@tangle-network/agent-eval 0.129.0 → 0.130.0

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 (427) hide show
  1. package/CHANGELOG.md +14 -0
  2. package/README.md +1 -0
  3. package/dist/active-curriculum-C4mk67HP.js +214 -0
  4. package/dist/active-curriculum-C4mk67HP.js.map +1 -0
  5. package/dist/adversarial-smnADNFS.d.ts +21 -0
  6. package/dist/adversarial-smnADNFS.d.ts.map +1 -0
  7. package/dist/analyst/index.d.ts +81 -2872
  8. package/dist/analyst/index.d.ts.map +1 -0
  9. package/dist/analyst/index.js +319 -360
  10. package/dist/analyst/index.js.map +1 -1
  11. package/dist/analyst-BkTS3C58.d.ts +89 -0
  12. package/dist/analyst-BkTS3C58.d.ts.map +1 -0
  13. package/dist/analyst-LsnNpSkm.js +152 -0
  14. package/dist/analyst-LsnNpSkm.js.map +1 -0
  15. package/dist/analyze-runs-C1CavBMk.js +1067 -0
  16. package/dist/analyze-runs-C1CavBMk.js.map +1 -0
  17. package/dist/analyze-runs-FsgYCinh.d.ts +72 -0
  18. package/dist/analyze-runs-FsgYCinh.d.ts.map +1 -0
  19. package/dist/attribute-vocabulary-DLJ6303h.d.ts +54 -0
  20. package/dist/attribute-vocabulary-DLJ6303h.d.ts.map +1 -0
  21. package/dist/authenticity/index.d.ts +81 -79
  22. package/dist/authenticity/index.d.ts.map +1 -0
  23. package/dist/authenticity/index.js +209 -193
  24. package/dist/authenticity/index.js.map +1 -1
  25. package/dist/baseline-HsBvw_dk.js +550 -0
  26. package/dist/baseline-HsBvw_dk.js.map +1 -0
  27. package/dist/baseline-hG3K85h4.d.ts +125 -0
  28. package/dist/baseline-hG3K85h4.d.ts.map +1 -0
  29. package/dist/belief-state/index.d.ts +448 -1188
  30. package/dist/belief-state/index.d.ts.map +1 -0
  31. package/dist/belief-state/index.js +1617 -1709
  32. package/dist/belief-state/index.js.map +1 -1
  33. package/dist/benchmarks/index.d.ts +2 -891
  34. package/dist/benchmarks/index.js +2 -60
  35. package/dist/benchmarks-DviOvUNr.js +754 -0
  36. package/dist/benchmarks-DviOvUNr.js.map +1 -0
  37. package/dist/builder-eval/index.d.ts +150 -662
  38. package/dist/builder-eval/index.d.ts.map +1 -0
  39. package/dist/builder-eval/index.js +356 -345
  40. package/dist/builder-eval/index.js.map +1 -1
  41. package/dist/calibration-CNWWA6K8.js +94 -0
  42. package/dist/calibration-CNWWA6K8.js.map +1 -0
  43. package/dist/campaign/index.d.ts +5 -6381
  44. package/dist/campaign/index.js +3 -213
  45. package/dist/campaign-CBKZvQ1H.js +3885 -0
  46. package/dist/campaign-CBKZvQ1H.js.map +1 -0
  47. package/dist/cli.d.ts +1 -1
  48. package/dist/cli.js +164 -175
  49. package/dist/cli.js.map +1 -1
  50. package/dist/client-C97NMzqi.d.ts +581 -0
  51. package/dist/client-C97NMzqi.d.ts.map +1 -0
  52. package/dist/client-CYzbdJOZ.js +637 -0
  53. package/dist/client-CYzbdJOZ.js.map +1 -0
  54. package/dist/code-agent-session-BjkMTQ7H.js +1390 -0
  55. package/dist/code-agent-session-BjkMTQ7H.js.map +1 -0
  56. package/dist/code-agent-session-DqqgOJaz.d.ts +143 -0
  57. package/dist/code-agent-session-DqqgOJaz.d.ts.map +1 -0
  58. package/dist/concurrency-DIxRZF_J.js +85 -0
  59. package/dist/concurrency-DIxRZF_J.js.map +1 -0
  60. package/dist/contract/index.d.ts +645 -5565
  61. package/dist/contract/index.d.ts.map +1 -0
  62. package/dist/contract/index.js +1730 -1938
  63. package/dist/contract/index.js.map +1 -1
  64. package/dist/control.d.ts +3 -1030
  65. package/dist/control.js +2 -33
  66. package/dist/cost-ledger-DIgQUFZZ.js +801 -0
  67. package/dist/cost-ledger-DIgQUFZZ.js.map +1 -0
  68. package/dist/cost-ledger-Dye6jCgg.d.ts +236 -0
  69. package/dist/cost-ledger-Dye6jCgg.d.ts.map +1 -0
  70. package/dist/dataset-BvtnC8Dc.d.ts +115 -0
  71. package/dist/dataset-BvtnC8Dc.d.ts.map +1 -0
  72. package/dist/default-registry-C-vFCSEc.js +2579 -0
  73. package/dist/default-registry-C-vFCSEc.js.map +1 -0
  74. package/dist/default-registry-CNPo-Vsb.d.ts +540 -0
  75. package/dist/default-registry-CNPo-Vsb.d.ts.map +1 -0
  76. package/dist/emitter-CPBAhxum.js +266 -0
  77. package/dist/emitter-CPBAhxum.js.map +1 -0
  78. package/dist/emitter-DGQGoLyj.d.ts +113 -0
  79. package/dist/emitter-DGQGoLyj.d.ts.map +1 -0
  80. package/dist/errors-8YnH8WlF.js +64 -0
  81. package/dist/errors-8YnH8WlF.js.map +1 -0
  82. package/dist/errors-CEk209JS.d.ts +76 -0
  83. package/dist/errors-CEk209JS.d.ts.map +1 -0
  84. package/dist/eval-campaign-DEm6c8ru.js +349 -0
  85. package/dist/eval-campaign-DEm6c8ru.js.map +1 -0
  86. package/dist/execution-tracks-CpgFPpS5.js +93 -0
  87. package/dist/execution-tracks-CpgFPpS5.js.map +1 -0
  88. package/dist/exporters-q9iL-2Jf.js +148 -0
  89. package/dist/exporters-q9iL-2Jf.js.map +1 -0
  90. package/dist/extract-usage-BrQ8mCLX.js +155 -0
  91. package/dist/extract-usage-BrQ8mCLX.js.map +1 -0
  92. package/dist/failure-cluster-CqcvCcdR.d.ts +59 -0
  93. package/dist/failure-cluster-CqcvCcdR.d.ts.map +1 -0
  94. package/dist/feedback-trajectory-CVaeREXV.d.ts +340 -0
  95. package/dist/feedback-trajectory-CVaeREXV.d.ts.map +1 -0
  96. package/dist/fuzz.d.ts +320 -646
  97. package/dist/fuzz.d.ts.map +1 -0
  98. package/dist/fuzz.js +670 -618
  99. package/dist/fuzz.js.map +1 -1
  100. package/dist/hf-dataset-DBJXXoY1.js +763 -0
  101. package/dist/hf-dataset-DBJXXoY1.js.map +1 -0
  102. package/dist/hosted/index.d.ts +11 -831
  103. package/dist/hosted/index.d.ts.map +1 -0
  104. package/dist/hosted/index.js +2 -37
  105. package/dist/index-2JJSA6-r2.d.ts +926 -0
  106. package/dist/index-2JJSA6-r2.d.ts.map +1 -0
  107. package/dist/index-6N0aYmpW.d.ts +217 -0
  108. package/dist/index-6N0aYmpW.d.ts.map +1 -0
  109. package/dist/index-BAvgST_9.d.ts +131 -0
  110. package/dist/index-BAvgST_9.d.ts.map +1 -0
  111. package/dist/index-BvnJuTGD.d.ts +68 -0
  112. package/dist/index-BvnJuTGD.d.ts.map +1 -0
  113. package/dist/index-C61Wi7yg.d.ts +547 -0
  114. package/dist/index-C61Wi7yg.d.ts.map +1 -0
  115. package/dist/index-CAPUUKaM.d.ts +335 -0
  116. package/dist/index-CAPUUKaM.d.ts.map +1 -0
  117. package/dist/index-DE5fb3EC.d.ts +2244 -0
  118. package/dist/index-DE5fb3EC.d.ts.map +1 -0
  119. package/dist/index-DSC51roc.d.ts +102 -0
  120. package/dist/index-DSC51roc.d.ts.map +1 -0
  121. package/dist/index.d.ts +3755 -15555
  122. package/dist/index.d.ts.map +1 -0
  123. package/dist/index.js +11182 -11216
  124. package/dist/index.js.map +1 -1
  125. package/dist/integrity-BzRbCHzi.js +424 -0
  126. package/dist/integrity-BzRbCHzi.js.map +1 -0
  127. package/dist/integrity-rmVhXWA7.d.ts +61 -0
  128. package/dist/integrity-rmVhXWA7.d.ts.map +1 -0
  129. package/dist/judge-calibration-DFtEMlde.d.ts +146 -0
  130. package/dist/judge-calibration-DFtEMlde.d.ts.map +1 -0
  131. package/dist/ledger-core/index.d.ts +2 -0
  132. package/dist/ledger-core/index.js +2 -0
  133. package/dist/ledger-core-DtZz1RG0.js +388 -0
  134. package/dist/ledger-core-DtZz1RG0.js.map +1 -0
  135. package/dist/llm-client--GR4JbZE.js +687 -0
  136. package/dist/llm-client--GR4JbZE.js.map +1 -0
  137. package/dist/llm-client-B_nIBlYo.d.ts +290 -0
  138. package/dist/llm-client-B_nIBlYo.d.ts.map +1 -0
  139. package/dist/matrix/index.d.ts +3 -155
  140. package/dist/matrix/index.js +2 -8
  141. package/dist/matrix-BzQnu2S6.js +270 -0
  142. package/dist/matrix-BzQnu2S6.js.map +1 -0
  143. package/dist/meta-eval/index.d.ts +4 -1027
  144. package/dist/meta-eval/index.js +393 -390
  145. package/dist/meta-eval/index.js.map +1 -1
  146. package/dist/metrics-C9YY1OcL.js +239 -0
  147. package/dist/metrics-C9YY1OcL.js.map +1 -0
  148. package/dist/mint-yN2M2eh0.js +201 -0
  149. package/dist/mint-yN2M2eh0.js.map +1 -0
  150. package/dist/multi-layer-verifier-BHY1gWAc.d.ts +138 -0
  151. package/dist/multi-layer-verifier-BHY1gWAc.d.ts.map +1 -0
  152. package/dist/multishot/index.d.ts +273 -480
  153. package/dist/multishot/index.d.ts.map +1 -0
  154. package/dist/multishot/index.js +595 -548
  155. package/dist/multishot/index.js.map +1 -1
  156. package/dist/off-policy-DvgzvtIx.js +220 -0
  157. package/dist/off-policy-DvgzvtIx.js.map +1 -0
  158. package/dist/off-policy-mskQw8Mb.d.ts +153 -0
  159. package/dist/off-policy-mskQw8Mb.d.ts.map +1 -0
  160. package/dist/openapi.json +1 -1
  161. package/dist/opencode-sqlite-BGrHeDu3.js +318 -0
  162. package/dist/opencode-sqlite-BGrHeDu3.js.map +1 -0
  163. package/dist/outcome-store-BYHIuO0e.d.ts +65 -0
  164. package/dist/outcome-store-BYHIuO0e.d.ts.map +1 -0
  165. package/dist/outcome-store-ChBKlTd_.js +75 -0
  166. package/dist/outcome-store-ChBKlTd_.js.map +1 -0
  167. package/dist/paired-arms-D9D0wXj2.js +260 -0
  168. package/dist/paired-arms-D9D0wXj2.js.map +1 -0
  169. package/dist/pipelines/index.d.ts +95 -532
  170. package/dist/pipelines/index.d.ts.map +1 -0
  171. package/dist/pipelines/index.js +497 -478
  172. package/dist/pipelines/index.js.map +1 -1
  173. package/dist/pre-registration-DakwTRXk.js +96 -0
  174. package/dist/pre-registration-DakwTRXk.js.map +1 -0
  175. package/dist/propose-review-control-LhIWGzHl.js +1458 -0
  176. package/dist/propose-review-control-LhIWGzHl.js.map +1 -0
  177. package/dist/query-CJ_DX8vl.d.ts +28 -0
  178. package/dist/query-CJ_DX8vl.d.ts.map +1 -0
  179. package/dist/query-Di7eEQ79.js +83 -0
  180. package/dist/query-Di7eEQ79.js.map +1 -0
  181. package/dist/raw-provider-sink-BQd7mzyT.js +201 -0
  182. package/dist/raw-provider-sink-BQd7mzyT.js.map +1 -0
  183. package/dist/raw-provider-sink-BU29Sh8h.d.ts +134 -0
  184. package/dist/raw-provider-sink-BU29Sh8h.d.ts.map +1 -0
  185. package/dist/redact-7Aq1ukl-.js +107 -0
  186. package/dist/redact-7Aq1ukl-.js.map +1 -0
  187. package/dist/release-report-Cz9NKH39.js +603 -0
  188. package/dist/release-report-Cz9NKH39.js.map +1 -0
  189. package/dist/release-report-mvB2G4_J.d.ts +244 -0
  190. package/dist/release-report-mvB2G4_J.d.ts.map +1 -0
  191. package/dist/replay-GnyotH0J.js +1741 -0
  192. package/dist/replay-GnyotH0J.js.map +1 -0
  193. package/dist/replay-OoidtG1E.d.ts +749 -0
  194. package/dist/replay-OoidtG1E.d.ts.map +1 -0
  195. package/dist/reporting.d.ts +6 -1312
  196. package/dist/reporting.js +6 -51
  197. package/dist/researcher-CwTdwXG1.d.ts +314 -0
  198. package/dist/researcher-CwTdwXG1.d.ts.map +1 -0
  199. package/dist/reward-hacking-eAnOsynk.d.ts +224 -0
  200. package/dist/reward-hacking-eAnOsynk.d.ts.map +1 -0
  201. package/dist/reward-hacking-qipEpKvY.js +596 -0
  202. package/dist/reward-hacking-qipEpKvY.js.map +1 -0
  203. package/dist/reward-nw2xZGZG.js +137 -0
  204. package/dist/reward-nw2xZGZG.js.map +1 -0
  205. package/dist/rl.d.ts +760 -4010
  206. package/dist/rl.d.ts.map +1 -0
  207. package/dist/rl.js +2325 -1958
  208. package/dist/rl.js.map +1 -1
  209. package/dist/rolldown-runtime-8H4AJuhK.js +14 -0
  210. package/dist/rollout/index.d.ts +3 -2087
  211. package/dist/rollout/index.js +8 -168
  212. package/dist/rollout-BOYjemfR.js +624 -0
  213. package/dist/rollout-BOYjemfR.js.map +1 -0
  214. package/dist/rubric-predictive-validity-B3xmbmS1.js +141 -0
  215. package/dist/rubric-predictive-validity-B3xmbmS1.js.map +1 -0
  216. package/dist/rubric-predictive-validity-Ku_clp_1.d.ts +74 -0
  217. package/dist/rubric-predictive-validity-Ku_clp_1.d.ts.map +1 -0
  218. package/dist/run-evidence-ClRX_8A9.d.ts +225 -0
  219. package/dist/run-evidence-ClRX_8A9.d.ts.map +1 -0
  220. package/dist/run-record-BuoE80Dq.js +467 -0
  221. package/dist/{chunk-56TAVBOK.js.map → run-record-BuoE80Dq.js.map} +1 -1
  222. package/dist/run-record-CnZu_gjl.d.ts +357 -0
  223. package/dist/run-record-CnZu_gjl.d.ts.map +1 -0
  224. package/dist/run-score-iEEAWiBY.js +41 -0
  225. package/dist/run-score-iEEAWiBY.js.map +1 -0
  226. package/dist/runtime-trajectory-1gyaTOoC.js +93 -0
  227. package/dist/runtime-trajectory-1gyaTOoC.js.map +1 -0
  228. package/dist/runtime-trajectory-BvSZcCHD.d.ts +50 -0
  229. package/dist/runtime-trajectory-BvSZcCHD.d.ts.map +1 -0
  230. package/dist/schema-BtVldJ3T.d.ts +206 -0
  231. package/dist/schema-BtVldJ3T.d.ts.map +1 -0
  232. package/dist/schema-C6DW4ZHR.js +821 -0
  233. package/dist/schema-C6DW4ZHR.js.map +1 -0
  234. package/dist/schema-CRhEY1SO.js +69 -0
  235. package/dist/schema-CRhEY1SO.js.map +1 -0
  236. package/dist/schema-Cef2cFmb.d.ts +408 -0
  237. package/dist/schema-Cef2cFmb.d.ts.map +1 -0
  238. package/dist/semantic-concept-judge-B6cWNJ2K.js +725 -0
  239. package/dist/semantic-concept-judge-B6cWNJ2K.js.map +1 -0
  240. package/dist/sequential-Br0mAPHA.js +148 -0
  241. package/dist/sequential-Br0mAPHA.js.map +1 -0
  242. package/dist/sequential-CYwq6Ff_.d.ts +141 -0
  243. package/dist/sequential-CYwq6Ff_.d.ts.map +1 -0
  244. package/dist/series-convergence-CjO2QdRW.js +43 -0
  245. package/dist/series-convergence-CjO2QdRW.js.map +1 -0
  246. package/dist/series-convergence-ofsqPWhs.d.ts +35 -0
  247. package/dist/series-convergence-ofsqPWhs.d.ts.map +1 -0
  248. package/dist/server-m5D9cvnG.js +1040 -0
  249. package/dist/server-m5D9cvnG.js.map +1 -0
  250. package/dist/skill-usage-C_pXm7lP.d.ts +534 -0
  251. package/dist/skill-usage-C_pXm7lP.d.ts.map +1 -0
  252. package/dist/skillopt-optimization-method-B7wX7XkF.d.ts +1739 -0
  253. package/dist/skillopt-optimization-method-B7wX7XkF.d.ts.map +1 -0
  254. package/dist/skillopt-optimization-method-D4ODwFVV.js +7712 -0
  255. package/dist/skillopt-optimization-method-D4ODwFVV.js.map +1 -0
  256. package/dist/statistics-Cmj6nynr.d.ts +514 -0
  257. package/dist/statistics-Cmj6nynr.d.ts.map +1 -0
  258. package/dist/statistics-CnnxdpOg.js +1437 -0
  259. package/dist/statistics-CnnxdpOg.js.map +1 -0
  260. package/dist/store-CT9YIIve.d.ts +117 -0
  261. package/dist/store-CT9YIIve.d.ts.map +1 -0
  262. package/dist/store-CxJry_cs.d.ts +229 -0
  263. package/dist/store-CxJry_cs.d.ts.map +1 -0
  264. package/dist/storyboard/index.d.ts +84 -203
  265. package/dist/storyboard/index.d.ts.map +1 -0
  266. package/dist/storyboard/index.js +609 -542
  267. package/dist/storyboard/index.js.map +1 -1
  268. package/dist/summary-report-BNs5nmXI.js +862 -0
  269. package/dist/summary-report-BNs5nmXI.js.map +1 -0
  270. package/dist/summary-report-Cj9gdw4i.d.ts +407 -0
  271. package/dist/summary-report-Cj9gdw4i.d.ts.map +1 -0
  272. package/dist/supervisor-run/index.d.ts +2 -959
  273. package/dist/supervisor-run/index.js +2 -65
  274. package/dist/supervisor-run-_lnTLM3z.js +1679 -0
  275. package/dist/supervisor-run-_lnTLM3z.js.map +1 -0
  276. package/dist/task-failure-attributes-CQZlB3et.js +311 -0
  277. package/dist/task-failure-attributes-CQZlB3et.js.map +1 -0
  278. package/dist/test-graded-scenario-BsqWLmPt.js +318 -0
  279. package/dist/test-graded-scenario-BsqWLmPt.js.map +1 -0
  280. package/dist/test-graded-scenario-D1TaI2va.d.ts +141 -0
  281. package/dist/test-graded-scenario-D1TaI2va.d.ts.map +1 -0
  282. package/dist/tools-BmuN627J.js +1085 -0
  283. package/dist/tools-BmuN627J.js.map +1 -0
  284. package/dist/trace-attributes.d.ts +2 -52
  285. package/dist/trace-attributes.js +131 -61
  286. package/dist/trace-attributes.js.map +1 -1
  287. package/dist/traces.d.ts +12 -2365
  288. package/dist/traces.js +12 -252
  289. package/dist/trajectory-D_7rLrvE.js +56 -0
  290. package/dist/trajectory-D_7rLrvE.js.map +1 -0
  291. package/dist/types-DGsxbAEd.d.ts +387 -0
  292. package/dist/types-DGsxbAEd.d.ts.map +1 -0
  293. package/dist/types-k9tZGKUg.d.ts +640 -0
  294. package/dist/types-k9tZGKUg.d.ts.map +1 -0
  295. package/dist/verdict-Dps8_okt.d.ts +37 -0
  296. package/dist/verdict-Dps8_okt.d.ts.map +1 -0
  297. package/dist/wire/index.d.ts +702 -1173
  298. package/dist/wire/index.d.ts.map +1 -0
  299. package/dist/wire/index.js +2 -81
  300. package/package.json +17 -9
  301. package/dist/benchmarks/index.js.map +0 -1
  302. package/dist/campaign/index.js.map +0 -1
  303. package/dist/chunk-2QU3YOPR.js +0 -7374
  304. package/dist/chunk-2QU3YOPR.js.map +0 -1
  305. package/dist/chunk-3OCR4R5I.js +0 -728
  306. package/dist/chunk-3OCR4R5I.js.map +0 -1
  307. package/dist/chunk-3RF76KTD.js +0 -84
  308. package/dist/chunk-3RF76KTD.js.map +0 -1
  309. package/dist/chunk-56TAVBOK.js +0 -698
  310. package/dist/chunk-5DTSBUL2.js +0 -159
  311. package/dist/chunk-5DTSBUL2.js.map +0 -1
  312. package/dist/chunk-7FO3TNPI.js +0 -232
  313. package/dist/chunk-7FO3TNPI.js.map +0 -1
  314. package/dist/chunk-7ZZMD7UK.js +0 -386
  315. package/dist/chunk-7ZZMD7UK.js.map +0 -1
  316. package/dist/chunk-BOD4O7OF.js +0 -40
  317. package/dist/chunk-BOD4O7OF.js.map +0 -1
  318. package/dist/chunk-BSO5JDQH.js +0 -2335
  319. package/dist/chunk-BSO5JDQH.js.map +0 -1
  320. package/dist/chunk-C6LXANRU.js +0 -1550
  321. package/dist/chunk-C6LXANRU.js.map +0 -1
  322. package/dist/chunk-DODXQREJ.js +0 -752
  323. package/dist/chunk-DODXQREJ.js.map +0 -1
  324. package/dist/chunk-DRYIUNWY.js +0 -622
  325. package/dist/chunk-DRYIUNWY.js.map +0 -1
  326. package/dist/chunk-E7QXT7SX.js +0 -183
  327. package/dist/chunk-E7QXT7SX.js.map +0 -1
  328. package/dist/chunk-EG66UGL4.js +0 -341
  329. package/dist/chunk-EG66UGL4.js.map +0 -1
  330. package/dist/chunk-FXTVJPYD.js +0 -576
  331. package/dist/chunk-FXTVJPYD.js.map +0 -1
  332. package/dist/chunk-G7MGMCZD.js +0 -153
  333. package/dist/chunk-G7MGMCZD.js.map +0 -1
  334. package/dist/chunk-GGE4NNQT.js +0 -65
  335. package/dist/chunk-GGE4NNQT.js.map +0 -1
  336. package/dist/chunk-H23X7XKK.js +0 -181
  337. package/dist/chunk-H23X7XKK.js.map +0 -1
  338. package/dist/chunk-HHWE3POT.js +0 -94
  339. package/dist/chunk-HHWE3POT.js.map +0 -1
  340. package/dist/chunk-HPWUNB47.js +0 -289
  341. package/dist/chunk-HPWUNB47.js.map +0 -1
  342. package/dist/chunk-IYCLP2N2.js +0 -766
  343. package/dist/chunk-IYCLP2N2.js.map +0 -1
  344. package/dist/chunk-JHCHEVET.js +0 -274
  345. package/dist/chunk-JHCHEVET.js.map +0 -1
  346. package/dist/chunk-JQSF5DQT.js +0 -701
  347. package/dist/chunk-JQSF5DQT.js.map +0 -1
  348. package/dist/chunk-K4DBDHLK.js +0 -158
  349. package/dist/chunk-K4DBDHLK.js.map +0 -1
  350. package/dist/chunk-K6N6XJJX.js +0 -306
  351. package/dist/chunk-K6N6XJJX.js.map +0 -1
  352. package/dist/chunk-M4YBQKIJ.js +0 -1040
  353. package/dist/chunk-M4YBQKIJ.js.map +0 -1
  354. package/dist/chunk-MA6HLL3S.js +0 -65
  355. package/dist/chunk-MA6HLL3S.js.map +0 -1
  356. package/dist/chunk-MAZ26DC7.js +0 -99
  357. package/dist/chunk-MAZ26DC7.js.map +0 -1
  358. package/dist/chunk-NPCTHQIO.js +0 -91
  359. package/dist/chunk-NPCTHQIO.js.map +0 -1
  360. package/dist/chunk-NY44NC4A.js +0 -1056
  361. package/dist/chunk-NY44NC4A.js.map +0 -1
  362. package/dist/chunk-OIUOT4QD.js +0 -44
  363. package/dist/chunk-OIUOT4QD.js.map +0 -1
  364. package/dist/chunk-ONWEPEDO.js +0 -57
  365. package/dist/chunk-ONWEPEDO.js.map +0 -1
  366. package/dist/chunk-OWN5NPMC.js +0 -152
  367. package/dist/chunk-OWN5NPMC.js.map +0 -1
  368. package/dist/chunk-P6FYH6K4.js +0 -1161
  369. package/dist/chunk-P6FYH6K4.js.map +0 -1
  370. package/dist/chunk-PC4UYEBM.js +0 -166
  371. package/dist/chunk-PC4UYEBM.js.map +0 -1
  372. package/dist/chunk-PC5DOSM7.js +0 -579
  373. package/dist/chunk-PC5DOSM7.js.map +0 -1
  374. package/dist/chunk-PXE2VKMX.js +0 -140
  375. package/dist/chunk-PXE2VKMX.js.map +0 -1
  376. package/dist/chunk-PZ5AY32C.js +0 -10
  377. package/dist/chunk-PZ5AY32C.js.map +0 -1
  378. package/dist/chunk-QB6BDBP2.js +0 -4464
  379. package/dist/chunk-QB6BDBP2.js.map +0 -1
  380. package/dist/chunk-RXHCETDZ.js +0 -536
  381. package/dist/chunk-RXHCETDZ.js.map +0 -1
  382. package/dist/chunk-RZTMDUO7.js +0 -49
  383. package/dist/chunk-RZTMDUO7.js.map +0 -1
  384. package/dist/chunk-SFLLL76A.js +0 -669
  385. package/dist/chunk-SFLLL76A.js.map +0 -1
  386. package/dist/chunk-SZLVEKMJ.js +0 -1446
  387. package/dist/chunk-SZLVEKMJ.js.map +0 -1
  388. package/dist/chunk-T4SQEITX.js +0 -95
  389. package/dist/chunk-T4SQEITX.js.map +0 -1
  390. package/dist/chunk-T6RLYGAD.js +0 -158
  391. package/dist/chunk-T6RLYGAD.js.map +0 -1
  392. package/dist/chunk-TJVT4QFF.js +0 -911
  393. package/dist/chunk-TJVT4QFF.js.map +0 -1
  394. package/dist/chunk-TQ7LNKZ3.js +0 -136
  395. package/dist/chunk-TQ7LNKZ3.js.map +0 -1
  396. package/dist/chunk-U4L7JRPZ.js +0 -1706
  397. package/dist/chunk-U4L7JRPZ.js.map +0 -1
  398. package/dist/chunk-U4PHLT2N.js +0 -419
  399. package/dist/chunk-U4PHLT2N.js.map +0 -1
  400. package/dist/chunk-VCZ5FQYW.js +0 -928
  401. package/dist/chunk-VCZ5FQYW.js.map +0 -1
  402. package/dist/chunk-VI2UW6B6.js +0 -162
  403. package/dist/chunk-VI2UW6B6.js.map +0 -1
  404. package/dist/chunk-VQMK5FMP.js +0 -247
  405. package/dist/chunk-VQMK5FMP.js.map +0 -1
  406. package/dist/chunk-WGXIEX7P.js +0 -116
  407. package/dist/chunk-WGXIEX7P.js.map +0 -1
  408. package/dist/chunk-WVATSFCP.js +0 -1553
  409. package/dist/chunk-WVATSFCP.js.map +0 -1
  410. package/dist/chunk-X4YIBDER.js +0 -1662
  411. package/dist/chunk-X4YIBDER.js.map +0 -1
  412. package/dist/chunk-YQN4ICPP.js +0 -355
  413. package/dist/chunk-YQN4ICPP.js.map +0 -1
  414. package/dist/chunk-ZET2UAYW.js +0 -89
  415. package/dist/chunk-ZET2UAYW.js.map +0 -1
  416. package/dist/chunk-ZHTZ4EYI.js +0 -1212
  417. package/dist/chunk-ZHTZ4EYI.js.map +0 -1
  418. package/dist/control.js.map +0 -1
  419. package/dist/hosted/index.js.map +0 -1
  420. package/dist/matrix/index.js.map +0 -1
  421. package/dist/reporting.js.map +0 -1
  422. package/dist/rollout/index.js.map +0 -1
  423. package/dist/run-campaign-OJJ7CZF4.js +0 -18
  424. package/dist/run-campaign-OJJ7CZF4.js.map +0 -1
  425. package/dist/supervisor-run/index.js.map +0 -1
  426. package/dist/traces.js.map +0 -1
  427. package/dist/wire/index.js.map +0 -1
package/dist/cli.d.ts CHANGED
@@ -1 +1 @@
1
- #!/usr/bin/env node
1
+ export {}
package/dist/cli.js CHANGED
@@ -1,78 +1,76 @@
1
1
  #!/usr/bin/env node
2
- import {
3
- buildOpenApi,
4
- handleVersion,
5
- runRpcBatch,
6
- runRpcOnce,
7
- startServerAsync
8
- } from "./chunk-NY44NC4A.js";
9
- import {
10
- runRolloutReleaseCli
11
- } from "./chunk-3OCR4R5I.js";
12
- import "./chunk-OWN5NPMC.js";
13
- import "./chunk-PC5DOSM7.js";
14
- import "./chunk-SFLLL76A.js";
15
- import "./chunk-VCZ5FQYW.js";
16
- import "./chunk-VI2UW6B6.js";
17
- import "./chunk-PC4UYEBM.js";
18
- import "./chunk-ONWEPEDO.js";
19
- import "./chunk-PZ5AY32C.js";
20
-
21
- // src/cli.ts
22
- import { writeFileSync } from "fs";
23
-
24
- // src/cli-config.ts
2
+ import { o as runRolloutReleaseCli } from "./hf-dataset-DBJXXoY1.js";
3
+ import { a as runRpcBatch, o as runRpcOnce, p as handleVersion, r as startServerAsync, s as buildOpenApi } from "./server-m5D9cvnG.js";
4
+ import { writeFileSync } from "node:fs";
5
+ //#region src/cli-config.ts
25
6
  function resolveCliLlmConfig(env = process.env) {
26
- const explicitBaseUrl = nonEmpty(env.AGENT_EVAL_LLM_BASE_URL);
27
- const explicitApiKey = nonEmpty(env.AGENT_EVAL_LLM_API_KEY);
28
- const openAiApiKey = nonEmpty(env.OPENAI_API_KEY);
29
- const tangleApiKey = nonEmpty(env.TANGLE_API_KEY);
30
- const baseUrl = explicitBaseUrl ?? nonEmpty(env.OPENAI_BASE_URL) ?? nonEmpty(env.TANGLE_ROUTER_URL) ?? (openAiApiKey ? "https://api.openai.com/v1" : void 0) ?? (tangleApiKey ? "https://router.tangle.tools/v1" : void 0);
31
- const apiKey = explicitApiKey ?? openAiApiKey ?? tangleApiKey;
32
- const model = nonEmpty(env.AGENT_EVAL_LLM_MODEL) ?? nonEmpty(env.OPENAI_MODEL) ?? nonEmpty(env.TANGLE_MODEL);
33
- const client = baseUrl || apiKey ? { ...baseUrl ? { baseUrl } : {}, ...apiKey ? { apiKey } : {} } : void 0;
34
- return { ...client ? { client } : {}, ...model ? { model } : {} };
7
+ const explicitBaseUrl = nonEmpty(env.AGENT_EVAL_LLM_BASE_URL);
8
+ const explicitApiKey = nonEmpty(env.AGENT_EVAL_LLM_API_KEY);
9
+ const openAiApiKey = nonEmpty(env.OPENAI_API_KEY);
10
+ const tangleApiKey = nonEmpty(env.TANGLE_API_KEY);
11
+ const baseUrl = explicitBaseUrl ?? nonEmpty(env.OPENAI_BASE_URL) ?? nonEmpty(env.TANGLE_ROUTER_URL) ?? (openAiApiKey ? "https://api.openai.com/v1" : void 0) ?? (tangleApiKey ? "https://router.tangle.tools/v1" : void 0);
12
+ const apiKey = explicitApiKey ?? openAiApiKey ?? tangleApiKey;
13
+ const model = nonEmpty(env.AGENT_EVAL_LLM_MODEL) ?? nonEmpty(env.OPENAI_MODEL) ?? nonEmpty(env.TANGLE_MODEL);
14
+ const client = baseUrl || apiKey ? {
15
+ ...baseUrl ? { baseUrl } : {},
16
+ ...apiKey ? { apiKey } : {}
17
+ } : void 0;
18
+ return {
19
+ ...client ? { client } : {},
20
+ ...model ? { model } : {}
21
+ };
35
22
  }
36
23
  function nonEmpty(value) {
37
- const trimmed = value?.trim();
38
- return trimmed ? trimmed : void 0;
24
+ const trimmed = value?.trim();
25
+ return trimmed ? trimmed : void 0;
39
26
  }
40
-
41
- // src/cli.ts
27
+ //#endregion
28
+ //#region src/cli.ts
29
+ /**
30
+ * agent-eval CLI.
31
+ *
32
+ * agent-eval serve [--port 5005] [--host 127.0.0.1]
33
+ * agent-eval rpc <method> # one request from stdin → one response on stdout
34
+ * agent-eval rpc-batch <method> # JSONL stdin → JSONL stdout
35
+ * agent-eval openapi [--out path] # write OpenAPI spec
36
+ * agent-eval version
37
+ *
38
+ * <method> is one of: judge, listRubrics, version. When omitted, the
39
+ * stdin payload must be a full {method, params} envelope.
40
+ */
42
41
  function parseArgs(argv) {
43
- const [command, ...rest] = argv;
44
- const positional = [];
45
- const flags = {};
46
- for (let i = 0; i < rest.length; i++) {
47
- const tok = rest[i];
48
- if (tok.startsWith("--")) {
49
- const raw = tok.slice(2);
50
- const equalsAt = raw.indexOf("=");
51
- if (equalsAt >= 0) {
52
- flags[raw.slice(0, equalsAt)] = raw.slice(equalsAt + 1);
53
- continue;
54
- }
55
- const key = raw;
56
- if (key === "help") {
57
- flags[key] = "true";
58
- continue;
59
- }
60
- const next = rest[i + 1];
61
- if (next != null && !next.startsWith("--")) {
62
- flags[key] = next;
63
- i++;
64
- } else {
65
- flags[key] = "true";
66
- }
67
- } else if (tok === "-h") {
68
- flags.help = "true";
69
- } else {
70
- positional.push(tok);
71
- }
72
- }
73
- return { command: command ?? "help", positional, flags };
42
+ const [command, ...rest] = argv;
43
+ const positional = [];
44
+ const flags = {};
45
+ for (let i = 0; i < rest.length; i++) {
46
+ const tok = rest[i];
47
+ if (tok.startsWith("--")) {
48
+ const raw = tok.slice(2);
49
+ const equalsAt = raw.indexOf("=");
50
+ if (equalsAt >= 0) {
51
+ flags[raw.slice(0, equalsAt)] = raw.slice(equalsAt + 1);
52
+ continue;
53
+ }
54
+ const key = raw;
55
+ if (key === "help") {
56
+ flags[key] = "true";
57
+ continue;
58
+ }
59
+ const next = rest[i + 1];
60
+ if (next != null && !next.startsWith("--")) {
61
+ flags[key] = next;
62
+ i++;
63
+ } else flags[key] = "true";
64
+ } else if (tok === "-h") flags.help = "true";
65
+ else positional.push(tok);
66
+ }
67
+ return {
68
+ command: command ?? "help",
69
+ positional,
70
+ flags
71
+ };
74
72
  }
75
- var HELP = `agent-eval: evaluation RPC and HTTP server.
73
+ const HELP = `agent-eval: evaluation RPC and HTTP server.
76
74
 
77
75
  Commands:
78
76
  serve [--port 5005] [--host 127.0.0.1]
@@ -96,119 +94,110 @@ Judge provider:
96
94
 
97
95
  Without arguments, prints this help.`;
98
96
  async function main() {
99
- const { command, positional, flags } = parseArgs(process.argv.slice(2));
100
- assertKnownFlags(command, flags);
101
- if (flags.help === "true") {
102
- process.stdout.write(`${HELP}
103
- `);
104
- return 0;
105
- }
106
- switch (command) {
107
- case "serve": {
108
- const port = parsePort(flags.port ?? "5005");
109
- const host = flags.host ?? "127.0.0.1";
110
- const llm = resolveCliLlmConfig();
111
- const { server } = await startServerAsync({
112
- port,
113
- host,
114
- llm: llm.client,
115
- judgeModel: llm.model,
116
- llmRouteRequirements: { requireExplicitBaseUrl: true }
117
- });
118
- const shutdown = (sig) => {
119
- console.log(`[agent-eval] received ${sig}, shutting down`);
120
- server.close(() => process.exit(0));
121
- setTimeout(() => process.exit(1), 5e3).unref();
122
- };
123
- process.on("SIGINT", () => shutdown("SIGINT"));
124
- process.on("SIGTERM", () => shutdown("SIGTERM"));
125
- await new Promise(() => {
126
- });
127
- return 0;
128
- }
129
- case "rpc": {
130
- const [method] = positional;
131
- const llm = resolveCliLlmConfig();
132
- return await runRpcOnce(method, {
133
- llm: llm.client,
134
- judgeModel: llm.model,
135
- llmRouteRequirements: { requireExplicitBaseUrl: true }
136
- });
137
- }
138
- case "rpc-batch": {
139
- const [method] = positional;
140
- const llm = resolveCliLlmConfig();
141
- return await runRpcBatch(method, {
142
- llm: llm.client,
143
- judgeModel: llm.model,
144
- llmRouteRequirements: { requireExplicitBaseUrl: true }
145
- });
146
- }
147
- case "openapi": {
148
- const out = flags.out ?? "openapi.json";
149
- const spec = buildOpenApi(handleVersion().version);
150
- writeFileSync(out, `${JSON.stringify(spec, null, 2)}
151
- `, "utf-8");
152
- console.log(`[agent-eval] wrote OpenAPI 3.1 spec to ${out}`);
153
- return 0;
154
- }
155
- case "rollout-release": {
156
- return await runRolloutReleaseCli(process.argv.slice(3));
157
- }
158
- case "version": {
159
- process.stdout.write(`${JSON.stringify(handleVersion(), null, 2)}
160
- `);
161
- return 0;
162
- }
163
- case "--version": {
164
- process.stdout.write(`${handleVersion().version}
165
- `);
166
- return 0;
167
- }
168
- case "help":
169
- case "--help":
170
- case "-h":
171
- case "":
172
- process.stdout.write(`${HELP}
173
- `);
174
- return 0;
175
- default:
176
- process.stderr.write(`unknown command: ${command}
177
- ${HELP}
178
- `);
179
- return 1;
180
- }
97
+ const { command, positional, flags } = parseArgs(process.argv.slice(2));
98
+ assertKnownFlags(command, flags);
99
+ if (flags.help === "true") {
100
+ process.stdout.write(`${HELP}\n`);
101
+ return 0;
102
+ }
103
+ switch (command) {
104
+ case "serve": {
105
+ const port = parsePort(flags.port ?? "5005");
106
+ const host = flags.host ?? "127.0.0.1";
107
+ const llm = resolveCliLlmConfig();
108
+ const { server } = await startServerAsync({
109
+ port,
110
+ host,
111
+ llm: llm.client,
112
+ judgeModel: llm.model,
113
+ llmRouteRequirements: { requireExplicitBaseUrl: true }
114
+ });
115
+ const shutdown = (sig) => {
116
+ console.log(`[agent-eval] received ${sig}, shutting down`);
117
+ server.close(() => process.exit(0));
118
+ setTimeout(() => process.exit(1), 5e3).unref();
119
+ };
120
+ process.on("SIGINT", () => shutdown("SIGINT"));
121
+ process.on("SIGTERM", () => shutdown("SIGTERM"));
122
+ await new Promise(() => {});
123
+ return 0;
124
+ }
125
+ case "rpc": {
126
+ const [method] = positional;
127
+ const llm = resolveCliLlmConfig();
128
+ return await runRpcOnce(method, {
129
+ llm: llm.client,
130
+ judgeModel: llm.model,
131
+ llmRouteRequirements: { requireExplicitBaseUrl: true }
132
+ });
133
+ }
134
+ case "rpc-batch": {
135
+ const [method] = positional;
136
+ const llm = resolveCliLlmConfig();
137
+ return await runRpcBatch(method, {
138
+ llm: llm.client,
139
+ judgeModel: llm.model,
140
+ llmRouteRequirements: { requireExplicitBaseUrl: true }
141
+ });
142
+ }
143
+ case "openapi": {
144
+ const out = flags.out ?? "openapi.json";
145
+ const spec = buildOpenApi(handleVersion().version);
146
+ writeFileSync(out, `${JSON.stringify(spec, null, 2)}\n`, "utf-8");
147
+ console.log(`[agent-eval] wrote OpenAPI 3.1 spec to ${out}`);
148
+ return 0;
149
+ }
150
+ case "rollout-release": return await runRolloutReleaseCli(process.argv.slice(3));
151
+ case "version":
152
+ process.stdout.write(`${JSON.stringify(handleVersion(), null, 2)}\n`);
153
+ return 0;
154
+ case "--version":
155
+ process.stdout.write(`${handleVersion().version}\n`);
156
+ return 0;
157
+ case "help":
158
+ case "--help":
159
+ case "-h":
160
+ case "":
161
+ process.stdout.write(`${HELP}\n`);
162
+ return 0;
163
+ default:
164
+ process.stderr.write(`unknown command: ${command}\n${HELP}\n`);
165
+ return 1;
166
+ }
181
167
  }
182
- var FLAGS_BY_COMMAND = {
183
- serve: /* @__PURE__ */ new Set(["help", "host", "port"]),
184
- rpc: /* @__PURE__ */ new Set(["help"]),
185
- "rpc-batch": /* @__PURE__ */ new Set(["help"]),
186
- openapi: /* @__PURE__ */ new Set(["help", "out"]),
187
- version: /* @__PURE__ */ new Set(["help"]),
188
- help: /* @__PURE__ */ new Set(),
189
- "--help": /* @__PURE__ */ new Set(),
190
- "-h": /* @__PURE__ */ new Set(),
191
- "--version": /* @__PURE__ */ new Set(),
192
- "": /* @__PURE__ */ new Set()
168
+ const FLAGS_BY_COMMAND = {
169
+ serve: /* @__PURE__ */ new Set([
170
+ "help",
171
+ "host",
172
+ "port"
173
+ ]),
174
+ rpc: /* @__PURE__ */ new Set(["help"]),
175
+ "rpc-batch": /* @__PURE__ */ new Set(["help"]),
176
+ openapi: /* @__PURE__ */ new Set(["help", "out"]),
177
+ version: /* @__PURE__ */ new Set(["help"]),
178
+ help: /* @__PURE__ */ new Set(),
179
+ "--help": /* @__PURE__ */ new Set(),
180
+ "-h": /* @__PURE__ */ new Set(),
181
+ "--version": /* @__PURE__ */ new Set(),
182
+ "": /* @__PURE__ */ new Set()
193
183
  };
194
184
  function assertKnownFlags(command, flags) {
195
- if (command === "rollout-release") return;
196
- const allowed = FLAGS_BY_COMMAND[command];
197
- if (!allowed) return;
198
- const unknown = Object.keys(flags).filter((flag) => !allowed.has(flag));
199
- if (unknown.length > 0) {
200
- throw new Error(`unknown flag for ${command || "help"}: --${unknown[0]}`);
201
- }
185
+ if (command === "rollout-release") return;
186
+ const allowed = FLAGS_BY_COMMAND[command];
187
+ if (!allowed) return;
188
+ const unknown = Object.keys(flags).filter((flag) => !allowed.has(flag));
189
+ if (unknown.length > 0) throw new Error(`unknown flag for ${command || "help"}: --${unknown[0]}`);
202
190
  }
203
191
  function parsePort(raw) {
204
- const port = Number(raw);
205
- if (!Number.isInteger(port) || port < 0 || port > 65535) {
206
- throw new Error(`--port must be an integer from 0 to 65535; received ${JSON.stringify(raw)}`);
207
- }
208
- return port;
192
+ const port = Number(raw);
193
+ if (!Number.isInteger(port) || port < 0 || port > 65535) throw new Error(`--port must be an integer from 0 to 65535; received ${JSON.stringify(raw)}`);
194
+ return port;
209
195
  }
210
196
  main().then((code) => process.exit(code)).catch((err) => {
211
- console.error("[agent-eval] cli error:", err);
212
- process.exit(1);
197
+ console.error("[agent-eval] cli error:", err);
198
+ process.exit(1);
213
199
  });
200
+ //#endregion
201
+ export {};
202
+
214
203
  //# sourceMappingURL=cli.js.map
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 { 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":[]}
1
+ {"version":3,"file":"cli.js","names":[],"sources":["../src/cli-config.ts","../src/cli.ts"],"sourcesContent":["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","#!/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"],"mappings":";;;;;AAOA,SAAgB,oBAAoB,MAAyB,QAAQ,KAAmB;CACtF,MAAM,kBAAkB,SAAS,IAAI,uBAAuB;CAC5D,MAAM,iBAAiB,SAAS,IAAI,sBAAsB;CAC1D,MAAM,eAAe,SAAS,IAAI,cAAc;CAChD,MAAM,eAAe,SAAS,IAAI,cAAc;CAChD,MAAM,UACJ,mBACA,SAAS,IAAI,eAAe,KAC5B,SAAS,IAAI,iBAAiB,MAC7B,eAAe,8BAA8B,KAAA,OAC7C,eAAe,mCAAmC,KAAA;CACrD,MAAM,SAAS,kBAAkB,gBAAgB;CACjD,MAAM,QACJ,SAAS,IAAI,oBAAoB,KAAK,SAAS,IAAI,YAAY,KAAK,SAAS,IAAI,YAAY;CAE/F,MAAM,SACJ,WAAW,SACP;EAAE,GAAI,UAAU,EAAE,QAAQ,IAAI,CAAC;EAAI,GAAI,SAAS,EAAE,OAAO,IAAI,CAAC;CAAG,IACjE,KAAA;CACN,OAAO;EAAE,GAAI,SAAS,EAAE,OAAO,IAAI,CAAC;EAAI,GAAI,QAAQ,EAAE,MAAM,IAAI,CAAC;CAAG;AACtE;AAEA,SAAS,SAAS,OAA+C;CAC/D,MAAM,UAAU,OAAO,KAAK;CAC5B,OAAO,UAAU,UAAU,KAAA;AAC7B;;;;;;;;;;;;;;;ACLA,SAAS,UAAU,MAAsB;CACvC,MAAM,CAAC,SAAS,GAAG,QAAQ;CAC3B,MAAM,aAAuB,CAAC;CAC9B,MAAM,QAAgC,CAAC;CACvC,KAAK,IAAI,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK;EACpC,MAAM,MAAM,KAAK;EACjB,IAAI,IAAI,WAAW,IAAI,GAAG;GACxB,MAAM,MAAM,IAAI,MAAM,CAAC;GACvB,MAAM,WAAW,IAAI,QAAQ,GAAG;GAChC,IAAI,YAAY,GAAG;IACjB,MAAM,IAAI,MAAM,GAAG,QAAQ,KAAK,IAAI,MAAM,WAAW,CAAC;IACtD;GACF;GACA,MAAM,MAAM;GACZ,IAAI,QAAQ,QAAQ;IAClB,MAAM,OAAO;IACb;GACF;GACA,MAAM,OAAO,KAAK,IAAI;GACtB,IAAI,QAAQ,QAAQ,CAAC,KAAK,WAAW,IAAI,GAAG;IAC1C,MAAM,OAAO;IACb;GACF,OACE,MAAM,OAAO;EAEjB,OAAO,IAAI,QAAQ,MACjB,MAAM,OAAO;OAEb,WAAW,KAAK,GAAG;CAEvB;CACA,OAAO;EAAE,SAAS,WAAW;EAAQ;EAAY;CAAM;AACzD;AAEA,MAAM,OAAO;;;;;;;;;;;;;;;;;;;;;;;AAwBb,eAAe,OAAwB;CACrC,MAAM,EAAE,SAAS,YAAY,UAAU,UAAU,QAAQ,KAAK,MAAM,CAAC,CAAC;CACtE,iBAAiB,SAAS,KAAK;CAE/B,IAAI,MAAM,SAAS,QAAQ;EACzB,QAAQ,OAAO,MAAM,GAAG,KAAK,GAAG;EAChC,OAAO;CACT;CAEA,QAAQ,SAAR;EACE,KAAK,SAAS;GACZ,MAAM,OAAO,UAAU,MAAM,QAAQ,MAAM;GAC3C,MAAM,OAAO,MAAM,QAAQ;GAC3B,MAAM,MAAM,oBAAoB;GAChC,MAAM,EAAE,WAAW,MAAM,iBAAiB;IACxC;IACA;IACA,KAAK,IAAI;IACT,YAAY,IAAI;IAChB,sBAAsB,EAAE,wBAAwB,KAAK;GACvD,CAAC;GAED,MAAM,YAAY,QAAgB;IAEhC,QAAQ,IAAI,yBAAyB,IAAI,gBAAgB;IACzD,OAAO,YAAY,QAAQ,KAAK,CAAC,CAAC;IAElC,iBAAiB,QAAQ,KAAK,CAAC,GAAG,GAAI,CAAC,CAAC,MAAM;GAChD;GACA,QAAQ,GAAG,gBAAgB,SAAS,QAAQ,CAAC;GAC7C,QAAQ,GAAG,iBAAiB,SAAS,SAAS,CAAC;GAE/C,MAAM,IAAI,cAAc,CAAC,CAAC;GAC1B,OAAO;EACT;EACA,KAAK,OAAO;GACV,MAAM,CAAC,UAAU;GACjB,MAAM,MAAM,oBAAoB;GAChC,OAAO,MAAM,WAAW,QAAQ;IAC9B,KAAK,IAAI;IACT,YAAY,IAAI;IAChB,sBAAsB,EAAE,wBAAwB,KAAK;GACvD,CAAC;EACH;EACA,KAAK,aAAa;GAChB,MAAM,CAAC,UAAU;GACjB,MAAM,MAAM,oBAAoB;GAChC,OAAO,MAAM,YAAY,QAAQ;IAC/B,KAAK,IAAI;IACT,YAAY,IAAI;IAChB,sBAAsB,EAAE,wBAAwB,KAAK;GACvD,CAAC;EACH;EACA,KAAK,WAAW;GACd,MAAM,MAAM,MAAM,OAAO;GACzB,MAAM,OAAO,aAAa,cAAc,CAAC,CAAC,OAAO;GACjD,cAAc,KAAK,GAAG,KAAK,UAAU,MAAM,MAAM,CAAC,EAAE,KAAK,OAAO;GAEhE,QAAQ,IAAI,0CAA0C,KAAK;GAC3D,OAAO;EACT;EACA,KAAK,mBAGH,OAAO,MAAM,qBAAqB,QAAQ,KAAK,MAAM,CAAC,CAAC;EAEzD,KAAK;GACH,QAAQ,OAAO,MAAM,GAAG,KAAK,UAAU,cAAc,GAAG,MAAM,CAAC,EAAE,GAAG;GACpE,OAAO;EAET,KAAK;GACH,QAAQ,OAAO,MAAM,GAAG,cAAc,CAAC,CAAC,QAAQ,GAAG;GACnD,OAAO;EAET,KAAK;EACL,KAAK;EACL,KAAK;EACL,KAAK;GACH,QAAQ,OAAO,MAAM,GAAG,KAAK,GAAG;GAChC,OAAO;EACT;GACE,QAAQ,OAAO,MAAM,oBAAoB,QAAQ,IAAI,KAAK,GAAG;GAC7D,OAAO;CACX;AACF;AAEA,MAAM,mBAAwD;CAC5D,uBAAO,IAAI,IAAI;EAAC;EAAQ;EAAQ;CAAM,CAAC;CACvC,qBAAK,IAAI,IAAI,CAAC,MAAM,CAAC;CACrB,6BAAa,IAAI,IAAI,CAAC,MAAM,CAAC;CAC7B,yBAAS,IAAI,IAAI,CAAC,QAAQ,KAAK,CAAC;CAChC,yBAAS,IAAI,IAAI,CAAC,MAAM,CAAC;CACzB,sBAAM,IAAI,IAAI;CACd,0BAAU,IAAI,IAAI;CAClB,sBAAM,IAAI,IAAI;CACd,6BAAa,IAAI,IAAI;CACrB,oBAAI,IAAI,IAAI;AACd;AAEA,SAAS,iBAAiB,SAAiB,OAAqC;CAE9E,IAAI,YAAY,mBAAmB;CACnC,MAAM,UAAU,iBAAiB;CACjC,IAAI,CAAC,SAAS;CACd,MAAM,UAAU,OAAO,KAAK,KAAK,CAAC,CAAC,QAAQ,SAAS,CAAC,QAAQ,IAAI,IAAI,CAAC;CACtE,IAAI,QAAQ,SAAS,GACnB,MAAM,IAAI,MAAM,oBAAoB,WAAW,OAAO,MAAM,QAAQ,IAAI;AAE5E;AAEA,SAAS,UAAU,KAAqB;CACtC,MAAM,OAAO,OAAO,GAAG;CACvB,IAAI,CAAC,OAAO,UAAU,IAAI,KAAK,OAAO,KAAK,OAAO,OAChD,MAAM,IAAI,MAAM,uDAAuD,KAAK,UAAU,GAAG,GAAG;CAE9F,OAAO;AACT;AAEA,KAAK,CAAC,CACH,MAAM,SAAS,QAAQ,KAAK,IAAI,CAAC,CAAC,CAClC,OAAO,QAAQ;CAEd,QAAQ,MAAM,2BAA2B,GAAG;CAC5C,QAAQ,KAAK,CAAC;AAChB,CAAC"}