agent.spec.ts 16 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396
  1. import { describe, expect, expectTypeOf, it } from 'vitest'
  2. import { Context, Service, symbols } from 'cordis'
  3. import { createUserMessage, freezeMessage } from '@deepseek-ai/dsh-llm'
  4. import { Session, SessionId, type UserMessage } from '@deepseek-ai/dsh-session'
  5. import AgentRegistry, {
  6. agentEvents,
  7. Inbox,
  8. } from '@deepseek-ai/dsh-agent'
  9. import type {
  10. Agent,
  11. AgentCancelCause,
  12. AgentFactory,
  13. CreateAgentOptions,
  14. ResumeAgentOptions,
  15. } from '@deepseek-ai/dsh-agent'
  16. function stubAgent(rawId: string, overrides: Partial<Agent> = {}): Agent {
  17. const id = SessionId(rawId)
  18. const session = Session.create(id)
  19. const agent: Agent = {
  20. id,
  21. options: {},
  22. session,
  23. inbox: new Inbox(session, { inserted: () => {}, discarded: () => {}, claimed: () => {} }),
  24. status: 'idle',
  25. ctx: new Context(),
  26. send: () => {},
  27. followup: () => {},
  28. steer: () => ({ outcome: Promise.resolve({ status: 'rejected' as const }) }),
  29. inject: () => {},
  30. cancel() {},
  31. runMaintenance: task => task(new AbortController().signal),
  32. whenIdle: () => Promise.resolve(),
  33. }
  34. return Object.assign(agent, overrides)
  35. }
  36. describe('Inbox', () => {
  37. it('rejects an invalid durable splice during reconstruction', () => {
  38. const session = Session.create(SessionId('invalid-inbox-replay'))
  39. session.append('agent/inbox/spliced', {
  40. target: 'next-turn',
  41. start: 1,
  42. inserted: [],
  43. })
  44. expect(() => new Inbox(session, { inserted: () => {}, discarded: () => {}, claimed: () => {} }))
  45. .toThrow('invalid persisted inbox splice at session seq 0')
  46. })
  47. it('replaces a pending message by identity across both lists', () => {
  48. const session = Session.create(SessionId('replace-inbox'))
  49. const inserted: UserMessage[] = []
  50. const discarded: UserMessage[] = []
  51. const inbox = new Inbox(session, {
  52. claimed: () => {},
  53. inserted: message => void inserted.push(message),
  54. discarded: message => void discarded.push(message),
  55. })
  56. const original = createUserMessage({
  57. content: [{ type: 'text', text: 'original' }],
  58. source: { kind: 'user' },
  59. })
  60. const nextStep = createUserMessage({
  61. content: [{ type: 'text', text: 'step' }],
  62. source: { kind: 'user' },
  63. })
  64. const replacement = createUserMessage({
  65. content: [{ type: 'text', text: 'replacement' }],
  66. source: { kind: 'user' },
  67. })
  68. const editedStep = freezeMessage({
  69. ...nextStep,
  70. content: [{ type: 'text', text: 'edited step' }],
  71. })
  72. inbox.append('next-turn', original)
  73. inbox.append('next-step', nextStep)
  74. expect(inbox.replace(createUserMessage({
  75. content: [{ type: 'text', text: 'missing' }],
  76. source: { kind: 'user' },
  77. }).id, replacement)).toBe(false)
  78. expect(inbox.replace(original.id, replacement)).toBe(true)
  79. expect(inbox.replace(nextStep.id, editedStep)).toBe(true)
  80. expect(inbox.nextTurn).toEqual([replacement])
  81. expect(inbox.nextStep).toEqual([editedStep])
  82. expect(discarded).toEqual([original, nextStep])
  83. expect(inserted).toEqual([original, nextStep, replacement, editedStep])
  84. expect(() => { inbox.replace(editedStep.id, replacement) })
  85. .toThrow(`message "${replacement.id}" is already pending`)
  86. })
  87. it('normalizes splice coordinates, rejects duplicate identities, and reports missing removals', () => {
  88. const session = Session.create(SessionId('splice-inbox'))
  89. const inbox = new Inbox(session, { inserted: () => {}, discarded: () => {}, claimed: () => {} })
  90. const first = createUserMessage({
  91. content: [{ type: 'text', text: 'first' }],
  92. source: { kind: 'user' },
  93. })
  94. const second = createUserMessage({
  95. content: [{ type: 'text', text: 'second' }],
  96. source: { kind: 'user' },
  97. })
  98. inbox.splice('next-turn', Number.NaN, Number.NaN, [first, second])
  99. expect(inbox.nextTurn).toEqual([first, second])
  100. expect(inbox.splice('next-turn', -1, 1, [])).toEqual([second])
  101. expect(inbox.remove(second.id)).toBe(false)
  102. expect(() => { inbox.append('next-step', first) }).toThrow(`message "${first.id}" is already pending`)
  103. })
  104. it('clears both pending lists as durable cancellations', () => {
  105. const session = Session.create(SessionId('clear-inbox'))
  106. const discarded: UserMessage[] = []
  107. const inbox = new Inbox(session, {
  108. claimed: () => {},
  109. inserted: () => {},
  110. discarded: message => void discarded.push(message),
  111. })
  112. const nextTurn = createUserMessage({ content: [{ type: 'text', text: 'turn' }], source: { kind: 'user' } })
  113. const nextStep = createUserMessage({ content: [{ type: 'text', text: 'step' }], source: { kind: 'user' } })
  114. inbox.append('next-turn', nextTurn)
  115. inbox.append('next-step', nextStep)
  116. const beforeClear = session.events.length
  117. inbox.clear()
  118. expect(inbox.hasPending).toBe(false)
  119. expect(discarded).toEqual([nextStep, nextTurn])
  120. expect(session.events.slice(beforeClear).map(event => event.type === 'agent/inbox/spliced'
  121. ? event.data
  122. : event.type)).toEqual([
  123. { target: 'next-step', start: 0, removedCount: 1, inserted: [], outcome: 'canceled' },
  124. { target: 'next-turn', start: 0, removedCount: 1, inserted: [], outcome: 'canceled' },
  125. ])
  126. inbox.clear()
  127. expect(session.events).toHaveLength(beforeClear + 2)
  128. })
  129. })
  130. describe('AgentRegistry', () => {
  131. it('registers exact entries, emits lifecycle events, and unregisters on owner disposal', async () => {
  132. const ctx = new Context()
  133. await ctx.plugin(AgentRegistry)
  134. const lifecycle: string[] = []
  135. ctx.on('agent/created', agent => void lifecycle.push(`created:${agent.id}`))
  136. ctx.on('agent/disposed', agent => void lifecycle.push(`disposed:${agent.id}`))
  137. const agent = stubAgent('a1')
  138. const dispose = ctx.agents.register(agent)
  139. expect(ctx.agents.get(agent.id)).toBe(agent)
  140. expect(ctx.agents.list()).toEqual([agent])
  141. expect(ctx.agents.roots()).toEqual([agent])
  142. expect(() => ctx.agents.register(stubAgent('a1'))).toThrow(/already registered/)
  143. dispose()
  144. expect(ctx.agents.get(agent.id)).toBeUndefined()
  145. expect(lifecycle).toEqual(['created:a1', 'disposed:a1'])
  146. })
  147. it('rejects an agent whose registry and session identities differ', async () => {
  148. const ctx = new Context()
  149. await ctx.plugin(AgentRegistry)
  150. const agent = stubAgent('agent-id', { session: Session.create(SessionId('session-id')) })
  151. expect(() => ctx.agents.enter(agent, undefined))
  152. .toThrow('agent id "agent-id" does not match session id "session-id"')
  153. expect(ctx.agents.list()).toEqual([])
  154. })
  155. it('tracks runtime creator ownership separately from registry order', async () => {
  156. const ctx = new Context()
  157. await ctx.plugin(AgentRegistry)
  158. const root = stubAgent('root')
  159. const child = stubAgent('child')
  160. const detachRoot = ctx.agents.enter(root, undefined)
  161. ctx.agents.announce(root)
  162. const detachChild = ctx.agents.enter(child, root)
  163. ctx.agents.announce(child)
  164. expect(ctx.agents.list()).toEqual([root, child])
  165. expect(ctx.agents.roots()).toEqual([root])
  166. expect(ctx.agents.isOwnedBy(child.id, root)).toBe(true)
  167. expect(ctx.agents.isOwnedBy(root.id, root)).toBe(false)
  168. expect(ctx.agents.isOwnedBy(SessionId('missing'), root)).toBe(false)
  169. detachChild()
  170. expect(ctx.agents.isOwnedBy(child.id, root)).toBe(false)
  171. detachRoot()
  172. })
  173. it('rolls an entry back and pairs a partially delivered creation when a listener throws', async () => {
  174. const ctx = new Context()
  175. await ctx.plugin(AgentRegistry)
  176. const lifecycle: string[] = []
  177. ctx.on('agent/created', agent => void lifecycle.push(`created:${agent.id}`))
  178. ctx.on('agent/created', () => { throw new Error('creation veto') })
  179. ctx.on('agent/disposed', agent => void lifecycle.push(`disposed:${agent.id}`))
  180. expect(() => ctx.agents.register(stubAgent('vetoed'))).toThrow('creation veto')
  181. expect(ctx.agents.get(SessionId('vetoed'))).toBeUndefined()
  182. expect(lifecycle).toEqual(['created:vetoed', 'disposed:vetoed'])
  183. })
  184. it('contains asynchronous creation rejection and every disposal-listener failure', async () => {
  185. const ctx = new Context()
  186. await ctx.plugin(AgentRegistry)
  187. const warnings: string[] = []
  188. const heard: string[] = []
  189. ctx.logger.warn = ((message: unknown) => { warnings.push(String(message)) }) as typeof ctx.logger.warn
  190. ctx.on('agent/created', () => Promise.reject(new Error('created async')) as never)
  191. ctx.on('agent/disposed', () => { throw new Error('disposed sync') })
  192. ctx.on('agent/disposed', () => Promise.reject(new Error('disposed async')) as never)
  193. ctx.on('agent/disposed', agent => void heard.push(agent.id))
  194. const dispose = ctx.agents.register(stubAgent('contained'))
  195. await Promise.resolve()
  196. dispose()
  197. await Promise.resolve()
  198. expect(heard).toEqual(['contained'])
  199. expect(warnings).toEqual([
  200. 'agent "contained": agent/created listener rejected: Error: created async',
  201. 'agent "contained": agent/disposed listener threw: Error: disposed sync',
  202. 'agent "contained": agent/disposed listener rejected: Error: disposed async',
  203. ])
  204. })
  205. it('separates entry from announcement and stale/idempotent detach cannot remove a replacement', async () => {
  206. const ctx = new Context()
  207. await ctx.plugin(AgentRegistry)
  208. const lifecycle: string[] = []
  209. ctx.on('agent/created', agent => void lifecycle.push(`created:${agent.id}`))
  210. ctx.on('agent/disposed', agent => void lifecycle.push(`disposed:${agent.id}`))
  211. const first = stubAgent('split')
  212. const detachFirst = ctx.agents.enter(first, undefined)
  213. expect(lifecycle).toEqual([])
  214. ctx.agents.announce(first)
  215. expect(() => { ctx.agents.announce(first) }).toThrow(/already announced/)
  216. detachFirst()
  217. detachFirst()
  218. const replacement = stubAgent('split')
  219. const detachReplacement = ctx.agents.enter(replacement, undefined)
  220. detachFirst()
  221. expect(ctx.agents.get(replacement.id)).toBe(replacement)
  222. expect(() => { ctx.agents.announce(first) }).toThrow(/not live/)
  223. detachReplacement()
  224. expect(lifecycle).toEqual(['created:split', 'disposed:split'])
  225. })
  226. it('defers detach requested by a creation listener until that dispatch unwinds', async () => {
  227. const ctx = new Context()
  228. await ctx.plugin(AgentRegistry)
  229. const order: string[] = []
  230. const agent = stubAgent('reentrant')
  231. ctx.on('agent/created', () => {
  232. order.push(`first:${ctx.agents.get(agent.id) === agent}`)
  233. detach()
  234. order.push(`after-detach:${ctx.agents.get(agent.id) === agent}`)
  235. })
  236. ctx.on('agent/created', () => void order.push(`second:${ctx.agents.get(agent.id) === agent}`))
  237. ctx.on('agent/disposed', () => void order.push('disposed'))
  238. const detach = ctx.agents.enter(agent, undefined)
  239. ctx.agents.announce(agent)
  240. expect(order).toEqual(['first:true', 'after-detach:true', 'second:true', 'disposed'])
  241. expect(ctx.agents.get(agent.id)).toBeUndefined()
  242. })
  243. })
  244. describe('agentEvents()', () => {
  245. it('contains each synchronous throw and returned-promise rejection', async () => {
  246. const ctx = new Context()
  247. const warnings: string[] = []
  248. const heard: string[] = []
  249. ctx.logger.warn = ((message: unknown) => { warnings.push(String(message)) }) as typeof ctx.logger.warn
  250. const agent = stubAgent('event')
  251. ctx.on('agent/status', () => { throw new Error('sync listener') })
  252. ctx.on('agent/status', () => Promise.reject(new Error('async listener')) as never)
  253. ctx.on('agent/status', (_agent, status) => void heard.push(status))
  254. agentEvents(ctx, agent).emit('agent/status', 'running')
  255. await Promise.resolve()
  256. expect(heard).toEqual(['running'])
  257. expect(warnings).toEqual([
  258. 'agent event "agent/status" listener threw: Error: sync listener',
  259. 'agent event "agent/status" listener rejected: Error: async listener',
  260. ])
  261. })
  262. it('dispatches serial listeners with the fused agent subject', async () => {
  263. const ctx = new Context()
  264. const agent = stubAgent('serial-event')
  265. const signal = new AbortController().signal
  266. const heard: Array<{ agent: Agent; turn: number; signal: AbortSignal }> = []
  267. ctx.on('agent/turn-stopping', async (subject, turn, receivedSignal) => {
  268. await Promise.resolve()
  269. heard.push({ agent: subject, turn, signal: receivedSignal })
  270. })
  271. await agentEvents(ctx, agent).serial('agent/turn-stopping', 3, signal)
  272. expect(heard).toEqual([{ agent, turn: 3, signal }])
  273. })
  274. })
  275. describe('explicit cancellation contract', () => {
  276. it('exposes the closed typed cancellation cause at the Agent seam', () => {
  277. expectTypeOf<Parameters<Agent['cancel']>[0]>().toEqualTypeOf<AgentCancelCause>()
  278. })
  279. })
  280. describe('AgentRegistry factory seam', () => {
  281. function stubFactory() {
  282. const calls: {
  283. create: Array<{ ownerCtx: Context; options: CreateAgentOptions }>
  284. resume: Array<{ ownerCtx: Context; options: ResumeAgentOptions }>
  285. } = { create: [], resume: [] }
  286. const factory: AgentFactory = {
  287. async createAgent(ownerCtx, options) {
  288. calls.create.push({ ownerCtx, options })
  289. return { agent: stubAgent(options.sessionId), dispose: () => Promise.resolve() }
  290. },
  291. async resume(ownerCtx, options) {
  292. calls.resume.push({ ownerCtx, options })
  293. return { agent: stubAgent(options.resumeSessionId), dispose: () => Promise.resolve() }
  294. },
  295. }
  296. return { factory, calls }
  297. }
  298. it('requires a factory and delegates through the calling context', async () => {
  299. const ctx = new Context()
  300. await ctx.plugin(AgentRegistry)
  301. await expect(ctx.agents.create({ sessionId: SessionId('s') })).rejects.toThrow(/no agent factory/)
  302. const { factory, calls } = stubFactory()
  303. ctx.agents.setFactory(factory)
  304. let callerFiber: Context['fiber'] | undefined
  305. await ctx.plugin(Object.assign(async (inner: Context) => {
  306. callerFiber = inner.fiber
  307. await inner.agents.create({ sessionId: SessionId('create-s') })
  308. await inner.agents.resume({ resumeSessionId: SessionId('resume-s') })
  309. }, { inject: ['agents'] }))
  310. expect(calls.create[0]?.ownerCtx.fiber).toBe(callerFiber)
  311. expect(calls.resume[0]?.ownerCtx.fiber).toBe(callerFiber)
  312. })
  313. it('rejects a second factory and clears the slot with its owner (HMR)', async () => {
  314. const ctx = new Context()
  315. await ctx.plugin(AgentRegistry)
  316. const owner = await ctx.plugin(Object.assign((inner: Context) => {
  317. inner.agents.setFactory(stubFactory().factory)
  318. expect(() => inner.agents.setFactory(stubFactory().factory)).toThrow(/already registered/)
  319. }, { inject: ['agents'] }))
  320. await expect(ctx.agents.create({ sessionId: SessionId('before-s') })).resolves.toBeDefined()
  321. await owner.dispose()
  322. await expect(ctx.agents.create({ sessionId: SessionId('after-s') })).rejects.toThrow(/no agent factory/)
  323. })
  324. it('canonicalizes an already traced Service before tracing it for the caller', async () => {
  325. const ctx = new Context()
  326. await ctx.plugin(AgentRegistry)
  327. const states = new WeakMap<object, string[]>()
  328. class TracedFactory extends Service implements AgentFactory {
  329. constructor(inner: Context) {
  330. super(inner, 'tracedFactory')
  331. states.set(this, [])
  332. }
  333. private calls(): string[] {
  334. const original = (this as unknown as { [symbols.original]?: TracedFactory })[symbols.original] ?? this
  335. const calls = states.get(original)
  336. if (calls === undefined) throw new Error('factory receiver was not canonicalized')
  337. return calls
  338. }
  339. async createAgent(_ownerCtx: Context, options: CreateAgentOptions) {
  340. this.calls().push('create')
  341. return { agent: stubAgent(options.sessionId), dispose: () => Promise.resolve() }
  342. }
  343. async resume(_ownerCtx: Context, options: ResumeAgentOptions) {
  344. this.calls().push('resume')
  345. return { agent: stubAgent(options.resumeSessionId), dispose: () => Promise.resolve() }
  346. }
  347. }
  348. await ctx.plugin(TracedFactory)
  349. const traced = (ctx as Context & { tracedFactory: TracedFactory }).tracedFactory
  350. ctx.agents.setFactory(traced)
  351. await ctx.agents.create({ sessionId: SessionId('create-s') })
  352. await ctx.agents.resume({ resumeSessionId: SessionId('resume-s') })
  353. const raw = (traced as unknown as { [symbols.original]?: TracedFactory })[symbols.original]
  354. expect(states.get(raw!)).toEqual(['create', 'resume'])
  355. })
  356. })