keyless-smoke.e2e.ts 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310
  1. import { createServer } from 'node:http'
  2. import { mkdtemp, readFile, readdir, rm } from 'node:fs/promises'
  3. import { tmpdir } from 'node:os'
  4. import { join } from 'node:path'
  5. import { fileURLToPath } from 'node:url'
  6. import { promisify } from 'node:util'
  7. import { zstdDecompress } from 'node:zlib'
  8. import { execa } from 'execa'
  9. import { describe, expect, it } from 'vitest'
  10. const binScript = fileURLToPath(new URL('../../../apps/cli/src/bin.ts', import.meta.url))
  11. const patchPath = fileURLToPath(new URL('./keyless.patch.yml', import.meta.url))
  12. const repoRoot = fileURLToPath(new URL('../../..', import.meta.url))
  13. const decompress = promisify(zstdDecompress)
  14. function waitForLine(
  15. lines: string[],
  16. predicate: (value: Record<string, unknown>) => boolean,
  17. stderr: () => string,
  18. ): Promise<Record<string, unknown>> {
  19. return new Promise((resolve, reject) => {
  20. const deadline = Date.now() + 30_000
  21. const poll = (): void => {
  22. while (lines.length > 0) {
  23. const line = lines.shift()!
  24. if (!line.trim()) continue
  25. try {
  26. const value = JSON.parse(line) as Record<string, unknown>
  27. if (predicate(value)) {
  28. resolve(value)
  29. return
  30. }
  31. } catch {
  32. reject(new Error(`non-JSON stdout from JSON-RPC agent runtime: ${line}`))
  33. return
  34. }
  35. }
  36. if (Date.now() >= deadline) {
  37. reject(new Error(`timed out waiting for JSON-RPC response; stderr=${stderr()}`))
  38. return
  39. }
  40. setTimeout(poll, 10)
  41. }
  42. poll()
  43. })
  44. }
  45. describe('Python SDK dsh profile keyless smoke', () => {
  46. it.each([
  47. { label: 'reports max-token turns with the default mapping config', envValue: undefined },
  48. { label: 'reports max-token turns with mapping enabled through env', envValue: 'true' },
  49. { label: 'reports max-token turns with mapping disabled through env', envValue: 'false' },
  50. ])('$label', async ({ envValue }) => {
  51. const root = await mkdtemp(join(tmpdir(), 'dsh-python-sdk-runtime-smoke-'))
  52. const modelRequests: Record<string, unknown>[] = []
  53. const modelServer = createServer((request, response) => {
  54. let body = ''
  55. request.setEncoding('utf8')
  56. request.on('data', (chunk: string) => { body += chunk })
  57. request.on('end', () => {
  58. modelRequests.push(JSON.parse(body) as Record<string, unknown>)
  59. response.writeHead(200, { 'content-type': 'text/event-stream' })
  60. response.write('data: {"choices":[{"delta":{"role":"assistant","content":null}}]}\n\n')
  61. response.write('data: {"choices":[{"delta":{"content":"done"}}]}\n\n')
  62. response.write('data: {"choices":[{"delta":{},"finish_reason":"length"}],"usage":{"prompt_tokens":3,"completion_tokens":1}}\n\n')
  63. response.end('data: [DONE]\n\n')
  64. })
  65. })
  66. await new Promise<void>(resolve => modelServer.listen(0, '127.0.0.1', resolve))
  67. const address = modelServer.address()
  68. if (address === null || typeof address === 'string') throw new Error('model server did not bind a TCP port')
  69. // The line-predicate protocol driving below is the genuinely custom part;
  70. // execa owns spawn, the deadline, and exit settlement around it.
  71. const child = execa(process.execPath, [
  72. '--import',
  73. 'tsx/esm',
  74. binScript,
  75. '--profile',
  76. 'sdk',
  77. '--patch',
  78. patchPath,
  79. ], {
  80. cwd: repoRoot,
  81. env: {
  82. DSH_HOME: join(root, '.dsh'),
  83. DSH_PERMISSION_MODE: 'danger-full-access',
  84. DSH_TELEMETRY_DISABLED: '1',
  85. DEEPSEEK_API_KEY: 'keyless-smoke-no-call',
  86. DEEPSEEK_BASE_URL: `http://127.0.0.1:${address.port}`,
  87. ...(envValue === undefined ? {} : { DSH_MAX_TOKENS_AS_SUCCESS: envValue }),
  88. },
  89. timeout: 35_000,
  90. killSignal: 'SIGKILL',
  91. reject: false,
  92. })
  93. const lines: string[] = []
  94. let stdoutBuffer = ''
  95. let stderr = ''
  96. child.stdout.on('data', (chunk: Buffer) => {
  97. stdoutBuffer += chunk.toString('utf8')
  98. const parts = stdoutBuffer.split('\n')
  99. stdoutBuffer = parts.pop() ?? ''
  100. lines.push(...parts)
  101. })
  102. child.stderr.on('data', (chunk: Buffer) => { stderr += chunk.toString('utf8') })
  103. try {
  104. child.stdin.write(`${JSON.stringify({
  105. jsonrpc: '2.0',
  106. id: 1,
  107. method: 'initialize',
  108. params: { cwd: root, provider: 'deepseek-official', model: 'deepseek-v4-pro', maxTokens: 1234 },
  109. })}\n`)
  110. const initialized = await waitForLine(lines, value => value.id === 1, () => stderr)
  111. expect(initialized).toMatchObject({
  112. jsonrpc: '2.0',
  113. id: 1,
  114. result: { serverInfo: { name: 'deepseek-harness-sdk-runtime' } },
  115. })
  116. child.stdin.write(`${JSON.stringify({
  117. jsonrpc: '2.0',
  118. id: 2,
  119. method: 'session/prompt',
  120. params: { sessionId: 'main', contentBlocks: [{ type: 'text', text: 'inspect tools' }] },
  121. })}\n`)
  122. const prompt = await waitForLine(lines, value => value.id === 2, () => stderr)
  123. expect(prompt).toMatchObject({
  124. jsonrpc: '2.0',
  125. id: 2,
  126. result: { messageId: expect.any(String) as unknown },
  127. })
  128. const turnEnd = await waitForLine(lines, (value) => {
  129. if (value.method !== 'session.event') return false
  130. const params = value.params as Record<string, unknown> | undefined
  131. const event = params?.event as Record<string, unknown> | undefined
  132. return params?.sessionId === 'main' && event?.type === 'turn/end'
  133. }, () => stderr)
  134. expect(turnEnd).toMatchObject({
  135. jsonrpc: '2.0',
  136. method: 'session.event',
  137. params: {
  138. sessionId: 'main',
  139. event: {
  140. type: 'turn/end',
  141. data: { reason: { kind: 'max-tokens' } },
  142. },
  143. },
  144. })
  145. const tools = modelRequests[0]?.tools as { function?: { name?: string } }[]
  146. expect(modelRequests[0]?.max_tokens).toBe(1234)
  147. expect(tools.map(tool => tool.function?.name).sort()).toEqual([
  148. 'bash',
  149. 'edit',
  150. 'read',
  151. 'read_image',
  152. 'subagent',
  153. 'todo_write',
  154. 'write',
  155. ])
  156. child.stdin.write(`${JSON.stringify({ jsonrpc: '2.0', id: 3, method: 'shutdown' })}\n`)
  157. const shutdown = await waitForLine(lines, value => value.id === 3, () => stderr)
  158. expect(shutdown).toMatchObject({ jsonrpc: '2.0', id: 3, result: {} })
  159. const exit = await child
  160. expect(exit.exitCode, `signal=${String(exit.signal)}; stderr=${stderr}`).toBe(0)
  161. const sessionsRoot = join(root, '.dsh', 'sessions')
  162. const files = await readdir(sessionsRoot, { recursive: true })
  163. const log = files.find(file => file.endsWith('.jsonl.zstd'))
  164. expect(log).toBeDefined()
  165. const compressed = await readFile(join(sessionsRoot, log!))
  166. expect(compressed.subarray(0, 4).toString('hex')).toBe('28b52ffd')
  167. expect(JSON.parse((await decompress(compressed)).toString())).toMatchObject({ type: 'session', id: 'main' })
  168. } finally {
  169. // No-op after exit; reject: false settles on every outcome, so cleanup never races teardown.
  170. child.kill('SIGKILL')
  171. await child
  172. await new Promise<void>(resolve => modelServer.close(() => { resolve() }))
  173. await rm(root, { recursive: true, force: true })
  174. }
  175. }, 40_000)
  176. it('boots the standalone minimal profile with its exact model-facing roster', async () => {
  177. const root = await mkdtemp(join(tmpdir(), 'dsh-python-sdk-minimal-'))
  178. const modelRequests: Record<string, unknown>[] = []
  179. const modelServer = createServer((request, response) => {
  180. let body = ''
  181. request.setEncoding('utf8')
  182. request.on('data', (chunk: string) => { body += chunk })
  183. request.on('end', () => {
  184. modelRequests.push(JSON.parse(body) as Record<string, unknown>)
  185. response.writeHead(200, { 'content-type': 'text/event-stream' })
  186. response.write('data: {"choices":[{"delta":{"role":"assistant","content":null}}]}\n\n')
  187. response.write('data: {"choices":[{"delta":{"content":"done"}}]}\n\n')
  188. response.write('data: {"choices":[{"delta":{},"finish_reason":"stop"}],"usage":{"prompt_tokens":3,"completion_tokens":1}}\n\n')
  189. response.end('data: [DONE]\n\n')
  190. })
  191. })
  192. await new Promise<void>(resolve => modelServer.listen(0, '127.0.0.1', resolve))
  193. const address = modelServer.address()
  194. if (address === null || typeof address === 'string') throw new Error('model server did not bind a TCP port')
  195. const child = execa(process.execPath, [
  196. '--import',
  197. 'tsx/esm',
  198. binScript,
  199. '--profile',
  200. 'sdk-minimal',
  201. ], {
  202. cwd: repoRoot,
  203. env: {
  204. DSH_HOME: join(root, '.dsh'),
  205. DSH_SYSTEM_PROMPT: 'Minimal allowlist prompt.',
  206. DEEPSEEK_API_KEY: 'keyless-smoke-no-call',
  207. DEEPSEEK_BASE_URL: `http://127.0.0.1:${address.port}`,
  208. },
  209. timeout: 35_000,
  210. killSignal: 'SIGKILL',
  211. reject: false,
  212. })
  213. const lines: string[] = []
  214. let stdoutBuffer = ''
  215. let stderr = ''
  216. child.stdout.on('data', (chunk: Buffer) => {
  217. stdoutBuffer += chunk.toString('utf8')
  218. const parts = stdoutBuffer.split('\n')
  219. stdoutBuffer = parts.pop() ?? ''
  220. lines.push(...parts)
  221. })
  222. child.stderr.on('data', (chunk: Buffer) => { stderr += chunk.toString('utf8') })
  223. try {
  224. child.stdin.write(`${JSON.stringify({
  225. jsonrpc: '2.0',
  226. id: 1,
  227. method: 'initialize',
  228. params: { cwd: root, provider: 'deepseek-official', model: 'deepseek-v4-pro' },
  229. })}\n`)
  230. await waitForLine(lines, value => value.id === 1, () => stderr)
  231. child.stdin.write(`${JSON.stringify({
  232. jsonrpc: '2.0',
  233. id: 2,
  234. method: 'session/prompt',
  235. params: { sessionId: 'minimal', contentBlocks: [{ type: 'text', text: 'inspect tools' }] },
  236. })}\n`)
  237. await waitForLine(lines, (value) => {
  238. const params = value.params as Record<string, unknown> | undefined
  239. const event = params?.event as Record<string, unknown> | undefined
  240. return params?.sessionId === 'minimal' && event?.type === 'turn/end'
  241. }, () => stderr)
  242. const request = modelRequests[0] as {
  243. messages?: Array<{ role?: string; content?: unknown }>
  244. tools?: Array<{ function?: { name?: string } }>
  245. }
  246. expect(request.messages?.[0]).toMatchObject({ role: 'system', content: 'Minimal allowlist prompt.' })
  247. expect(request.tools?.map(tool => tool.function?.name).sort()).toEqual(['bash', 'str_replace_editor'])
  248. const profile = JSON.parse(
  249. await readFile(join(root, '.dsh', 'profiles', 'sdk-minimal', 'package.json'), 'utf8'),
  250. ) as { dsh?: { profile?: { bundles?: string[]; patchReload?: string } } }
  251. expect(profile.dsh?.profile).toEqual({
  252. bundles: ['@deepseek-ai/dsh-sdk-minimal'],
  253. patchReload: 'startup',
  254. })
  255. child.stdin.write(`${JSON.stringify({ jsonrpc: '2.0', id: 3, method: 'shutdown' })}\n`)
  256. await waitForLine(lines, value => value.id === 3, () => stderr)
  257. const exit = await child
  258. expect(exit.exitCode, `signal=${String(exit.signal)}; stderr=${stderr}`).toBe(0)
  259. } finally {
  260. child.kill('SIGKILL')
  261. await child
  262. await new Promise<void>(resolve => modelServer.close(() => { resolve() }))
  263. await rm(root, { recursive: true, force: true })
  264. }
  265. }, 40_000)
  266. it('rejects an invalid max-token success env value', async () => {
  267. const root = await mkdtemp(join(tmpdir(), 'dsh-python-sdk-runtime-invalid-'))
  268. try {
  269. const { exitCode, stdout, stderr } = await execa(process.execPath, [
  270. '--import',
  271. 'tsx/esm',
  272. binScript,
  273. '--profile',
  274. 'sdk',
  275. '--patch',
  276. patchPath,
  277. ], {
  278. cwd: repoRoot,
  279. env: {
  280. DSH_HOME: join(root, '.dsh'),
  281. DEEPSEEK_API_KEY: 'keyless-smoke-no-call',
  282. DSH_MAX_TOKENS_AS_SUCCESS: 'sometimes',
  283. },
  284. stdin: 'ignore',
  285. timeout: 25_000,
  286. killSignal: 'SIGKILL',
  287. reject: false,
  288. })
  289. expect(exitCode, stderr).toBe(1)
  290. expect(stdout).toBe('')
  291. expect(stderr).toContain('plugin tree failed to load')
  292. expect(stderr).toContain('failed to apply loader entry sdk-jsonrpc-server (@deepseek-ai/dsh-sdk-jsonrpc-server)')
  293. expect(stderr).toContain('sometimes')
  294. } finally {
  295. await rm(root, { recursive: true, force: true })
  296. }
  297. }, 30_000)
  298. })