('core')
  56. // Profile 选择 all:只禁用始终禁用的系统技能
  57. expect(cfg.skills).toEqual({ config: [{ path: '/sys/skill-installer/SKILL.md', enabled: false }] })
  58. })
  59. it('拒绝审批时工具标记为 denied', async () => {
  60. const t = makeCtx('请执行命令', { answer: () => ({ resolution: 'denied' }) })
  61. await provider.run(t.ctx)
  62. expect(t.events.find(e => e.type === 'tool.completed')!.payload).toMatchObject({ status: 'denied' })
  63. })
  64. it('命令联网请求一律拒绝并审计,不打扰用户', async () => {
  65. const t = makeCtx('需要联网')
  66. await provider.run(t.ctx)
  67. expect(t.interactions).toHaveLength(0)
  68. expect(t.events.some(e => (e.payload as { text?: string }).text === 'network:decline')).toBe(true)
  69. expect(t.audits.some(a => a.startsWith('network_denied'))).toBe(true)
  70. })
  71. it('提问映射为问题卡片,回答按 questionId 返回', async () => {
  72. const t = makeCtx('请提问', { answer: () => ({ resolution: 'answered', answers: { chart: '折线图' } }) })
  73. await provider.run(t.ctx)
  74. expect(t.interactions[0]).toMatchObject({ kind: 'question', questions: [{ id: 'chart', options: [{ label: '柱状图' }, { label: '折线图' }] }] })
  75. expect(t.events.some(e => (e.payload as { text?: string }).text === 'answer:{"chart":{"answers":["折线图"]}}')).toBe(true)
  76. })
  77. it('恢复不存在的会话时新建;中断正在运行的 turn', async () => {
  78. const t = makeCtx('你好', { providerSessionId: 't-999' })
  79. await provider.run(t.ctx)
  80. expect(t.sessionId()).toMatch(/^t-\d+$/)
  81. expect(t.sessionId()).not.toBe('t-999')
  82. const ac = new AbortController()
  83. const t2 = makeCtx('等待', { signal: ac.signal })
  84. const p = provider.run(t2.ctx)
  85. setTimeout(() => ac.abort(), 200)
  86. await expect(p).rejects.toThrow(/中断/)
  87. })
  88. it('设备码登录、账户状态与模型列表', async () => {
  89. const before = await provider.status(settings)
  90. expect(before.loggedIn).toBe(false)
  91. const login = await provider.startLogin(settings)
  92. expect(login).toMatchObject({ verificationUrl: 'https://auth.example.com/device', userCode: 'ABCD-1234', status: 'pending' })
  93. await new Promise(r => setTimeout(r, 300))
  94. expect(provider.loginState()?.status).toBe('success')
  95. const after = await provider.status(settings)
  96. expect(after).toMatchObject({ loggedIn: true, available: true, account: 'me@example.com (plus)', version: '9.9.9', models: ['gpt-test'] })
  97. expect(provider.knownModels()).toEqual(['gpt-test'])
  98. })
  99. })
  100. describe('commandTitle', () => {
  101. it('去掉 Codex 的 shell 包装', async () => {
  102. const { commandTitle } = await import('../src/providers/codex/adapter')
  103. expect(commandTitle(`/usr/bin/bash -lc "python - <<'PY'\nimport x\nPY"`)).toBe(`python - <<'PY'`)
  104. expect(commandTitle(`/usr/bin/bash -lc 'curl -I https://example.com'`)).toBe('curl -I https://example.com')
  105. expect(commandTitle('ls -la')).toBe('ls -la')
  106. })
  107. })