@tangle-network/agent-eval 0.123.8 → 0.125.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (61) hide show
  1. package/CHANGELOG.md +29 -4
  2. package/dist/analyst/index.d.ts +5 -0
  3. package/dist/analyst/index.js +4 -4
  4. package/dist/benchmarks/index.js +4 -4
  5. package/dist/campaign/index.d.ts +20 -1
  6. package/dist/campaign/index.js +4 -4
  7. package/dist/{chunk-ZU3QWGZE.js → chunk-A62YMFWA.js} +89 -8
  8. package/dist/chunk-A62YMFWA.js.map +1 -0
  9. package/dist/{chunk-BUR5R4R4.js → chunk-A6GT67HT.js} +3 -3
  10. package/dist/{chunk-EEHHBAWH.js → chunk-DT7OXY3C.js} +2 -2
  11. package/dist/{chunk-PDHIOKRE.js → chunk-EQUK3RFS.js} +8 -4
  12. package/dist/chunk-EQUK3RFS.js.map +1 -0
  13. package/dist/{chunk-V7HQGZBT.js → chunk-GID26AN4.js} +2 -2
  14. package/dist/{chunk-27UXRPIQ.js → chunk-HM6V7F3M.js} +2 -2
  15. package/dist/chunk-IPYXE555.js +594 -0
  16. package/dist/chunk-IPYXE555.js.map +1 -0
  17. package/dist/chunk-LKKT3IVV.js +1644 -0
  18. package/dist/chunk-LKKT3IVV.js.map +1 -0
  19. package/dist/chunk-M7AH34KV.js +155 -0
  20. package/dist/chunk-M7AH34KV.js.map +1 -0
  21. package/dist/chunk-MAX3TN3C.js +249 -0
  22. package/dist/chunk-MAX3TN3C.js.map +1 -0
  23. package/dist/{chunk-QVGVJQMR.js → chunk-PMITBABE.js} +7 -5
  24. package/dist/{chunk-QVGVJQMR.js.map → chunk-PMITBABE.js.map} +1 -1
  25. package/dist/{chunk-J3LHTAAB.js → chunk-QOTFXW5L.js} +1 -47
  26. package/dist/chunk-QOTFXW5L.js.map +1 -0
  27. package/dist/chunk-RZTMDUO7.js +49 -0
  28. package/dist/chunk-RZTMDUO7.js.map +1 -0
  29. package/dist/chunk-VBQ3CRKH.js +291 -0
  30. package/dist/chunk-VBQ3CRKH.js.map +1 -0
  31. package/dist/{chunk-RQ5TP2TV.js → chunk-W5B3ZGP3.js} +3 -3
  32. package/dist/cli.js +13 -2
  33. package/dist/cli.js.map +1 -1
  34. package/dist/contract/index.d.ts +7 -0
  35. package/dist/contract/index.js +4 -4
  36. package/dist/index.d.ts +833 -82
  37. package/dist/index.js +61 -121
  38. package/dist/index.js.map +1 -1
  39. package/dist/openapi.json +1 -1
  40. package/dist/pipelines/index.js +4 -2
  41. package/dist/pipelines/index.js.map +1 -1
  42. package/dist/rl.d.ts +3 -0
  43. package/dist/rl.js +2 -2
  44. package/dist/rollout/index.d.ts +1061 -0
  45. package/dist/rollout/index.js +109 -0
  46. package/dist/rollout/index.js.map +1 -0
  47. package/dist/supervisor-run/index.d.ts +857 -0
  48. package/dist/supervisor-run/index.js +64 -0
  49. package/dist/supervisor-run/index.js.map +1 -0
  50. package/dist/wire/index.d.ts +3 -0
  51. package/dist/wire/index.js +2 -2
  52. package/docs/rollout.md +48 -0
  53. package/package.json +11 -1
  54. package/dist/chunk-J3LHTAAB.js.map +0 -1
  55. package/dist/chunk-PDHIOKRE.js.map +0 -1
  56. package/dist/chunk-ZU3QWGZE.js.map +0 -1
  57. /package/dist/{chunk-BUR5R4R4.js.map → chunk-A6GT67HT.js.map} +0 -0
  58. /package/dist/{chunk-EEHHBAWH.js.map → chunk-DT7OXY3C.js.map} +0 -0
  59. /package/dist/{chunk-V7HQGZBT.js.map → chunk-GID26AN4.js.map} +0 -0
  60. /package/dist/{chunk-27UXRPIQ.js.map → chunk-HM6V7F3M.js.map} +0 -0
  61. /package/dist/{chunk-RQ5TP2TV.js.map → chunk-W5B3ZGP3.js.map} +0 -0
package/dist/cli.js.map CHANGED
@@ -1 +1 @@
1
- {"version":3,"sources":["../src/cli.ts","../src/cli-config.ts"],"sourcesContent":["#!/usr/bin/env node\n/**\n * agent-eval CLI.\n *\n * agent-eval serve [--port 5005] [--host 127.0.0.1]\n * agent-eval rpc <method> # one request from stdin → one response on stdout\n * agent-eval rpc-batch <method> # JSONL stdin → JSONL stdout\n * agent-eval openapi [--out path] # write OpenAPI spec\n * agent-eval version\n *\n * <method> is one of: judge, listRubrics, version. When omitted, the\n * stdin payload must be a full {method, params} envelope.\n */\nimport { writeFileSync } from 'node:fs'\nimport { resolveCliLlmConfig } from './cli-config'\nimport { handleVersion } from './wire/handlers'\nimport { buildOpenApi } from './wire/openapi'\nimport { runRpcBatch, runRpcOnce } from './wire/rpc'\nimport { startServerAsync } from './wire/server'\n\ninterface Args {\n command: string\n positional: string[]\n flags: Record<string, string>\n}\n\nfunction parseArgs(argv: string[]): Args {\n const [command, ...rest] = argv\n const positional: string[] = []\n const flags: Record<string, string> = {}\n for (let i = 0; i < rest.length; i++) {\n const tok = rest[i]!\n if (tok.startsWith('--')) {\n const raw = tok.slice(2)\n const equalsAt = raw.indexOf('=')\n if (equalsAt >= 0) {\n flags[raw.slice(0, equalsAt)] = raw.slice(equalsAt + 1)\n continue\n }\n const key = raw\n if (key === 'help') {\n flags[key] = 'true'\n continue\n }\n const next = rest[i + 1]\n if (next != null && !next.startsWith('--')) {\n flags[key] = next\n i++\n } else {\n flags[key] = 'true'\n }\n } else if (tok === '-h') {\n flags.help = 'true'\n } else {\n positional.push(tok)\n }\n }\n return { command: command ?? 'help', positional, flags }\n}\n\nconst HELP = `agent-eval: evaluation RPC and HTTP server.\n\nCommands:\n serve [--port 5005] [--host 127.0.0.1]\n Start the HTTP server. POST /v1/judge, GET /v1/rubrics, GET /v1/version, GET /openapi.json.\n rpc <method>\n Read one JSON object from stdin (the params for <method>), write one\n JSON object to stdout. Methods: judge, listRubrics, version.\n rpc-batch <method>\n Like 'rpc' but JSONL in / JSONL out.\n openapi [--out openapi.json]\n Write the OpenAPI 3.1 spec.\n version\n Print server + wire-protocol version JSON.\n\nJudge provider:\n Set AGENT_EVAL_LLM_BASE_URL, AGENT_EVAL_LLM_API_KEY, and AGENT_EVAL_LLM_MODEL.\n OPENAI_* and TANGLE_* equivalents are also accepted.\n\nWithout arguments, prints this help.`\n\nasync function main(): Promise<number> {\n const { command, positional, flags } = parseArgs(process.argv.slice(2))\n assertKnownFlags(command, flags)\n\n if (flags.help === 'true') {\n process.stdout.write(`${HELP}\\n`)\n return 0\n }\n\n switch (command) {\n case 'serve': {\n const port = parsePort(flags.port ?? '5005')\n const host = flags.host ?? '127.0.0.1'\n const llm = resolveCliLlmConfig()\n const { server } = await startServerAsync({\n port,\n host,\n llm: llm.client,\n judgeModel: llm.model,\n llmRouteRequirements: { requireExplicitBaseUrl: true },\n })\n // Keep process alive on SIGINT/SIGTERM\n const shutdown = (sig: string) => {\n // eslint-disable-next-line no-console\n console.log(`[agent-eval] received ${sig}, shutting down`)\n server.close(() => process.exit(0))\n // Force exit after 5s if close hangs\n setTimeout(() => process.exit(1), 5000).unref()\n }\n process.on('SIGINT', () => shutdown('SIGINT'))\n process.on('SIGTERM', () => shutdown('SIGTERM'))\n // Block forever\n await new Promise(() => {})\n return 0\n }\n case 'rpc': {\n const [method] = positional\n const llm = resolveCliLlmConfig()\n return await runRpcOnce(method, {\n llm: llm.client,\n judgeModel: llm.model,\n llmRouteRequirements: { requireExplicitBaseUrl: true },\n })\n }\n case 'rpc-batch': {\n const [method] = positional\n const llm = resolveCliLlmConfig()\n return await runRpcBatch(method, {\n llm: llm.client,\n judgeModel: llm.model,\n llmRouteRequirements: { requireExplicitBaseUrl: true },\n })\n }\n case 'openapi': {\n const out = flags.out ?? 'openapi.json'\n const spec = buildOpenApi(handleVersion().version)\n writeFileSync(out, `${JSON.stringify(spec, null, 2)}\\n`, 'utf-8')\n // eslint-disable-next-line no-console\n console.log(`[agent-eval] wrote OpenAPI 3.1 spec to ${out}`)\n return 0\n }\n case 'version': {\n process.stdout.write(`${JSON.stringify(handleVersion(), null, 2)}\\n`)\n return 0\n }\n case '--version': {\n process.stdout.write(`${handleVersion().version}\\n`)\n return 0\n }\n case 'help':\n case '--help':\n case '-h':\n case '':\n process.stdout.write(`${HELP}\\n`)\n return 0\n default:\n process.stderr.write(`unknown command: ${command}\\n${HELP}\\n`)\n return 1\n }\n}\n\nconst FLAGS_BY_COMMAND: Record<string, ReadonlySet<string>> = {\n serve: new Set(['help', 'host', 'port']),\n rpc: new Set(['help']),\n 'rpc-batch': new Set(['help']),\n openapi: new Set(['help', 'out']),\n version: new Set(['help']),\n help: new Set(),\n '--help': new Set(),\n '-h': new Set(),\n '--version': new Set(),\n '': new Set(),\n}\n\nfunction assertKnownFlags(command: string, flags: Record<string, string>): void {\n const allowed = FLAGS_BY_COMMAND[command]\n if (!allowed) return\n const unknown = Object.keys(flags).filter((flag) => !allowed.has(flag))\n if (unknown.length > 0) {\n throw new Error(`unknown flag for ${command || 'help'}: --${unknown[0]}`)\n }\n}\n\nfunction parsePort(raw: string): number {\n const port = Number(raw)\n if (!Number.isInteger(port) || port < 0 || port > 65_535) {\n throw new Error(`--port must be an integer from 0 to 65535; received ${JSON.stringify(raw)}`)\n }\n return port\n}\n\nmain()\n .then((code) => process.exit(code))\n .catch((err) => {\n // eslint-disable-next-line no-console\n console.error('[agent-eval] cli error:', err)\n process.exit(1)\n })\n","import type { LlmClientOptions } from './llm-client'\n\nexport interface CliLlmConfig {\n client?: LlmClientOptions\n model?: string\n}\n\nexport function resolveCliLlmConfig(env: NodeJS.ProcessEnv = process.env): CliLlmConfig {\n const explicitBaseUrl = nonEmpty(env.AGENT_EVAL_LLM_BASE_URL)\n const explicitApiKey = nonEmpty(env.AGENT_EVAL_LLM_API_KEY)\n const openAiApiKey = nonEmpty(env.OPENAI_API_KEY)\n const tangleApiKey = nonEmpty(env.TANGLE_API_KEY)\n const baseUrl =\n explicitBaseUrl ??\n nonEmpty(env.OPENAI_BASE_URL) ??\n nonEmpty(env.TANGLE_ROUTER_URL) ??\n (openAiApiKey ? 'https://api.openai.com/v1' : undefined) ??\n (tangleApiKey ? 'https://router.tangle.tools/v1' : undefined)\n const apiKey = explicitApiKey ?? openAiApiKey ?? tangleApiKey\n const model =\n nonEmpty(env.AGENT_EVAL_LLM_MODEL) ?? nonEmpty(env.OPENAI_MODEL) ?? nonEmpty(env.TANGLE_MODEL)\n\n const client =\n baseUrl || apiKey\n ? { ...(baseUrl ? { baseUrl } : {}), ...(apiKey ? { apiKey } : {}) }\n : undefined\n return { ...(client ? { client } : {}), ...(model ? { model } : {}) }\n}\n\nfunction nonEmpty(value: string | undefined): string | undefined {\n const trimmed = value?.trim()\n return trimmed ? trimmed : undefined\n}\n"],"mappings":";;;;;;;;;;;;;;;;AAaA,SAAS,qBAAqB;;;ACNvB,SAAS,oBAAoB,MAAyB,QAAQ,KAAmB;AACtF,QAAM,kBAAkB,SAAS,IAAI,uBAAuB;AAC5D,QAAM,iBAAiB,SAAS,IAAI,sBAAsB;AAC1D,QAAM,eAAe,SAAS,IAAI,cAAc;AAChD,QAAM,eAAe,SAAS,IAAI,cAAc;AAChD,QAAM,UACJ,mBACA,SAAS,IAAI,eAAe,KAC5B,SAAS,IAAI,iBAAiB,MAC7B,eAAe,8BAA8B,YAC7C,eAAe,mCAAmC;AACrD,QAAM,SAAS,kBAAkB,gBAAgB;AACjD,QAAM,QACJ,SAAS,IAAI,oBAAoB,KAAK,SAAS,IAAI,YAAY,KAAK,SAAS,IAAI,YAAY;AAE/F,QAAM,SACJ,WAAW,SACP,EAAE,GAAI,UAAU,EAAE,QAAQ,IAAI,CAAC,GAAI,GAAI,SAAS,EAAE,OAAO,IAAI,CAAC,EAAG,IACjE;AACN,SAAO,EAAE,GAAI,SAAS,EAAE,OAAO,IAAI,CAAC,GAAI,GAAI,QAAQ,EAAE,MAAM,IAAI,CAAC,EAAG;AACtE;AAEA,SAAS,SAAS,OAA+C;AAC/D,QAAM,UAAU,OAAO,KAAK;AAC5B,SAAO,UAAU,UAAU;AAC7B;;;ADNA,SAAS,UAAU,MAAsB;AACvC,QAAM,CAAC,SAAS,GAAG,IAAI,IAAI;AAC3B,QAAM,aAAuB,CAAC;AAC9B,QAAM,QAAgC,CAAC;AACvC,WAAS,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK;AACpC,UAAM,MAAM,KAAK,CAAC;AAClB,QAAI,IAAI,WAAW,IAAI,GAAG;AACxB,YAAM,MAAM,IAAI,MAAM,CAAC;AACvB,YAAM,WAAW,IAAI,QAAQ,GAAG;AAChC,UAAI,YAAY,GAAG;AACjB,cAAM,IAAI,MAAM,GAAG,QAAQ,CAAC,IAAI,IAAI,MAAM,WAAW,CAAC;AACtD;AAAA,MACF;AACA,YAAM,MAAM;AACZ,UAAI,QAAQ,QAAQ;AAClB,cAAM,GAAG,IAAI;AACb;AAAA,MACF;AACA,YAAM,OAAO,KAAK,IAAI,CAAC;AACvB,UAAI,QAAQ,QAAQ,CAAC,KAAK,WAAW,IAAI,GAAG;AAC1C,cAAM,GAAG,IAAI;AACb;AAAA,MACF,OAAO;AACL,cAAM,GAAG,IAAI;AAAA,MACf;AAAA,IACF,WAAW,QAAQ,MAAM;AACvB,YAAM,OAAO;AAAA,IACf,OAAO;AACL,iBAAW,KAAK,GAAG;AAAA,IACrB;AAAA,EACF;AACA,SAAO,EAAE,SAAS,WAAW,QAAQ,YAAY,MAAM;AACzD;AAEA,IAAM,OAAO;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAqBb,eAAe,OAAwB;AACrC,QAAM,EAAE,SAAS,YAAY,MAAM,IAAI,UAAU,QAAQ,KAAK,MAAM,CAAC,CAAC;AACtE,mBAAiB,SAAS,KAAK;AAE/B,MAAI,MAAM,SAAS,QAAQ;AACzB,YAAQ,OAAO,MAAM,GAAG,IAAI;AAAA,CAAI;AAChC,WAAO;AAAA,EACT;AAEA,UAAQ,SAAS;AAAA,IACf,KAAK,SAAS;AACZ,YAAM,OAAO,UAAU,MAAM,QAAQ,MAAM;AAC3C,YAAM,OAAO,MAAM,QAAQ;AAC3B,YAAM,MAAM,oBAAoB;AAChC,YAAM,EAAE,OAAO,IAAI,MAAM,iBAAiB;AAAA,QACxC;AAAA,QACA;AAAA,QACA,KAAK,IAAI;AAAA,QACT,YAAY,IAAI;AAAA,QAChB,sBAAsB,EAAE,wBAAwB,KAAK;AAAA,MACvD,CAAC;AAED,YAAM,WAAW,CAAC,QAAgB;AAEhC,gBAAQ,IAAI,yBAAyB,GAAG,iBAAiB;AACzD,eAAO,MAAM,MAAM,QAAQ,KAAK,CAAC,CAAC;AAElC,mBAAW,MAAM,QAAQ,KAAK,CAAC,GAAG,GAAI,EAAE,MAAM;AAAA,MAChD;AACA,cAAQ,GAAG,UAAU,MAAM,SAAS,QAAQ,CAAC;AAC7C,cAAQ,GAAG,WAAW,MAAM,SAAS,SAAS,CAAC;AAE/C,YAAM,IAAI,QAAQ,MAAM;AAAA,MAAC,CAAC;AAC1B,aAAO;AAAA,IACT;AAAA,IACA,KAAK,OAAO;AACV,YAAM,CAAC,MAAM,IAAI;AACjB,YAAM,MAAM,oBAAoB;AAChC,aAAO,MAAM,WAAW,QAAQ;AAAA,QAC9B,KAAK,IAAI;AAAA,QACT,YAAY,IAAI;AAAA,QAChB,sBAAsB,EAAE,wBAAwB,KAAK;AAAA,MACvD,CAAC;AAAA,IACH;AAAA,IACA,KAAK,aAAa;AAChB,YAAM,CAAC,MAAM,IAAI;AACjB,YAAM,MAAM,oBAAoB;AAChC,aAAO,MAAM,YAAY,QAAQ;AAAA,QAC/B,KAAK,IAAI;AAAA,QACT,YAAY,IAAI;AAAA,QAChB,sBAAsB,EAAE,wBAAwB,KAAK;AAAA,MACvD,CAAC;AAAA,IACH;AAAA,IACA,KAAK,WAAW;AACd,YAAM,MAAM,MAAM,OAAO;AACzB,YAAM,OAAO,aAAa,cAAc,EAAE,OAAO;AACjD,oBAAc,KAAK,GAAG,KAAK,UAAU,MAAM,MAAM,CAAC,CAAC;AAAA,GAAM,OAAO;AAEhE,cAAQ,IAAI,0CAA0C,GAAG,EAAE;AAC3D,aAAO;AAAA,IACT;AAAA,IACA,KAAK,WAAW;AACd,cAAQ,OAAO,MAAM,GAAG,KAAK,UAAU,cAAc,GAAG,MAAM,CAAC,CAAC;AAAA,CAAI;AACpE,aAAO;AAAA,IACT;AAAA,IACA,KAAK,aAAa;AAChB,cAAQ,OAAO,MAAM,GAAG,cAAc,EAAE,OAAO;AAAA,CAAI;AACnD,aAAO;AAAA,IACT;AAAA,IACA,KAAK;AAAA,IACL,KAAK;AAAA,IACL,KAAK;AAAA,IACL,KAAK;AACH,cAAQ,OAAO,MAAM,GAAG,IAAI;AAAA,CAAI;AAChC,aAAO;AAAA,IACT;AACE,cAAQ,OAAO,MAAM,oBAAoB,OAAO;AAAA,EAAK,IAAI;AAAA,CAAI;AAC7D,aAAO;AAAA,EACX;AACF;AAEA,IAAM,mBAAwD;AAAA,EAC5D,OAAO,oBAAI,IAAI,CAAC,QAAQ,QAAQ,MAAM,CAAC;AAAA,EACvC,KAAK,oBAAI,IAAI,CAAC,MAAM,CAAC;AAAA,EACrB,aAAa,oBAAI,IAAI,CAAC,MAAM,CAAC;AAAA,EAC7B,SAAS,oBAAI,IAAI,CAAC,QAAQ,KAAK,CAAC;AAAA,EAChC,SAAS,oBAAI,IAAI,CAAC,MAAM,CAAC;AAAA,EACzB,MAAM,oBAAI,IAAI;AAAA,EACd,UAAU,oBAAI,IAAI;AAAA,EAClB,MAAM,oBAAI,IAAI;AAAA,EACd,aAAa,oBAAI,IAAI;AAAA,EACrB,IAAI,oBAAI,IAAI;AACd;AAEA,SAAS,iBAAiB,SAAiB,OAAqC;AAC9E,QAAM,UAAU,iBAAiB,OAAO;AACxC,MAAI,CAAC,QAAS;AACd,QAAM,UAAU,OAAO,KAAK,KAAK,EAAE,OAAO,CAAC,SAAS,CAAC,QAAQ,IAAI,IAAI,CAAC;AACtE,MAAI,QAAQ,SAAS,GAAG;AACtB,UAAM,IAAI,MAAM,oBAAoB,WAAW,MAAM,OAAO,QAAQ,CAAC,CAAC,EAAE;AAAA,EAC1E;AACF;AAEA,SAAS,UAAU,KAAqB;AACtC,QAAM,OAAO,OAAO,GAAG;AACvB,MAAI,CAAC,OAAO,UAAU,IAAI,KAAK,OAAO,KAAK,OAAO,OAAQ;AACxD,UAAM,IAAI,MAAM,uDAAuD,KAAK,UAAU,GAAG,CAAC,EAAE;AAAA,EAC9F;AACA,SAAO;AACT;AAEA,KAAK,EACF,KAAK,CAAC,SAAS,QAAQ,KAAK,IAAI,CAAC,EACjC,MAAM,CAAC,QAAQ;AAEd,UAAQ,MAAM,2BAA2B,GAAG;AAC5C,UAAQ,KAAK,CAAC;AAChB,CAAC;","names":[]}
1
+ {"version":3,"sources":["../src/cli.ts","../src/cli-config.ts"],"sourcesContent":["#!/usr/bin/env node\n/**\n * agent-eval CLI.\n *\n * agent-eval serve [--port 5005] [--host 127.0.0.1]\n * agent-eval rpc <method> # one request from stdin → one response on stdout\n * agent-eval rpc-batch <method> # JSONL stdin → JSONL stdout\n * agent-eval openapi [--out path] # write OpenAPI spec\n * agent-eval version\n *\n * <method> is one of: judge, listRubrics, version. When omitted, the\n * stdin payload must be a full {method, params} envelope.\n */\nimport { writeFileSync } from 'node:fs'\nimport { resolveCliLlmConfig } from './cli-config'\nimport { runRolloutReleaseCli } from './rollout/release/hf-dataset'\nimport { handleVersion } from './wire/handlers'\nimport { buildOpenApi } from './wire/openapi'\nimport { runRpcBatch, runRpcOnce } from './wire/rpc'\nimport { startServerAsync } from './wire/server'\n\ninterface Args {\n command: string\n positional: string[]\n flags: Record<string, string>\n}\n\nfunction parseArgs(argv: string[]): Args {\n const [command, ...rest] = argv\n const positional: string[] = []\n const flags: Record<string, string> = {}\n for (let i = 0; i < rest.length; i++) {\n const tok = rest[i]!\n if (tok.startsWith('--')) {\n const raw = tok.slice(2)\n const equalsAt = raw.indexOf('=')\n if (equalsAt >= 0) {\n flags[raw.slice(0, equalsAt)] = raw.slice(equalsAt + 1)\n continue\n }\n const key = raw\n if (key === 'help') {\n flags[key] = 'true'\n continue\n }\n const next = rest[i + 1]\n if (next != null && !next.startsWith('--')) {\n flags[key] = next\n i++\n } else {\n flags[key] = 'true'\n }\n } else if (tok === '-h') {\n flags.help = 'true'\n } else {\n positional.push(tok)\n }\n }\n return { command: command ?? 'help', positional, flags }\n}\n\nconst HELP = `agent-eval: evaluation RPC and HTTP server.\n\nCommands:\n serve [--port 5005] [--host 127.0.0.1]\n Start the HTTP server. POST /v1/judge, GET /v1/rubrics, GET /v1/version, GET /openapi.json.\n rpc <method>\n Read one JSON object from stdin (the params for <method>), write one\n JSON object to stdout. Methods: judge, listRubrics, version.\n rpc-batch <method>\n Like 'rpc' but JSONL in / JSONL out.\n openapi [--out openapi.json]\n Write the OpenAPI 3.1 spec.\n rollout-release <ledger.jsonl...> --out <dir> [--formats sft,verifiers,rft,raw] [--include-proposers] [--push <org/name>]\n Build a HuggingFace-ready dataset dir from tangle.rollout.v1 ledgers:\n validate, fail-closed split filter, scrub, export formats + card.\n version\n Print server + wire-protocol version JSON.\n\nJudge provider:\n Set AGENT_EVAL_LLM_BASE_URL, AGENT_EVAL_LLM_API_KEY, and AGENT_EVAL_LLM_MODEL.\n OPENAI_* and TANGLE_* equivalents are also accepted.\n\nWithout arguments, prints this help.`\n\nasync function main(): Promise<number> {\n const { command, positional, flags } = parseArgs(process.argv.slice(2))\n assertKnownFlags(command, flags)\n\n if (flags.help === 'true') {\n process.stdout.write(`${HELP}\\n`)\n return 0\n }\n\n switch (command) {\n case 'serve': {\n const port = parsePort(flags.port ?? '5005')\n const host = flags.host ?? '127.0.0.1'\n const llm = resolveCliLlmConfig()\n const { server } = await startServerAsync({\n port,\n host,\n llm: llm.client,\n judgeModel: llm.model,\n llmRouteRequirements: { requireExplicitBaseUrl: true },\n })\n // Keep process alive on SIGINT/SIGTERM\n const shutdown = (sig: string) => {\n // eslint-disable-next-line no-console\n console.log(`[agent-eval] received ${sig}, shutting down`)\n server.close(() => process.exit(0))\n // Force exit after 5s if close hangs\n setTimeout(() => process.exit(1), 5000).unref()\n }\n process.on('SIGINT', () => shutdown('SIGINT'))\n process.on('SIGTERM', () => shutdown('SIGTERM'))\n // Block forever\n await new Promise(() => {})\n return 0\n }\n case 'rpc': {\n const [method] = positional\n const llm = resolveCliLlmConfig()\n return await runRpcOnce(method, {\n llm: llm.client,\n judgeModel: llm.model,\n llmRouteRequirements: { requireExplicitBaseUrl: true },\n })\n }\n case 'rpc-batch': {\n const [method] = positional\n const llm = resolveCliLlmConfig()\n return await runRpcBatch(method, {\n llm: llm.client,\n judgeModel: llm.model,\n llmRouteRequirements: { requireExplicitBaseUrl: true },\n })\n }\n case 'openapi': {\n const out = flags.out ?? 'openapi.json'\n const spec = buildOpenApi(handleVersion().version)\n writeFileSync(out, `${JSON.stringify(spec, null, 2)}\\n`, 'utf-8')\n // eslint-disable-next-line no-console\n console.log(`[agent-eval] wrote OpenAPI 3.1 spec to ${out}`)\n return 0\n }\n case 'rollout-release': {\n // The subcommand owns its own flag grammar (multi-value --formats,\n // boolean --include-proposers) — pass raw argv through untouched.\n return await runRolloutReleaseCli(process.argv.slice(3))\n }\n case 'version': {\n process.stdout.write(`${JSON.stringify(handleVersion(), null, 2)}\\n`)\n return 0\n }\n case '--version': {\n process.stdout.write(`${handleVersion().version}\\n`)\n return 0\n }\n case 'help':\n case '--help':\n case '-h':\n case '':\n process.stdout.write(`${HELP}\\n`)\n return 0\n default:\n process.stderr.write(`unknown command: ${command}\\n${HELP}\\n`)\n return 1\n }\n}\n\nconst FLAGS_BY_COMMAND: Record<string, ReadonlySet<string>> = {\n serve: new Set(['help', 'host', 'port']),\n rpc: new Set(['help']),\n 'rpc-batch': new Set(['help']),\n openapi: new Set(['help', 'out']),\n version: new Set(['help']),\n help: new Set(),\n '--help': new Set(),\n '-h': new Set(),\n '--version': new Set(),\n '': new Set(),\n}\n\nfunction assertKnownFlags(command: string, flags: Record<string, string>): void {\n // rollout-release parses its own argv (multi-value flags).\n if (command === 'rollout-release') return\n const allowed = FLAGS_BY_COMMAND[command]\n if (!allowed) return\n const unknown = Object.keys(flags).filter((flag) => !allowed.has(flag))\n if (unknown.length > 0) {\n throw new Error(`unknown flag for ${command || 'help'}: --${unknown[0]}`)\n }\n}\n\nfunction parsePort(raw: string): number {\n const port = Number(raw)\n if (!Number.isInteger(port) || port < 0 || port > 65_535) {\n throw new Error(`--port must be an integer from 0 to 65535; received ${JSON.stringify(raw)}`)\n }\n return port\n}\n\nmain()\n .then((code) => process.exit(code))\n .catch((err) => {\n // eslint-disable-next-line no-console\n console.error('[agent-eval] cli error:', err)\n process.exit(1)\n })\n","import type { LlmClientOptions } from './llm-client'\n\nexport interface CliLlmConfig {\n client?: LlmClientOptions\n model?: string\n}\n\nexport function resolveCliLlmConfig(env: NodeJS.ProcessEnv = process.env): CliLlmConfig {\n const explicitBaseUrl = nonEmpty(env.AGENT_EVAL_LLM_BASE_URL)\n const explicitApiKey = nonEmpty(env.AGENT_EVAL_LLM_API_KEY)\n const openAiApiKey = nonEmpty(env.OPENAI_API_KEY)\n const tangleApiKey = nonEmpty(env.TANGLE_API_KEY)\n const baseUrl =\n explicitBaseUrl ??\n nonEmpty(env.OPENAI_BASE_URL) ??\n nonEmpty(env.TANGLE_ROUTER_URL) ??\n (openAiApiKey ? 'https://api.openai.com/v1' : undefined) ??\n (tangleApiKey ? 'https://router.tangle.tools/v1' : undefined)\n const apiKey = explicitApiKey ?? openAiApiKey ?? tangleApiKey\n const model =\n nonEmpty(env.AGENT_EVAL_LLM_MODEL) ?? nonEmpty(env.OPENAI_MODEL) ?? nonEmpty(env.TANGLE_MODEL)\n\n const client =\n baseUrl || apiKey\n ? { ...(baseUrl ? { baseUrl } : {}), ...(apiKey ? { apiKey } : {}) }\n : undefined\n return { ...(client ? { client } : {}), ...(model ? { model } : {}) }\n}\n\nfunction nonEmpty(value: string | undefined): string | undefined {\n const trimmed = value?.trim()\n return trimmed ? trimmed : undefined\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;AAaA,SAAS,qBAAqB;;;ACNvB,SAAS,oBAAoB,MAAyB,QAAQ,KAAmB;AACtF,QAAM,kBAAkB,SAAS,IAAI,uBAAuB;AAC5D,QAAM,iBAAiB,SAAS,IAAI,sBAAsB;AAC1D,QAAM,eAAe,SAAS,IAAI,cAAc;AAChD,QAAM,eAAe,SAAS,IAAI,cAAc;AAChD,QAAM,UACJ,mBACA,SAAS,IAAI,eAAe,KAC5B,SAAS,IAAI,iBAAiB,MAC7B,eAAe,8BAA8B,YAC7C,eAAe,mCAAmC;AACrD,QAAM,SAAS,kBAAkB,gBAAgB;AACjD,QAAM,QACJ,SAAS,IAAI,oBAAoB,KAAK,SAAS,IAAI,YAAY,KAAK,SAAS,IAAI,YAAY;AAE/F,QAAM,SACJ,WAAW,SACP,EAAE,GAAI,UAAU,EAAE,QAAQ,IAAI,CAAC,GAAI,GAAI,SAAS,EAAE,OAAO,IAAI,CAAC,EAAG,IACjE;AACN,SAAO,EAAE,GAAI,SAAS,EAAE,OAAO,IAAI,CAAC,GAAI,GAAI,QAAQ,EAAE,MAAM,IAAI,CAAC,EAAG;AACtE;AAEA,SAAS,SAAS,OAA+C;AAC/D,QAAM,UAAU,OAAO,KAAK;AAC5B,SAAO,UAAU,UAAU;AAC7B;;;ADLA,SAAS,UAAU,MAAsB;AACvC,QAAM,CAAC,SAAS,GAAG,IAAI,IAAI;AAC3B,QAAM,aAAuB,CAAC;AAC9B,QAAM,QAAgC,CAAC;AACvC,WAAS,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK;AACpC,UAAM,MAAM,KAAK,CAAC;AAClB,QAAI,IAAI,WAAW,IAAI,GAAG;AACxB,YAAM,MAAM,IAAI,MAAM,CAAC;AACvB,YAAM,WAAW,IAAI,QAAQ,GAAG;AAChC,UAAI,YAAY,GAAG;AACjB,cAAM,IAAI,MAAM,GAAG,QAAQ,CAAC,IAAI,IAAI,MAAM,WAAW,CAAC;AACtD;AAAA,MACF;AACA,YAAM,MAAM;AACZ,UAAI,QAAQ,QAAQ;AAClB,cAAM,GAAG,IAAI;AACb;AAAA,MACF;AACA,YAAM,OAAO,KAAK,IAAI,CAAC;AACvB,UAAI,QAAQ,QAAQ,CAAC,KAAK,WAAW,IAAI,GAAG;AAC1C,cAAM,GAAG,IAAI;AACb;AAAA,MACF,OAAO;AACL,cAAM,GAAG,IAAI;AAAA,MACf;AAAA,IACF,WAAW,QAAQ,MAAM;AACvB,YAAM,OAAO;AAAA,IACf,OAAO;AACL,iBAAW,KAAK,GAAG;AAAA,IACrB;AAAA,EACF;AACA,SAAO,EAAE,SAAS,WAAW,QAAQ,YAAY,MAAM;AACzD;AAEA,IAAM,OAAO;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAwBb,eAAe,OAAwB;AACrC,QAAM,EAAE,SAAS,YAAY,MAAM,IAAI,UAAU,QAAQ,KAAK,MAAM,CAAC,CAAC;AACtE,mBAAiB,SAAS,KAAK;AAE/B,MAAI,MAAM,SAAS,QAAQ;AACzB,YAAQ,OAAO,MAAM,GAAG,IAAI;AAAA,CAAI;AAChC,WAAO;AAAA,EACT;AAEA,UAAQ,SAAS;AAAA,IACf,KAAK,SAAS;AACZ,YAAM,OAAO,UAAU,MAAM,QAAQ,MAAM;AAC3C,YAAM,OAAO,MAAM,QAAQ;AAC3B,YAAM,MAAM,oBAAoB;AAChC,YAAM,EAAE,OAAO,IAAI,MAAM,iBAAiB;AAAA,QACxC;AAAA,QACA;AAAA,QACA,KAAK,IAAI;AAAA,QACT,YAAY,IAAI;AAAA,QAChB,sBAAsB,EAAE,wBAAwB,KAAK;AAAA,MACvD,CAAC;AAED,YAAM,WAAW,CAAC,QAAgB;AAEhC,gBAAQ,IAAI,yBAAyB,GAAG,iBAAiB;AACzD,eAAO,MAAM,MAAM,QAAQ,KAAK,CAAC,CAAC;AAElC,mBAAW,MAAM,QAAQ,KAAK,CAAC,GAAG,GAAI,EAAE,MAAM;AAAA,MAChD;AACA,cAAQ,GAAG,UAAU,MAAM,SAAS,QAAQ,CAAC;AAC7C,cAAQ,GAAG,WAAW,MAAM,SAAS,SAAS,CAAC;AAE/C,YAAM,IAAI,QAAQ,MAAM;AAAA,MAAC,CAAC;AAC1B,aAAO;AAAA,IACT;AAAA,IACA,KAAK,OAAO;AACV,YAAM,CAAC,MAAM,IAAI;AACjB,YAAM,MAAM,oBAAoB;AAChC,aAAO,MAAM,WAAW,QAAQ;AAAA,QAC9B,KAAK,IAAI;AAAA,QACT,YAAY,IAAI;AAAA,QAChB,sBAAsB,EAAE,wBAAwB,KAAK;AAAA,MACvD,CAAC;AAAA,IACH;AAAA,IACA,KAAK,aAAa;AAChB,YAAM,CAAC,MAAM,IAAI;AACjB,YAAM,MAAM,oBAAoB;AAChC,aAAO,MAAM,YAAY,QAAQ;AAAA,QAC/B,KAAK,IAAI;AAAA,QACT,YAAY,IAAI;AAAA,QAChB,sBAAsB,EAAE,wBAAwB,KAAK;AAAA,MACvD,CAAC;AAAA,IACH;AAAA,IACA,KAAK,WAAW;AACd,YAAM,MAAM,MAAM,OAAO;AACzB,YAAM,OAAO,aAAa,cAAc,EAAE,OAAO;AACjD,oBAAc,KAAK,GAAG,KAAK,UAAU,MAAM,MAAM,CAAC,CAAC;AAAA,GAAM,OAAO;AAEhE,cAAQ,IAAI,0CAA0C,GAAG,EAAE;AAC3D,aAAO;AAAA,IACT;AAAA,IACA,KAAK,mBAAmB;AAGtB,aAAO,MAAM,qBAAqB,QAAQ,KAAK,MAAM,CAAC,CAAC;AAAA,IACzD;AAAA,IACA,KAAK,WAAW;AACd,cAAQ,OAAO,MAAM,GAAG,KAAK,UAAU,cAAc,GAAG,MAAM,CAAC,CAAC;AAAA,CAAI;AACpE,aAAO;AAAA,IACT;AAAA,IACA,KAAK,aAAa;AAChB,cAAQ,OAAO,MAAM,GAAG,cAAc,EAAE,OAAO;AAAA,CAAI;AACnD,aAAO;AAAA,IACT;AAAA,IACA,KAAK;AAAA,IACL,KAAK;AAAA,IACL,KAAK;AAAA,IACL,KAAK;AACH,cAAQ,OAAO,MAAM,GAAG,IAAI;AAAA,CAAI;AAChC,aAAO;AAAA,IACT;AACE,cAAQ,OAAO,MAAM,oBAAoB,OAAO;AAAA,EAAK,IAAI;AAAA,CAAI;AAC7D,aAAO;AAAA,EACX;AACF;AAEA,IAAM,mBAAwD;AAAA,EAC5D,OAAO,oBAAI,IAAI,CAAC,QAAQ,QAAQ,MAAM,CAAC;AAAA,EACvC,KAAK,oBAAI,IAAI,CAAC,MAAM,CAAC;AAAA,EACrB,aAAa,oBAAI,IAAI,CAAC,MAAM,CAAC;AAAA,EAC7B,SAAS,oBAAI,IAAI,CAAC,QAAQ,KAAK,CAAC;AAAA,EAChC,SAAS,oBAAI,IAAI,CAAC,MAAM,CAAC;AAAA,EACzB,MAAM,oBAAI,IAAI;AAAA,EACd,UAAU,oBAAI,IAAI;AAAA,EAClB,MAAM,oBAAI,IAAI;AAAA,EACd,aAAa,oBAAI,IAAI;AAAA,EACrB,IAAI,oBAAI,IAAI;AACd;AAEA,SAAS,iBAAiB,SAAiB,OAAqC;AAE9E,MAAI,YAAY,kBAAmB;AACnC,QAAM,UAAU,iBAAiB,OAAO;AACxC,MAAI,CAAC,QAAS;AACd,QAAM,UAAU,OAAO,KAAK,KAAK,EAAE,OAAO,CAAC,SAAS,CAAC,QAAQ,IAAI,IAAI,CAAC;AACtE,MAAI,QAAQ,SAAS,GAAG;AACtB,UAAM,IAAI,MAAM,oBAAoB,WAAW,MAAM,OAAO,QAAQ,CAAC,CAAC,EAAE;AAAA,EAC1E;AACF;AAEA,SAAS,UAAU,KAAqB;AACtC,QAAM,OAAO,OAAO,GAAG;AACvB,MAAI,CAAC,OAAO,UAAU,IAAI,KAAK,OAAO,KAAK,OAAO,OAAQ;AACxD,UAAM,IAAI,MAAM,uDAAuD,KAAK,UAAU,GAAG,CAAC,EAAE;AAAA,EAC9F;AACA,SAAO;AACT;AAEA,KAAK,EACF,KAAK,CAAC,SAAS,QAAQ,KAAK,IAAI,CAAC,EACjC,MAAM,CAAC,QAAQ;AAEd,UAAQ,MAAM,2BAA2B,GAAG;AAC5C,UAAQ,KAAK,CAAC;AAChB,CAAC;","names":[]}
@@ -838,6 +838,7 @@ interface LlmMessage {
838
838
  };
839
839
  }>;
840
840
  }
841
+ type LlmThinkingMode = 'enabled' | 'disabled';
841
842
  interface LlmCallRequest {
842
843
  model: string;
843
844
  messages: LlmMessage[];
@@ -850,6 +851,8 @@ interface LlmCallRequest {
850
851
  };
851
852
  temperature?: number;
852
853
  maxTokens?: number;
854
+ /** OpenAI-compatible reasoning mode. Omitted when the provider default should apply. */
855
+ thinking?: LlmThinkingMode;
853
856
  /** Per-call timeout, default 300s. */
854
857
  timeoutMs?: number;
855
858
  }
@@ -944,6 +947,8 @@ interface LlmClientOptions {
944
947
  * Default: `extract`.
945
948
  */
946
949
  jsonPayloadMode?: 'extract' | 'exact';
950
+ /** Default provider reasoning mode. A per-call request value takes precedence. */
951
+ thinking?: LlmThinkingMode;
947
952
  /** Fetch implementation — defaults to global `fetch`. Override for custom transport (e.g. tests). */
948
953
  fetch?: typeof fetch;
949
954
  /**
@@ -2664,6 +2669,8 @@ interface GepaProposerOptions {
2664
2669
  temperature?: number;
2665
2670
  /** Reflection max tokens. Default 6000. */
2666
2671
  maxTokens?: number;
2672
+ /** Provider reasoning mode for each reflection call. Omitted to preserve the provider default. */
2673
+ thinking?: LlmCallRequest['thinking'];
2667
2674
  /** Structured-doc constraints. Candidates violating any are rejected
2668
2675
  * post-parse and dropped from the returned population. */
2669
2676
  constraints?: GepaProposerConstraints;
@@ -37,7 +37,7 @@ import {
37
37
  runReferenceEquivalenceJudge,
38
38
  surfaceContentHash,
39
39
  surfaceHash
40
- } from "../chunk-QVGVJQMR.js";
40
+ } from "../chunk-PMITBABE.js";
41
41
  import {
42
42
  campaignSplitDigest,
43
43
  createRunCostLedger,
@@ -49,15 +49,15 @@ import {
49
49
  import {
50
50
  buildDefaultAnalystRegistry,
51
51
  createChatClient
52
- } from "../chunk-BUR5R4R4.js";
52
+ } from "../chunk-A6GT67HT.js";
53
53
  import "../chunk-HHWE3POT.js";
54
- import "../chunk-EEHHBAWH.js";
54
+ import "../chunk-DT7OXY3C.js";
55
55
  import {
56
56
  FileSystemOutcomeStore,
57
57
  InMemoryOutcomeStore
58
58
  } from "../chunk-3RF76KTD.js";
59
59
  import "../chunk-ARU2PZFM.js";
60
- import "../chunk-PDHIOKRE.js";
60
+ import "../chunk-EQUK3RFS.js";
61
61
  import "../chunk-DPZAEKA6.js";
62
62
  import {
63
63
  pairedBootstrap