github-webhook-real.e2e.ts 17 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468
  1. /** Real CLI and DeepSeek evidence for a GitHub webhook-created Session. */
  2. import type { ChildProcess } from 'node:child_process'
  3. import { spawn } from 'node:child_process'
  4. import { createHmac, randomUUID } from 'node:crypto'
  5. import { existsSync } from 'node:fs'
  6. import { mkdir, mkdtemp, realpath, rm } from 'node:fs/promises'
  7. import { createServer } from 'node:net'
  8. import type { AddressInfo } from 'node:net'
  9. import { tmpdir } from 'node:os'
  10. import { join } from 'node:path'
  11. import { setTimeout as delay } from 'node:timers/promises'
  12. import { fileURLToPath } from 'node:url'
  13. import { decodeStorageRecord } from '@deepseek-ai/dsh-session/chunk-rows'
  14. import { describe, expect, it } from 'vitest'
  15. import WebSocket from 'ws'
  16. const REPO_ROOT = fileURLToPath(new URL('../../..', import.meta.url))
  17. const BUILT_BIN = join(REPO_ROOT, 'apps/cli/lib/bin.js')
  18. const OVERLAY = fileURLToPath(new URL(
  19. './fixtures/github-webhook/cordis.yml',
  20. import.meta.url,
  21. ))
  22. const SECRET = 'github-webhook-real-e2e-secret'
  23. const DELIVERY = 'github-webhook-real-e2e-delivery'
  24. const MARKER = 'DSH_GITHUB_WEBHOOK_REAL_E2E_OK'
  25. const TITLE = 'GitHub webhook real e2e'
  26. const authenticatedCookies = new Map<string, Promise<{ origin: string; cookie: string }>>()
  27. /** Exchange the printed process token once for Node-side API probes. */
  28. function authenticatedWeb(launchUrl: string): Promise<{ origin: string; cookie: string }> {
  29. const existing = authenticatedCookies.get(launchUrl)
  30. if (existing !== undefined) return existing
  31. const exchange = (async () => {
  32. const response = await fetch(launchUrl, { redirect: 'manual' })
  33. const setCookie = response.headers.get('set-cookie')
  34. if (response.status !== 303 || setCookie === null) {
  35. throw new Error(`dsh web authentication returned HTTP ${String(response.status)}`)
  36. }
  37. return { origin: new URL(launchUrl).origin, cookie: setCookie.split(';', 1)[0]! }
  38. })()
  39. authenticatedCookies.set(launchUrl, exchange)
  40. return exchange
  41. }
  42. interface SessionList {
  43. items: Array<{
  44. sessionId: string
  45. cwd?: string
  46. blank: boolean
  47. projections?: { values: { agentPreset?: string | null } }
  48. }>
  49. }
  50. interface WorkspaceBaseline {
  51. items: Array<{
  52. path: string
  53. sessionIds: string[]
  54. }>
  55. }
  56. interface HistoryPage {
  57. records: Array<{ event: HistoryEvent } | { chunks: unknown }>
  58. hasMore: boolean
  59. }
  60. interface HistoryEvent {
  61. type: string
  62. data: unknown
  63. }
  64. interface ProcessObservation {
  65. readonly ready: Promise<string>
  66. readonly text: () => string
  67. }
  68. function isRecord(value: unknown): value is Record<string, unknown> {
  69. return typeof value === 'object' && value !== null
  70. }
  71. /** Capture bounded process output and resolve the public Web URL after settled boot. */
  72. function observeProcess(child: ChildProcess): ProcessObservation {
  73. let output = ''
  74. let settled = false
  75. let resolveReady!: (url: string) => void
  76. let rejectReady!: (error: Error) => void
  77. const ready = new Promise<string>((resolve, reject) => {
  78. resolveReady = resolve
  79. rejectReady = reject
  80. })
  81. const timer = setTimeout(() => {
  82. if (!settled) rejectReady(new Error(`dsh web did not become ready within 90s:\n${output}`))
  83. }, 90_000)
  84. timer.unref()
  85. const append = (chunk: Buffer | string): void => {
  86. output = `${output}${String(chunk)}`.slice(-100_000)
  87. const match = /dsh web: (http:\/\/[^\s]+)/u.exec(output)
  88. if (settled || match?.[1] === undefined) return
  89. settled = true
  90. clearTimeout(timer)
  91. resolveReady(match[1].replace('0.0.0.0', '127.0.0.1'))
  92. }
  93. child.stdout?.on('data', append)
  94. child.stderr?.on('data', append)
  95. child.once('error', (error) => {
  96. if (!settled) rejectReady(error)
  97. })
  98. child.once('exit', (code) => {
  99. if (!settled) rejectReady(new Error(`dsh web exited before readiness (code ${String(code)}):\n${output}`))
  100. })
  101. return { ready, text: () => output }
  102. }
  103. /** Reserve and release one loopback port for the isolated webhook listener. */
  104. async function freePort(): Promise<number> {
  105. const server = createServer()
  106. await new Promise<void>((resolve, reject) => {
  107. server.once('error', reject)
  108. server.listen(0, '127.0.0.1', resolve)
  109. })
  110. const port = (server.address() as AddressInfo).port
  111. await new Promise<void>((resolve, reject) => {
  112. server.close((error) => {
  113. if (error === undefined) resolve()
  114. else reject(error)
  115. })
  116. })
  117. return port
  118. }
  119. /** Invoke one public Remote method over its HTTP carrier. */
  120. async function remoteRpc<T>(baseUrl: string, endpoint: string, args: object): Promise<T> {
  121. const authenticated = await authenticatedWeb(baseUrl)
  122. const response = await fetch(`${authenticated.origin}/api/${endpoint}`, {
  123. method: 'POST',
  124. headers: { 'content-type': 'application/json', cookie: authenticated.cookie },
  125. body: JSON.stringify({
  126. type: 'client-request',
  127. rpcId: `github-webhook-real-${endpoint}-${randomUUID()}`,
  128. method: endpoint,
  129. payload: { args },
  130. }),
  131. })
  132. if (!response.ok) {
  133. throw new Error(`${endpoint} returned HTTP ${String(response.status)}: ${await response.text()}`)
  134. }
  135. const envelope = await response.json() as {
  136. result: { ok: true; value: T } | { ok: false; error: { code: string; message: string } }
  137. }
  138. if (!envelope.result.ok) {
  139. throw new Error(`${endpoint} failed: ${envelope.result.error.code}: ${envelope.result.error.message}`)
  140. }
  141. return envelope.result.value
  142. }
  143. /** Read one opening item from a public Remote stream. */
  144. async function openingStreamItem(
  145. baseUrl: string,
  146. endpoint: string,
  147. args: object,
  148. accepts: (value: unknown) => boolean,
  149. ): Promise<Record<string, unknown>> {
  150. const authenticated = await authenticatedWeb(baseUrl)
  151. const socket = new WebSocket(`${authenticated.origin.replace(/^http/u, 'ws')}/api/remote.mux`, {
  152. headers: { cookie: authenticated.cookie },
  153. })
  154. const streamId = `github-webhook-real-${endpoint}-${randomUUID()}`
  155. try {
  156. await new Promise<void>((resolve, reject) => {
  157. const cleanup = (): void => {
  158. socket.removeEventListener('open', opened)
  159. socket.removeEventListener('error', failed)
  160. socket.removeEventListener('close', closed)
  161. }
  162. const opened = (): void => {
  163. cleanup()
  164. resolve()
  165. }
  166. const failed = (): void => {
  167. cleanup()
  168. reject(new Error(`${endpoint} carrier failed before opening`))
  169. }
  170. const closed = (): void => {
  171. cleanup()
  172. reject(new Error(`${endpoint} carrier closed before opening`))
  173. }
  174. socket.addEventListener('open', opened)
  175. socket.addEventListener('error', failed)
  176. socket.addEventListener('close', closed)
  177. })
  178. return await new Promise<Record<string, unknown>>((resolve, reject) => {
  179. const timer = setTimeout(() => { finish(new Error(`${endpoint} did not publish its opening item`)) }, 10_000)
  180. const cleanup = (): void => {
  181. clearTimeout(timer)
  182. socket.removeEventListener('message', message)
  183. socket.removeEventListener('error', failed)
  184. socket.removeEventListener('close', closed)
  185. }
  186. const finish = (error: Error | undefined, value?: Record<string, unknown>): void => {
  187. cleanup()
  188. if (error !== undefined) reject(error)
  189. else if (value === undefined) reject(new Error(`${endpoint} opening item was absent`))
  190. else resolve(value)
  191. }
  192. const message = (event: WebSocket.MessageEvent): void => {
  193. try {
  194. const text = typeof event.data === 'string'
  195. ? event.data
  196. : Buffer.isBuffer(event.data) ? event.data.toString('utf8') : undefined
  197. if (text === undefined) throw new Error(`${endpoint} published a non-text frame`)
  198. const frame: unknown = JSON.parse(text)
  199. if (!isRecord(frame) || frame.streamId !== streamId) return
  200. if (frame.type === 'error') {
  201. finish(new Error(`${endpoint} failed: ${JSON.stringify(frame.error)}`))
  202. return
  203. }
  204. if (frame.type === 'end') {
  205. finish(new Error(`${endpoint} ended before its opening item`))
  206. return
  207. }
  208. if (frame.type === 'item' && isRecord(frame.value) && accepts(frame.value)) {
  209. finish(undefined, frame.value)
  210. }
  211. } catch (error) {
  212. finish(error instanceof Error ? error : new Error(String(error)))
  213. }
  214. }
  215. const failed = (): void => { finish(new Error(`${endpoint} carrier failed before its opening item`)) }
  216. const closed = (): void => { finish(new Error(`${endpoint} carrier closed before its opening item`)) }
  217. socket.addEventListener('message', message)
  218. socket.addEventListener('error', failed)
  219. socket.addEventListener('close', closed)
  220. socket.send(JSON.stringify({ type: 'open', streamId, endpoint, payload: { args } }))
  221. })
  222. } finally {
  223. socket.close()
  224. }
  225. }
  226. /** Read the current Workspace baseline from a fresh follow generation. */
  227. async function workspaceBaseline(baseUrl: string): Promise<WorkspaceBaseline> {
  228. const frame = await openingStreamItem(
  229. baseUrl,
  230. 'workspace/follow',
  231. {},
  232. value => isRecord(value) && value.type === 'baseline' && isRecord(value.value),
  233. )
  234. return frame.value as WorkspaceBaseline
  235. }
  236. /** Read the complete opening page from a fresh Session follow generation. */
  237. async function history(baseUrl: string, sessionId: string): Promise<HistoryPage> {
  238. const frame = await openingStreamItem(
  239. baseUrl,
  240. 'session/follow',
  241. { request: { address: { kind: 'session', sessionId }, maxMessages: 100 } },
  242. value => isRecord(value)
  243. && value.type === 'snapshot'
  244. && Array.isArray(value.records)
  245. && typeof value.hasMore === 'boolean',
  246. )
  247. return { records: frame.records as HistoryPage['records'], hasMore: frame.hasMore as boolean }
  248. }
  249. /** Poll a public observation until it satisfies the test's behavior predicate. */
  250. async function eventually<T>(
  251. child: ChildProcess,
  252. processOutput: () => string,
  253. label: string,
  254. probe: () => Promise<T>,
  255. accepts: (value: T) => boolean,
  256. timeoutMs: number,
  257. ): Promise<T> {
  258. const deadline = Date.now() + timeoutMs
  259. let lastValue: T | undefined
  260. let lastError: unknown
  261. while (Date.now() < deadline) {
  262. if (child.exitCode !== null) {
  263. throw new Error(`dsh web exited while waiting for ${label} (code ${String(child.exitCode)}):\n${processOutput()}`)
  264. }
  265. try {
  266. lastValue = await probe()
  267. if (accepts(lastValue)) return lastValue
  268. } catch (error) {
  269. lastError = error
  270. }
  271. await delay(300)
  272. }
  273. throw new Error(
  274. `timed out waiting for ${label}; last value=${JSON.stringify(lastValue)}; `
  275. + `last error=${String(lastError)}; process output:\n${processOutput()}`,
  276. )
  277. }
  278. /** Return every text block from durable assistant messages. */
  279. function assistantText(page: HistoryPage): string {
  280. const text: string[] = []
  281. for (const event of historyEvents(page)) {
  282. if (event.type !== 'assistant/message' || !isRecord(event.data) || !isRecord(event.data.message)) continue
  283. const content = event.data.message.content
  284. if (!Array.isArray(content)) continue
  285. for (const block of content) {
  286. if (isRecord(block) && block.type === 'text' && typeof block.text === 'string') text.push(block.text)
  287. }
  288. }
  289. return text.join('\n')
  290. }
  291. /** Expand lossless history records for assertions over the public event stream. */
  292. function historyEvents(page: HistoryPage): HistoryEvent[] {
  293. return page.records.flatMap(record => 'event' in record ? [record.event] : decodeStorageRecord(record.chunks))
  294. }
  295. /** Stop the spawned CLI through its normal signal path, escalating only on a stuck teardown. */
  296. async function stop(child: ChildProcess): Promise<void> {
  297. if (child.exitCode !== null) return
  298. let resolveClosed!: () => void
  299. const closed = new Promise<void>((resolve) => { resolveClosed = resolve })
  300. child.once('close', resolveClosed)
  301. child.kill('SIGTERM')
  302. if (await Promise.race([closed.then(() => true), delay(10_000, false, { ref: false })])) return
  303. if (child.exitCode === null) child.kill('SIGKILL')
  304. await Promise.race([closed, delay(5_000, undefined, { ref: false })])
  305. }
  306. /** Send the sole synthetic external interaction: one signed GitHub delivery. */
  307. async function sendGitHubDelivery(origin: string): Promise<Response> {
  308. const body = JSON.stringify({
  309. action: 'ready_for_review',
  310. number: 4242,
  311. repository: { full_name: 'deepseek-harness/deepseek-harness' },
  312. pull_request: {
  313. title: 'Real CLI webhook e2e',
  314. html_url: 'https://github.com/deepseek-harness/deepseek-harness/pull/4242',
  315. draft: false,
  316. user: { login: 'octocat' },
  317. base: { ref: 'master', sha: 'aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa' },
  318. head: { ref: 'webhook-e2e', sha: 'bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb' },
  319. },
  320. })
  321. const signature = `sha256=${createHmac('sha256', SECRET).update(body).digest('hex')}`
  322. return await fetch(`${origin}/github`, {
  323. method: 'POST',
  324. headers: {
  325. 'content-type': 'application/json',
  326. 'x-github-delivery': DELIVERY,
  327. 'x-github-event': 'pull_request',
  328. 'x-hub-signature-256': signature,
  329. },
  330. body,
  331. })
  332. }
  333. describe.skipIf(!process.env.DEEPSEEK_API_KEY)('GitHub webhook through the real dsh CLI and model', () => {
  334. it('creates, attaches, prompts, and completes a Workspace Session', async () => {
  335. expect(existsSync(BUILT_BIN), `missing built CLI ${BUILT_BIN}; run pnpm run build:official`).toBe(true)
  336. const root = await mkdtemp(join(tmpdir(), 'dsh-github-webhook-real-'))
  337. const workspacePath = join(root, 'workspace')
  338. await mkdir(workspacePath)
  339. const canonicalWorkspacePath = await realpath(workspacePath)
  340. const webhookPort = await freePort()
  341. const child = spawn(process.execPath, [
  342. BUILT_BIN,
  343. 'web',
  344. '--patch', OVERLAY,
  345. '--no-open',
  346. '--host', '127.0.0.1',
  347. '--port', '0',
  348. ], {
  349. cwd: root,
  350. env: {
  351. ...process.env,
  352. DSH_AGENTS_HOME: join(root, '.agents'),
  353. DSH_GITHUB_E2E_MARKER: MARKER,
  354. DSH_GITHUB_E2E_WORKSPACE: workspacePath,
  355. DSH_GITHUB_WEBHOOK_PORT: String(webhookPort),
  356. DSH_GITHUB_WEBHOOK_SECRET: SECRET,
  357. DSH_HOME: join(root, '.dsh'),
  358. DSH_TELEMETRY_DISABLED: '1',
  359. },
  360. stdio: ['ignore', 'pipe', 'pipe'],
  361. })
  362. const observation = observeProcess(child)
  363. try {
  364. const baseUrl = await observation.ready
  365. const webhookOrigin = `http://127.0.0.1:${String(webhookPort)}`
  366. expect((await fetch(`${webhookOrigin}/api`)).status).toBe(404)
  367. expect((await sendGitHubDelivery(new URL(baseUrl).origin)).status).not.toBe(202)
  368. expect((await sendGitHubDelivery(webhookOrigin)).status).toBe(202)
  369. const workspaces = await eventually(
  370. child,
  371. observation.text,
  372. 'one Workspace-attached Session',
  373. async () => await workspaceBaseline(baseUrl),
  374. value => value.items.some(workspace =>
  375. workspace.path === canonicalWorkspacePath && workspace.sessionIds.length === 1),
  376. 30_000,
  377. )
  378. const workspace = workspaces.items.find(item => item.path === canonicalWorkspacePath)
  379. const sessionId = workspace?.sessionIds[0]
  380. if (sessionId === undefined) throw new Error('workspace/follow did not expose the webhook Session')
  381. const sessions = await remoteRpc<SessionList>(baseUrl, 'session/list', { _request: {} })
  382. expect(sessions.items.find(session => session.sessionId === sessionId)).toMatchObject({
  383. blank: false,
  384. cwd: canonicalWorkspacePath,
  385. projections: { values: { agentPreset: 'minimal' } },
  386. })
  387. const admitted = await eventually(
  388. child,
  389. observation.text,
  390. 'webhook provenance, title, and permission events',
  391. async () => await history(baseUrl, sessionId),
  392. (page) => {
  393. const events = historyEvents(page)
  394. const title = events.find(event => event.type === 'session/title')
  395. const permission = events.find(event =>
  396. event.type === 'permission/preset'
  397. && isRecord(event.data)
  398. && event.data.preset === 'read-only')
  399. const message = events.find(event =>
  400. event.type === 'user/message'
  401. && isRecord(event.data)
  402. && isRecord(event.data.source)
  403. && event.data.source.kind === 'webhook')
  404. return isRecord(title?.data) && title.data.title === TITLE
  405. && permission !== undefined
  406. && isRecord(message?.data) && isRecord(message.data.source)
  407. && message.data.source.provider === 'github'
  408. && message.data.source.deliveryId === DELIVERY
  409. },
  410. 30_000,
  411. )
  412. const webhookMessage = historyEvents(admitted)
  413. .find(event => event.type === 'user/message'
  414. && isRecord(event.data)
  415. && isRecord(event.data.source)
  416. && event.data.source.kind === 'webhook')
  417. expect(webhookMessage?.data).toMatchObject({
  418. content: [{ type: 'text', text: `Reply with exactly ${MARKER} and no other text. Do not call tools.` }],
  419. source: {
  420. kind: 'webhook',
  421. provider: 'github',
  422. deliveryId: DELIVERY,
  423. ruleId: 'github-real-e2e',
  424. source: 'github-real-e2e',
  425. },
  426. })
  427. const completed = await eventually(
  428. child,
  429. observation.text,
  430. 'a real DeepSeek assistant response',
  431. async () => await history(baseUrl, sessionId),
  432. page => assistantText(page).includes(MARKER),
  433. 150_000,
  434. )
  435. expect(assistantText(completed)).toContain(MARKER)
  436. } finally {
  437. await stop(child)
  438. await rm(root, { recursive: true, force: true })
  439. }
  440. }, 330_000)
  441. })