github-webhook-real.e2e.ts 17 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467
  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 { describe, expect, it } from 'vitest'
  14. import WebSocket from 'ws'
  15. const REPO_ROOT = fileURLToPath(new URL('../../..', import.meta.url))
  16. const BUILT_BIN = join(REPO_ROOT, 'apps/cli/lib/bin.js')
  17. const OVERLAY = fileURLToPath(new URL(
  18. './fixtures/github-webhook/cordis.yml',
  19. import.meta.url,
  20. ))
  21. const SECRET = 'github-webhook-real-e2e-secret'
  22. const DELIVERY = 'github-webhook-real-e2e-delivery'
  23. const MARKER = 'DSH_GITHUB_WEBHOOK_REAL_E2E_OK'
  24. const TITLE = 'GitHub webhook real e2e'
  25. const authenticatedCookies = new Map<string, Promise<{ origin: string; cookie: string }>>()
  26. /** Exchange the printed process token once for Node-side API probes. */
  27. function authenticatedWeb(launchUrl: string): Promise<{ origin: string; cookie: string }> {
  28. const existing = authenticatedCookies.get(launchUrl)
  29. if (existing !== undefined) return existing
  30. const exchange = (async () => {
  31. const response = await fetch(launchUrl, { redirect: 'manual' })
  32. const setCookie = response.headers.get('set-cookie')
  33. if (response.status !== 303 || setCookie === null) {
  34. throw new Error(`dsh web authentication returned HTTP ${String(response.status)}`)
  35. }
  36. return { origin: new URL(launchUrl).origin, cookie: setCookie.split(';', 1)[0]! }
  37. })()
  38. authenticatedCookies.set(launchUrl, exchange)
  39. return exchange
  40. }
  41. interface SessionList {
  42. items: Array<{
  43. sessionId: string
  44. cwd?: string
  45. blank: boolean
  46. projections?: { values: { agentPreset?: string | null } }
  47. }>
  48. }
  49. interface WorkspaceBaseline {
  50. items: Array<{
  51. path: string
  52. sessionIds: string[]
  53. }>
  54. }
  55. interface HistoryPage {
  56. records: Array<{ type: 'event'; event: HistoryEvent }>
  57. hasMore: boolean
  58. }
  59. interface HistoryEvent {
  60. type: string
  61. data: unknown
  62. }
  63. interface ProcessObservation {
  64. readonly ready: Promise<string>
  65. readonly text: () => string
  66. }
  67. function isRecord(value: unknown): value is Record<string, unknown> {
  68. return typeof value === 'object' && value !== null
  69. }
  70. /** Capture bounded process output and resolve the public Web URL after settled boot. */
  71. function observeProcess(child: ChildProcess): ProcessObservation {
  72. let output = ''
  73. let settled = false
  74. let resolveReady!: (url: string) => void
  75. let rejectReady!: (error: Error) => void
  76. const ready = new Promise<string>((resolve, reject) => {
  77. resolveReady = resolve
  78. rejectReady = reject
  79. })
  80. const timer = setTimeout(() => {
  81. if (!settled) rejectReady(new Error(`dsh web did not become ready within 90s:\n${output}`))
  82. }, 90_000)
  83. timer.unref()
  84. const append = (chunk: Buffer | string): void => {
  85. output = `${output}${String(chunk)}`.slice(-100_000)
  86. const match = /dsh web: (http:\/\/[^\s]+)/u.exec(output)
  87. if (settled || match?.[1] === undefined) return
  88. settled = true
  89. clearTimeout(timer)
  90. resolveReady(match[1].replace('0.0.0.0', '127.0.0.1'))
  91. }
  92. child.stdout?.on('data', append)
  93. child.stderr?.on('data', append)
  94. child.once('error', (error) => {
  95. if (!settled) rejectReady(error)
  96. })
  97. child.once('exit', (code) => {
  98. if (!settled) rejectReady(new Error(`dsh web exited before readiness (code ${String(code)}):\n${output}`))
  99. })
  100. return { ready, text: () => output }
  101. }
  102. /** Reserve and release one loopback port for the isolated webhook listener. */
  103. async function freePort(): Promise<number> {
  104. const server = createServer()
  105. await new Promise<void>((resolve, reject) => {
  106. server.once('error', reject)
  107. server.listen(0, '127.0.0.1', resolve)
  108. })
  109. const port = (server.address() as AddressInfo).port
  110. await new Promise<void>((resolve, reject) => {
  111. server.close((error) => {
  112. if (error === undefined) resolve()
  113. else reject(error)
  114. })
  115. })
  116. return port
  117. }
  118. /** Invoke one public Remote method over its HTTP carrier. */
  119. async function remoteRpc<T>(baseUrl: string, endpoint: string, args: object): Promise<T> {
  120. const authenticated = await authenticatedWeb(baseUrl)
  121. const response = await fetch(`${authenticated.origin}/api/${endpoint}`, {
  122. method: 'POST',
  123. headers: { 'content-type': 'application/json', cookie: authenticated.cookie },
  124. body: JSON.stringify({
  125. type: 'client-request',
  126. rpcId: `github-webhook-real-${endpoint}-${randomUUID()}`,
  127. method: endpoint,
  128. payload: { args },
  129. }),
  130. })
  131. if (!response.ok) {
  132. throw new Error(`${endpoint} returned HTTP ${String(response.status)}: ${await response.text()}`)
  133. }
  134. const envelope = await response.json() as {
  135. result: { ok: true; value: T } | { ok: false; error: { code: string; message: string } }
  136. }
  137. if (!envelope.result.ok) {
  138. throw new Error(`${endpoint} failed: ${envelope.result.error.code}: ${envelope.result.error.message}`)
  139. }
  140. return envelope.result.value
  141. }
  142. /** Read one opening item from a public Remote stream. */
  143. async function openingStreamItem(
  144. baseUrl: string,
  145. endpoint: string,
  146. args: object,
  147. accepts: (value: unknown) => boolean,
  148. ): Promise<Record<string, unknown>> {
  149. const authenticated = await authenticatedWeb(baseUrl)
  150. const socket = new WebSocket(`${authenticated.origin.replace(/^http/u, 'ws')}/api/remote.mux`, {
  151. headers: { cookie: authenticated.cookie },
  152. })
  153. const streamId = `github-webhook-real-${endpoint}-${randomUUID()}`
  154. try {
  155. await new Promise<void>((resolve, reject) => {
  156. const cleanup = (): void => {
  157. socket.removeEventListener('open', opened)
  158. socket.removeEventListener('error', failed)
  159. socket.removeEventListener('close', closed)
  160. }
  161. const opened = (): void => {
  162. cleanup()
  163. resolve()
  164. }
  165. const failed = (): void => {
  166. cleanup()
  167. reject(new Error(`${endpoint} carrier failed before opening`))
  168. }
  169. const closed = (): void => {
  170. cleanup()
  171. reject(new Error(`${endpoint} carrier closed before opening`))
  172. }
  173. socket.addEventListener('open', opened)
  174. socket.addEventListener('error', failed)
  175. socket.addEventListener('close', closed)
  176. })
  177. return await new Promise<Record<string, unknown>>((resolve, reject) => {
  178. const timer = setTimeout(() => { finish(new Error(`${endpoint} did not publish its opening item`)) }, 10_000)
  179. const cleanup = (): void => {
  180. clearTimeout(timer)
  181. socket.removeEventListener('message', message)
  182. socket.removeEventListener('error', failed)
  183. socket.removeEventListener('close', closed)
  184. }
  185. const finish = (error: Error | undefined, value?: Record<string, unknown>): void => {
  186. cleanup()
  187. if (error !== undefined) reject(error)
  188. else if (value === undefined) reject(new Error(`${endpoint} opening item was absent`))
  189. else resolve(value)
  190. }
  191. const message = (event: WebSocket.MessageEvent): void => {
  192. try {
  193. const text = typeof event.data === 'string'
  194. ? event.data
  195. : Buffer.isBuffer(event.data) ? event.data.toString('utf8') : undefined
  196. if (text === undefined) throw new Error(`${endpoint} published a non-text frame`)
  197. const frame: unknown = JSON.parse(text)
  198. if (!isRecord(frame) || frame.streamId !== streamId) return
  199. if (frame.type === 'error') {
  200. finish(new Error(`${endpoint} failed: ${JSON.stringify(frame.error)}`))
  201. return
  202. }
  203. if (frame.type === 'end') {
  204. finish(new Error(`${endpoint} ended before its opening item`))
  205. return
  206. }
  207. if (frame.type === 'item' && isRecord(frame.value) && accepts(frame.value)) {
  208. finish(undefined, frame.value)
  209. }
  210. } catch (error) {
  211. finish(error instanceof Error ? error : new Error(String(error)))
  212. }
  213. }
  214. const failed = (): void => { finish(new Error(`${endpoint} carrier failed before its opening item`)) }
  215. const closed = (): void => { finish(new Error(`${endpoint} carrier closed before its opening item`)) }
  216. socket.addEventListener('message', message)
  217. socket.addEventListener('error', failed)
  218. socket.addEventListener('close', closed)
  219. socket.send(JSON.stringify({ type: 'open', streamId, endpoint, payload: { args } }))
  220. })
  221. } finally {
  222. socket.close()
  223. }
  224. }
  225. /** Read the current Workspace baseline from a fresh follow generation. */
  226. async function workspaceBaseline(baseUrl: string): Promise<WorkspaceBaseline> {
  227. const frame = await openingStreamItem(
  228. baseUrl,
  229. 'workspace/follow',
  230. {},
  231. value => isRecord(value) && value.type === 'baseline' && isRecord(value.value),
  232. )
  233. return frame.value as WorkspaceBaseline
  234. }
  235. /** Read the complete opening page from a fresh Session follow generation. */
  236. async function history(baseUrl: string, sessionId: string): Promise<HistoryPage> {
  237. const frame = await openingStreamItem(
  238. baseUrl,
  239. 'session/follow',
  240. { request: { address: { kind: 'session', sessionId }, maxMessages: 100 } },
  241. value => isRecord(value)
  242. && value.type === 'snapshot'
  243. && Array.isArray(value.records)
  244. && typeof value.hasMore === 'boolean',
  245. )
  246. return { records: frame.records as HistoryPage['records'], hasMore: frame.hasMore as boolean }
  247. }
  248. /** Poll a public observation until it satisfies the test's behavior predicate. */
  249. async function eventually<T>(
  250. child: ChildProcess,
  251. processOutput: () => string,
  252. label: string,
  253. probe: () => Promise<T>,
  254. accepts: (value: T) => boolean,
  255. timeoutMs: number,
  256. ): Promise<T> {
  257. const deadline = Date.now() + timeoutMs
  258. let lastValue: T | undefined
  259. let lastError: unknown
  260. while (Date.now() < deadline) {
  261. if (child.exitCode !== null) {
  262. throw new Error(`dsh web exited while waiting for ${label} (code ${String(child.exitCode)}):\n${processOutput()}`)
  263. }
  264. try {
  265. lastValue = await probe()
  266. if (accepts(lastValue)) return lastValue
  267. } catch (error) {
  268. lastError = error
  269. }
  270. await delay(300)
  271. }
  272. throw new Error(
  273. `timed out waiting for ${label}; last value=${JSON.stringify(lastValue)}; `
  274. + `last error=${String(lastError)}; process output:\n${processOutput()}`,
  275. )
  276. }
  277. /** Return every text block from durable assistant messages. */
  278. function assistantText(page: HistoryPage): string {
  279. const text: string[] = []
  280. for (const event of historyEvents(page)) {
  281. if (event.type !== 'assistant/message' || !isRecord(event.data) || !isRecord(event.data.message)) continue
  282. const content = event.data.message.content
  283. if (!Array.isArray(content)) continue
  284. for (const block of content) {
  285. if (isRecord(block) && block.type === 'text' && typeof block.text === 'string') text.push(block.text)
  286. }
  287. }
  288. return text.join('\n')
  289. }
  290. /** Read scalar v2 history records for assertions over the public event stream. */
  291. function historyEvents(page: HistoryPage): HistoryEvent[] {
  292. return page.records.map(record => record.event)
  293. }
  294. /** Stop the spawned CLI through its normal signal path, escalating only on a stuck teardown. */
  295. async function stop(child: ChildProcess): Promise<void> {
  296. if (child.exitCode !== null) return
  297. let resolveClosed!: () => void
  298. const closed = new Promise<void>((resolve) => { resolveClosed = resolve })
  299. child.once('close', resolveClosed)
  300. child.kill('SIGTERM')
  301. if (await Promise.race([closed.then(() => true), delay(10_000, false, { ref: false })])) return
  302. if (child.exitCode === null) child.kill('SIGKILL')
  303. await Promise.race([closed, delay(5_000, undefined, { ref: false })])
  304. }
  305. /** Send the sole synthetic external interaction: one signed GitHub delivery. */
  306. async function sendGitHubDelivery(origin: string): Promise<Response> {
  307. const body = JSON.stringify({
  308. action: 'ready_for_review',
  309. number: 4242,
  310. repository: { full_name: 'deepseek-harness/deepseek-harness' },
  311. pull_request: {
  312. title: 'Real CLI webhook e2e',
  313. html_url: 'https://github.com/deepseek-harness/deepseek-harness/pull/4242',
  314. draft: false,
  315. user: { login: 'octocat' },
  316. base: { ref: 'master', sha: 'aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa' },
  317. head: { ref: 'webhook-e2e', sha: 'bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb' },
  318. },
  319. })
  320. const signature = `sha256=${createHmac('sha256', SECRET).update(body).digest('hex')}`
  321. return await fetch(`${origin}/github`, {
  322. method: 'POST',
  323. headers: {
  324. 'content-type': 'application/json',
  325. 'x-github-delivery': DELIVERY,
  326. 'x-github-event': 'pull_request',
  327. 'x-hub-signature-256': signature,
  328. },
  329. body,
  330. })
  331. }
  332. describe.skipIf(!process.env.DEEPSEEK_API_KEY)('GitHub webhook through the real dsh CLI and model', () => {
  333. it('creates, attaches, prompts, and completes a Workspace Session', async () => {
  334. expect(existsSync(BUILT_BIN), `missing built CLI ${BUILT_BIN}; run pnpm run build:official`).toBe(true)
  335. const root = await mkdtemp(join(tmpdir(), 'dsh-github-webhook-real-'))
  336. const workspacePath = join(root, 'workspace')
  337. await mkdir(workspacePath)
  338. const canonicalWorkspacePath = await realpath(workspacePath)
  339. const webhookPort = await freePort()
  340. const child = spawn(process.execPath, [
  341. BUILT_BIN,
  342. 'web',
  343. '--patch', OVERLAY,
  344. '--no-open',
  345. '--host', '127.0.0.1',
  346. '--port', '0',
  347. ], {
  348. cwd: root,
  349. env: {
  350. ...process.env,
  351. DSH_AGENTS_HOME: join(root, '.agents'),
  352. DSH_GITHUB_E2E_MARKER: MARKER,
  353. DSH_GITHUB_E2E_WORKSPACE: workspacePath,
  354. DSH_GITHUB_WEBHOOK_PORT: String(webhookPort),
  355. DSH_GITHUB_WEBHOOK_SECRET: SECRET,
  356. DSH_HOME: join(root, '.dsh'),
  357. DSH_TELEMETRY_DISABLED: '1',
  358. },
  359. stdio: ['ignore', 'pipe', 'pipe'],
  360. })
  361. const observation = observeProcess(child)
  362. try {
  363. const baseUrl = await observation.ready
  364. const webhookOrigin = `http://127.0.0.1:${String(webhookPort)}`
  365. expect((await fetch(`${webhookOrigin}/api`)).status).toBe(404)
  366. expect((await sendGitHubDelivery(new URL(baseUrl).origin)).status).not.toBe(202)
  367. expect((await sendGitHubDelivery(webhookOrigin)).status).toBe(202)
  368. const workspaces = await eventually(
  369. child,
  370. observation.text,
  371. 'one Workspace-attached Session',
  372. async () => await workspaceBaseline(baseUrl),
  373. value => value.items.some(workspace =>
  374. workspace.path === canonicalWorkspacePath && workspace.sessionIds.length === 1),
  375. 30_000,
  376. )
  377. const workspace = workspaces.items.find(item => item.path === canonicalWorkspacePath)
  378. const sessionId = workspace?.sessionIds[0]
  379. if (sessionId === undefined) throw new Error('workspace/follow did not expose the webhook Session')
  380. const sessions = await remoteRpc<SessionList>(baseUrl, 'session/list', { _request: {} })
  381. expect(sessions.items.find(session => session.sessionId === sessionId)).toMatchObject({
  382. blank: false,
  383. cwd: canonicalWorkspacePath,
  384. projections: { values: { agentPreset: 'minimal' } },
  385. })
  386. const admitted = await eventually(
  387. child,
  388. observation.text,
  389. 'webhook provenance, title, and permission events',
  390. async () => await history(baseUrl, sessionId),
  391. (page) => {
  392. const events = historyEvents(page)
  393. const title = events.find(event => event.type === 'session/title')
  394. const permission = events.find(event =>
  395. event.type === 'permission/preset'
  396. && isRecord(event.data)
  397. && event.data.preset === 'read-only')
  398. const message = events.find(event =>
  399. event.type === 'user/message'
  400. && isRecord(event.data)
  401. && isRecord(event.data.source)
  402. && event.data.source.kind === 'webhook')
  403. return isRecord(title?.data) && title.data.title === TITLE
  404. && permission !== undefined
  405. && isRecord(message?.data) && isRecord(message.data.source)
  406. && message.data.source.provider === 'github'
  407. && message.data.source.deliveryId === DELIVERY
  408. },
  409. 30_000,
  410. )
  411. const webhookMessage = historyEvents(admitted)
  412. .find(event => event.type === 'user/message'
  413. && isRecord(event.data)
  414. && isRecord(event.data.source)
  415. && event.data.source.kind === 'webhook')
  416. expect(webhookMessage?.data).toMatchObject({
  417. content: [{ type: 'text', text: `Reply with exactly ${MARKER} and no other text. Do not call tools.` }],
  418. source: {
  419. kind: 'webhook',
  420. provider: 'github',
  421. deliveryId: DELIVERY,
  422. ruleId: 'github-real-e2e',
  423. source: 'github-real-e2e',
  424. },
  425. })
  426. const completed = await eventually(
  427. child,
  428. observation.text,
  429. 'a real DeepSeek assistant response',
  430. async () => await history(baseUrl, sessionId),
  431. page => assistantText(page).includes(MARKER),
  432. 150_000,
  433. )
  434. expect(assistantText(completed)).toContain(MARKER)
  435. } finally {
  436. await stop(child)
  437. await rm(root, { recursive: true, force: true })
  438. }
  439. }, 330_000)
  440. })