@meetkai/mka1 0.54.22 → 0.54.24

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 (497) hide show
  1. package/README.md +302 -285
  2. package/bin/mcp-server.js +5917 -3364
  3. package/bin/mcp-server.js.map +47 -18
  4. package/dist/commonjs/funcs/llmEvalsAppendAgentTrials.d.ts +19 -0
  5. package/dist/commonjs/funcs/llmEvalsAppendAgentTrials.d.ts.map +1 -0
  6. package/dist/commonjs/funcs/llmEvalsAppendAgentTrials.js +136 -0
  7. package/dist/commonjs/funcs/llmEvalsAppendAgentTrials.js.map +1 -0
  8. package/dist/commonjs/funcs/llmEvalsGetRunOverview.d.ts +20 -0
  9. package/dist/commonjs/funcs/llmEvalsGetRunOverview.d.ts.map +1 -0
  10. package/dist/commonjs/funcs/llmEvalsGetRunOverview.js +138 -0
  11. package/dist/commonjs/funcs/llmEvalsGetRunOverview.js.map +1 -0
  12. package/dist/commonjs/funcs/llmEvalsGetSampleTrajectory.d.ts +19 -0
  13. package/dist/commonjs/funcs/llmEvalsGetSampleTrajectory.d.ts.map +1 -0
  14. package/dist/commonjs/funcs/llmEvalsGetSampleTrajectory.js +142 -0
  15. package/dist/commonjs/funcs/llmEvalsGetSampleTrajectory.js.map +1 -0
  16. package/dist/commonjs/funcs/llmEvalsGetSuiteLeaderboard.d.ts +20 -0
  17. package/dist/commonjs/funcs/llmEvalsGetSuiteLeaderboard.d.ts.map +1 -0
  18. package/dist/commonjs/funcs/llmEvalsGetSuiteLeaderboard.js +142 -0
  19. package/dist/commonjs/funcs/llmEvalsGetSuiteLeaderboard.js.map +1 -0
  20. package/dist/commonjs/funcs/llmEvalsImportAgentRun.d.ts +19 -0
  21. package/dist/commonjs/funcs/llmEvalsImportAgentRun.d.ts.map +1 -0
  22. package/dist/commonjs/funcs/llmEvalsImportAgentRun.js +130 -0
  23. package/dist/commonjs/funcs/llmEvalsImportAgentRun.js.map +1 -0
  24. package/dist/commonjs/funcs/llmEvalsUploadSampleTrajectory.d.ts +20 -0
  25. package/dist/commonjs/funcs/llmEvalsUploadSampleTrajectory.d.ts.map +1 -0
  26. package/dist/commonjs/funcs/llmEvalsUploadSampleTrajectory.js +146 -0
  27. package/dist/commonjs/funcs/llmEvalsUploadSampleTrajectory.js.map +1 -0
  28. package/dist/commonjs/funcs/llmResponsesCreate.d.ts +25 -0
  29. package/dist/commonjs/funcs/llmResponsesCreate.d.ts.map +1 -0
  30. package/dist/commonjs/funcs/llmResponsesCreate.js +137 -0
  31. package/dist/commonjs/funcs/llmResponsesCreate.js.map +1 -0
  32. package/dist/commonjs/lib/config.d.ts +2 -2
  33. package/dist/commonjs/lib/config.js +2 -2
  34. package/dist/commonjs/mcp-server/mcp-server.js +1 -1
  35. package/dist/commonjs/mcp-server/server.d.ts.map +1 -1
  36. package/dist/commonjs/mcp-server/server.js +15 -1
  37. package/dist/commonjs/mcp-server/server.js.map +1 -1
  38. package/dist/commonjs/mcp-server/tools/llmEvalsAppendAgentTrials.d.ts +8 -0
  39. package/dist/commonjs/mcp-server/tools/llmEvalsAppendAgentTrials.d.ts.map +1 -0
  40. package/dist/commonjs/mcp-server/tools/llmEvalsAppendAgentTrials.js +64 -0
  41. package/dist/commonjs/mcp-server/tools/llmEvalsAppendAgentTrials.js.map +1 -0
  42. package/dist/commonjs/mcp-server/tools/llmEvalsGetRunOverview.d.ts +8 -0
  43. package/dist/commonjs/mcp-server/tools/llmEvalsGetRunOverview.d.ts.map +1 -0
  44. package/dist/commonjs/mcp-server/tools/llmEvalsGetRunOverview.js +64 -0
  45. package/dist/commonjs/mcp-server/tools/llmEvalsGetRunOverview.js.map +1 -0
  46. package/dist/commonjs/mcp-server/tools/llmEvalsGetSampleTrajectory.d.ts +8 -0
  47. package/dist/commonjs/mcp-server/tools/llmEvalsGetSampleTrajectory.d.ts.map +1 -0
  48. package/dist/commonjs/mcp-server/tools/llmEvalsGetSampleTrajectory.js +64 -0
  49. package/dist/commonjs/mcp-server/tools/llmEvalsGetSampleTrajectory.js.map +1 -0
  50. package/dist/commonjs/mcp-server/tools/llmEvalsGetSuiteLeaderboard.d.ts +8 -0
  51. package/dist/commonjs/mcp-server/tools/llmEvalsGetSuiteLeaderboard.d.ts.map +1 -0
  52. package/dist/commonjs/mcp-server/tools/llmEvalsGetSuiteLeaderboard.js +64 -0
  53. package/dist/commonjs/mcp-server/tools/llmEvalsGetSuiteLeaderboard.js.map +1 -0
  54. package/dist/commonjs/mcp-server/tools/llmEvalsImportAgentRun.d.ts +8 -0
  55. package/dist/commonjs/mcp-server/tools/llmEvalsImportAgentRun.d.ts.map +1 -0
  56. package/dist/commonjs/mcp-server/tools/llmEvalsImportAgentRun.js +64 -0
  57. package/dist/commonjs/mcp-server/tools/llmEvalsImportAgentRun.js.map +1 -0
  58. package/dist/commonjs/mcp-server/tools/llmEvalsUploadSampleTrajectory.d.ts +8 -0
  59. package/dist/commonjs/mcp-server/tools/llmEvalsUploadSampleTrajectory.d.ts.map +1 -0
  60. package/dist/commonjs/mcp-server/tools/llmEvalsUploadSampleTrajectory.js +64 -0
  61. package/dist/commonjs/mcp-server/tools/llmEvalsUploadSampleTrajectory.js.map +1 -0
  62. package/dist/commonjs/mcp-server/tools/llmResponsesCreate.d.ts +8 -0
  63. package/dist/commonjs/mcp-server/tools/llmResponsesCreate.d.ts.map +1 -0
  64. package/dist/commonjs/mcp-server/tools/llmResponsesCreate.js +64 -0
  65. package/dist/commonjs/mcp-server/tools/llmResponsesCreate.js.map +1 -0
  66. package/dist/commonjs/models/components/appendagenttrialsrequest.d.ts +111 -0
  67. package/dist/commonjs/models/components/appendagenttrialsrequest.d.ts.map +1 -0
  68. package/dist/commonjs/models/components/appendagenttrialsrequest.js +136 -0
  69. package/dist/commonjs/models/components/appendagenttrialsrequest.js.map +1 -0
  70. package/dist/commonjs/models/components/evalleaderboardrow.d.ts +85 -0
  71. package/dist/commonjs/models/components/evalleaderboardrow.d.ts.map +1 -0
  72. package/dist/commonjs/models/components/evalleaderboardrow.js +129 -0
  73. package/dist/commonjs/models/components/evalleaderboardrow.js.map +1 -0
  74. package/dist/commonjs/models/components/evalrunoverview.d.ts +101 -0
  75. package/dist/commonjs/models/components/evalrunoverview.d.ts.map +1 -0
  76. package/dist/commonjs/models/components/evalrunoverview.js +150 -0
  77. package/dist/commonjs/models/components/evalrunoverview.js.map +1 -0
  78. package/dist/commonjs/models/components/evalsampleobject.d.ts +32 -0
  79. package/dist/commonjs/models/components/evalsampleobject.d.ts.map +1 -1
  80. package/dist/commonjs/models/components/evalsampleobject.js +20 -0
  81. package/dist/commonjs/models/components/evalsampleobject.js.map +1 -1
  82. package/dist/commonjs/models/components/evalsampletrajectoryuploadedobject.d.ts +32 -0
  83. package/dist/commonjs/models/components/evalsampletrajectoryuploadedobject.d.ts.map +1 -0
  84. package/dist/commonjs/models/components/evalsampletrajectoryuploadedobject.js +87 -0
  85. package/dist/commonjs/models/components/evalsampletrajectoryuploadedobject.js.map +1 -0
  86. package/dist/commonjs/models/components/evalsuiteleaderboard.d.ts +35 -0
  87. package/dist/commonjs/models/components/evalsuiteleaderboard.d.ts.map +1 -0
  88. package/dist/commonjs/models/components/evalsuiteleaderboard.js +78 -0
  89. package/dist/commonjs/models/components/evalsuiteleaderboard.js.map +1 -0
  90. package/dist/commonjs/models/components/importagentrunrequest.d.ts +164 -0
  91. package/dist/commonjs/models/components/importagentrunrequest.d.ts.map +1 -0
  92. package/dist/commonjs/models/components/importagentrunrequest.js +182 -0
  93. package/dist/commonjs/models/components/importagentrunrequest.js.map +1 -0
  94. package/dist/commonjs/models/components/index.d.ts +8 -0
  95. package/dist/commonjs/models/components/index.d.ts.map +1 -1
  96. package/dist/commonjs/models/components/index.js +8 -0
  97. package/dist/commonjs/models/components/index.js.map +1 -1
  98. package/dist/commonjs/models/components/responsescreaterequest.d.ts +432 -0
  99. package/dist/commonjs/models/components/responsescreaterequest.d.ts.map +1 -0
  100. package/dist/commonjs/models/components/responsescreaterequest.js +455 -0
  101. package/dist/commonjs/models/components/responsescreaterequest.js.map +1 -0
  102. package/dist/commonjs/models/components/uploadevalsampletrajectoryrequest.d.ts +34 -0
  103. package/dist/commonjs/models/components/uploadevalsampletrajectoryrequest.d.ts.map +1 -0
  104. package/dist/commonjs/models/components/uploadevalsampletrajectoryrequest.js +71 -0
  105. package/dist/commonjs/models/components/uploadevalsampletrajectoryrequest.js.map +1 -0
  106. package/dist/commonjs/models/operations/appendagentevaltrials.d.ts +264 -0
  107. package/dist/commonjs/models/operations/appendagentevaltrials.d.ts.map +1 -0
  108. package/dist/commonjs/models/operations/appendagentevaltrials.js +371 -0
  109. package/dist/commonjs/models/operations/appendagentevaltrials.js.map +1 -0
  110. package/dist/commonjs/models/operations/cancelevalrun.d.ts +20 -0
  111. package/dist/commonjs/models/operations/cancelevalrun.d.ts.map +1 -1
  112. package/dist/commonjs/models/operations/cancelevalrun.js +16 -0
  113. package/dist/commonjs/models/operations/cancelevalrun.js.map +1 -1
  114. package/dist/commonjs/models/operations/createevalrun.d.ts +20 -0
  115. package/dist/commonjs/models/operations/createevalrun.d.ts.map +1 -1
  116. package/dist/commonjs/models/operations/createevalrun.js +16 -0
  117. package/dist/commonjs/models/operations/createevalrun.js.map +1 -1
  118. package/dist/commonjs/models/operations/createresponse.d.ts +64 -0
  119. package/dist/commonjs/models/operations/createresponse.d.ts.map +1 -0
  120. package/dist/commonjs/models/operations/createresponse.js +128 -0
  121. package/dist/commonjs/models/operations/createresponse.js.map +1 -0
  122. package/dist/commonjs/models/operations/getevalrun.d.ts +20 -0
  123. package/dist/commonjs/models/operations/getevalrun.d.ts.map +1 -1
  124. package/dist/commonjs/models/operations/getevalrun.js +16 -0
  125. package/dist/commonjs/models/operations/getevalrun.js.map +1 -1
  126. package/dist/commonjs/models/operations/getevalrunoverview.d.ts +24 -0
  127. package/dist/commonjs/models/operations/getevalrunoverview.d.ts.map +1 -0
  128. package/dist/commonjs/models/operations/getevalrunoverview.js +73 -0
  129. package/dist/commonjs/models/operations/getevalrunoverview.js.map +1 -0
  130. package/dist/commonjs/models/operations/getevalsampletrajectory.d.ts +75 -0
  131. package/dist/commonjs/models/operations/getevalsampletrajectory.d.ts.map +1 -0
  132. package/dist/commonjs/models/operations/getevalsampletrajectory.js +133 -0
  133. package/dist/commonjs/models/operations/getevalsampletrajectory.js.map +1 -0
  134. package/dist/commonjs/models/operations/getevalsuiteleaderboard.d.ts +32 -0
  135. package/dist/commonjs/models/operations/getevalsuiteleaderboard.d.ts.map +1 -0
  136. package/dist/commonjs/models/operations/getevalsuiteleaderboard.js +83 -0
  137. package/dist/commonjs/models/operations/getevalsuiteleaderboard.js.map +1 -0
  138. package/dist/commonjs/models/operations/importagentevalrun.d.ts +262 -0
  139. package/dist/commonjs/models/operations/importagentevalrun.d.ts.map +1 -0
  140. package/dist/commonjs/models/operations/importagentevalrun.js +367 -0
  141. package/dist/commonjs/models/operations/importagentevalrun.js.map +1 -0
  142. package/dist/commonjs/models/operations/index.d.ts +7 -0
  143. package/dist/commonjs/models/operations/index.d.ts.map +1 -1
  144. package/dist/commonjs/models/operations/index.js +7 -0
  145. package/dist/commonjs/models/operations/index.js.map +1 -1
  146. package/dist/commonjs/models/operations/listevalruns.d.ts +20 -0
  147. package/dist/commonjs/models/operations/listevalruns.d.ts.map +1 -1
  148. package/dist/commonjs/models/operations/listevalruns.js +16 -0
  149. package/dist/commonjs/models/operations/listevalruns.js.map +1 -1
  150. package/dist/commonjs/models/operations/listevalscheduleruns.d.ts +20 -0
  151. package/dist/commonjs/models/operations/listevalscheduleruns.d.ts.map +1 -1
  152. package/dist/commonjs/models/operations/listevalscheduleruns.js +16 -0
  153. package/dist/commonjs/models/operations/listevalscheduleruns.js.map +1 -1
  154. package/dist/commonjs/models/operations/listevalsuiteleaderboardruns.d.ts +20 -0
  155. package/dist/commonjs/models/operations/listevalsuiteleaderboardruns.d.ts.map +1 -1
  156. package/dist/commonjs/models/operations/listevalsuiteleaderboardruns.js +16 -0
  157. package/dist/commonjs/models/operations/listevalsuiteleaderboardruns.js.map +1 -1
  158. package/dist/commonjs/models/operations/rerunfailedevalsamples.d.ts +20 -0
  159. package/dist/commonjs/models/operations/rerunfailedevalsamples.d.ts.map +1 -1
  160. package/dist/commonjs/models/operations/rerunfailedevalsamples.js +16 -0
  161. package/dist/commonjs/models/operations/rerunfailedevalsamples.js.map +1 -1
  162. package/dist/commonjs/models/operations/retryfailedevalrun.d.ts +20 -0
  163. package/dist/commonjs/models/operations/retryfailedevalrun.d.ts.map +1 -1
  164. package/dist/commonjs/models/operations/retryfailedevalrun.js +16 -0
  165. package/dist/commonjs/models/operations/retryfailedevalrun.js.map +1 -1
  166. package/dist/commonjs/models/operations/uploadevalsampletrajectory.d.ts +31 -0
  167. package/dist/commonjs/models/operations/uploadevalsampletrajectory.d.ts.map +1 -0
  168. package/dist/commonjs/models/operations/uploadevalsampletrajectory.js +86 -0
  169. package/dist/commonjs/models/operations/uploadevalsampletrajectory.js.map +1 -0
  170. package/dist/commonjs/react-query/index.d.ts +7 -0
  171. package/dist/commonjs/react-query/index.d.ts.map +1 -1
  172. package/dist/commonjs/react-query/index.js +7 -0
  173. package/dist/commonjs/react-query/index.js.map +1 -1
  174. package/dist/commonjs/react-query/llmEvalsAppendAgentTrials.d.ts +28 -0
  175. package/dist/commonjs/react-query/llmEvalsAppendAgentTrials.d.ts.map +1 -0
  176. package/dist/commonjs/react-query/llmEvalsAppendAgentTrials.js +47 -0
  177. package/dist/commonjs/react-query/llmEvalsAppendAgentTrials.js.map +1 -0
  178. package/dist/commonjs/react-query/llmEvalsGetRunOverview.core.d.ts +16 -0
  179. package/dist/commonjs/react-query/llmEvalsGetRunOverview.core.d.ts.map +1 -0
  180. package/dist/commonjs/react-query/llmEvalsGetRunOverview.core.js +37 -0
  181. package/dist/commonjs/react-query/llmEvalsGetRunOverview.core.js.map +1 -0
  182. package/dist/commonjs/react-query/llmEvalsGetRunOverview.d.ts +40 -0
  183. package/dist/commonjs/react-query/llmEvalsGetRunOverview.d.ts.map +1 -0
  184. package/dist/commonjs/react-query/llmEvalsGetRunOverview.js +60 -0
  185. package/dist/commonjs/react-query/llmEvalsGetRunOverview.js.map +1 -0
  186. package/dist/commonjs/react-query/llmEvalsGetSampleTrajectory.core.d.ts +16 -0
  187. package/dist/commonjs/react-query/llmEvalsGetSampleTrajectory.core.d.ts.map +1 -0
  188. package/dist/commonjs/react-query/llmEvalsGetSampleTrajectory.core.js +45 -0
  189. package/dist/commonjs/react-query/llmEvalsGetSampleTrajectory.core.js.map +1 -0
  190. package/dist/commonjs/react-query/llmEvalsGetSampleTrajectory.d.ts +44 -0
  191. package/dist/commonjs/react-query/llmEvalsGetSampleTrajectory.d.ts.map +1 -0
  192. package/dist/commonjs/react-query/llmEvalsGetSampleTrajectory.js +65 -0
  193. package/dist/commonjs/react-query/llmEvalsGetSampleTrajectory.js.map +1 -0
  194. package/dist/commonjs/react-query/llmEvalsGetSuiteLeaderboard.core.d.ts +20 -0
  195. package/dist/commonjs/react-query/llmEvalsGetSuiteLeaderboard.core.d.ts.map +1 -0
  196. package/dist/commonjs/react-query/llmEvalsGetSuiteLeaderboard.core.js +41 -0
  197. package/dist/commonjs/react-query/llmEvalsGetSuiteLeaderboard.core.js.map +1 -0
  198. package/dist/commonjs/react-query/llmEvalsGetSuiteLeaderboard.d.ts +48 -0
  199. package/dist/commonjs/react-query/llmEvalsGetSuiteLeaderboard.d.ts.map +1 -0
  200. package/dist/commonjs/react-query/llmEvalsGetSuiteLeaderboard.js +65 -0
  201. package/dist/commonjs/react-query/llmEvalsGetSuiteLeaderboard.js.map +1 -0
  202. package/dist/commonjs/react-query/llmEvalsImportAgentRun.d.ts +28 -0
  203. package/dist/commonjs/react-query/llmEvalsImportAgentRun.d.ts.map +1 -0
  204. package/dist/commonjs/react-query/llmEvalsImportAgentRun.js +47 -0
  205. package/dist/commonjs/react-query/llmEvalsImportAgentRun.js.map +1 -0
  206. package/dist/commonjs/react-query/llmEvalsUploadSampleTrajectory.d.ts +29 -0
  207. package/dist/commonjs/react-query/llmEvalsUploadSampleTrajectory.d.ts.map +1 -0
  208. package/dist/commonjs/react-query/llmEvalsUploadSampleTrajectory.js +47 -0
  209. package/dist/commonjs/react-query/llmEvalsUploadSampleTrajectory.js.map +1 -0
  210. package/dist/commonjs/react-query/llmResponsesCreate.d.ts +34 -0
  211. package/dist/commonjs/react-query/llmResponsesCreate.d.ts.map +1 -0
  212. package/dist/commonjs/react-query/llmResponsesCreate.js +50 -0
  213. package/dist/commonjs/react-query/llmResponsesCreate.js.map +1 -0
  214. package/dist/commonjs/sdk/evals.d.ts +42 -0
  215. package/dist/commonjs/sdk/evals.d.ts.map +1 -1
  216. package/dist/commonjs/sdk/evals.js +60 -0
  217. package/dist/commonjs/sdk/evals.js.map +1 -1
  218. package/dist/commonjs/sdk/responses.d.ts +11 -0
  219. package/dist/commonjs/sdk/responses.d.ts.map +1 -1
  220. package/dist/commonjs/sdk/responses.js +13 -1
  221. package/dist/commonjs/sdk/responses.js.map +1 -1
  222. package/dist/esm/funcs/llmEvalsAppendAgentTrials.d.ts +19 -0
  223. package/dist/esm/funcs/llmEvalsAppendAgentTrials.d.ts.map +1 -0
  224. package/dist/esm/funcs/llmEvalsAppendAgentTrials.js +100 -0
  225. package/dist/esm/funcs/llmEvalsAppendAgentTrials.js.map +1 -0
  226. package/dist/esm/funcs/llmEvalsGetRunOverview.d.ts +20 -0
  227. package/dist/esm/funcs/llmEvalsGetRunOverview.d.ts.map +1 -0
  228. package/dist/esm/funcs/llmEvalsGetRunOverview.js +102 -0
  229. package/dist/esm/funcs/llmEvalsGetRunOverview.js.map +1 -0
  230. package/dist/esm/funcs/llmEvalsGetSampleTrajectory.d.ts +19 -0
  231. package/dist/esm/funcs/llmEvalsGetSampleTrajectory.d.ts.map +1 -0
  232. package/dist/esm/funcs/llmEvalsGetSampleTrajectory.js +106 -0
  233. package/dist/esm/funcs/llmEvalsGetSampleTrajectory.js.map +1 -0
  234. package/dist/esm/funcs/llmEvalsGetSuiteLeaderboard.d.ts +20 -0
  235. package/dist/esm/funcs/llmEvalsGetSuiteLeaderboard.d.ts.map +1 -0
  236. package/dist/esm/funcs/llmEvalsGetSuiteLeaderboard.js +106 -0
  237. package/dist/esm/funcs/llmEvalsGetSuiteLeaderboard.js.map +1 -0
  238. package/dist/esm/funcs/llmEvalsImportAgentRun.d.ts +19 -0
  239. package/dist/esm/funcs/llmEvalsImportAgentRun.d.ts.map +1 -0
  240. package/dist/esm/funcs/llmEvalsImportAgentRun.js +94 -0
  241. package/dist/esm/funcs/llmEvalsImportAgentRun.js.map +1 -0
  242. package/dist/esm/funcs/llmEvalsUploadSampleTrajectory.d.ts +20 -0
  243. package/dist/esm/funcs/llmEvalsUploadSampleTrajectory.d.ts.map +1 -0
  244. package/dist/esm/funcs/llmEvalsUploadSampleTrajectory.js +110 -0
  245. package/dist/esm/funcs/llmEvalsUploadSampleTrajectory.js.map +1 -0
  246. package/dist/esm/funcs/llmResponsesCreate.d.ts +25 -0
  247. package/dist/esm/funcs/llmResponsesCreate.d.ts.map +1 -0
  248. package/dist/esm/funcs/llmResponsesCreate.js +100 -0
  249. package/dist/esm/funcs/llmResponsesCreate.js.map +1 -0
  250. package/dist/esm/lib/config.d.ts +2 -2
  251. package/dist/esm/lib/config.js +2 -2
  252. package/dist/esm/mcp-server/mcp-server.js +1 -1
  253. package/dist/esm/mcp-server/server.d.ts.map +1 -1
  254. package/dist/esm/mcp-server/server.js +15 -1
  255. package/dist/esm/mcp-server/server.js.map +1 -1
  256. package/dist/esm/mcp-server/tools/llmEvalsAppendAgentTrials.d.ts +8 -0
  257. package/dist/esm/mcp-server/tools/llmEvalsAppendAgentTrials.d.ts.map +1 -0
  258. package/dist/esm/mcp-server/tools/llmEvalsAppendAgentTrials.js +28 -0
  259. package/dist/esm/mcp-server/tools/llmEvalsAppendAgentTrials.js.map +1 -0
  260. package/dist/esm/mcp-server/tools/llmEvalsGetRunOverview.d.ts +8 -0
  261. package/dist/esm/mcp-server/tools/llmEvalsGetRunOverview.d.ts.map +1 -0
  262. package/dist/esm/mcp-server/tools/llmEvalsGetRunOverview.js +28 -0
  263. package/dist/esm/mcp-server/tools/llmEvalsGetRunOverview.js.map +1 -0
  264. package/dist/esm/mcp-server/tools/llmEvalsGetSampleTrajectory.d.ts +8 -0
  265. package/dist/esm/mcp-server/tools/llmEvalsGetSampleTrajectory.d.ts.map +1 -0
  266. package/dist/esm/mcp-server/tools/llmEvalsGetSampleTrajectory.js +28 -0
  267. package/dist/esm/mcp-server/tools/llmEvalsGetSampleTrajectory.js.map +1 -0
  268. package/dist/esm/mcp-server/tools/llmEvalsGetSuiteLeaderboard.d.ts +8 -0
  269. package/dist/esm/mcp-server/tools/llmEvalsGetSuiteLeaderboard.d.ts.map +1 -0
  270. package/dist/esm/mcp-server/tools/llmEvalsGetSuiteLeaderboard.js +28 -0
  271. package/dist/esm/mcp-server/tools/llmEvalsGetSuiteLeaderboard.js.map +1 -0
  272. package/dist/esm/mcp-server/tools/llmEvalsImportAgentRun.d.ts +8 -0
  273. package/dist/esm/mcp-server/tools/llmEvalsImportAgentRun.d.ts.map +1 -0
  274. package/dist/esm/mcp-server/tools/llmEvalsImportAgentRun.js +28 -0
  275. package/dist/esm/mcp-server/tools/llmEvalsImportAgentRun.js.map +1 -0
  276. package/dist/esm/mcp-server/tools/llmEvalsUploadSampleTrajectory.d.ts +8 -0
  277. package/dist/esm/mcp-server/tools/llmEvalsUploadSampleTrajectory.d.ts.map +1 -0
  278. package/dist/esm/mcp-server/tools/llmEvalsUploadSampleTrajectory.js +28 -0
  279. package/dist/esm/mcp-server/tools/llmEvalsUploadSampleTrajectory.js.map +1 -0
  280. package/dist/esm/mcp-server/tools/llmResponsesCreate.d.ts +8 -0
  281. package/dist/esm/mcp-server/tools/llmResponsesCreate.d.ts.map +1 -0
  282. package/dist/esm/mcp-server/tools/llmResponsesCreate.js +28 -0
  283. package/dist/esm/mcp-server/tools/llmResponsesCreate.js.map +1 -0
  284. package/dist/esm/models/components/appendagenttrialsrequest.d.ts +111 -0
  285. package/dist/esm/models/components/appendagenttrialsrequest.d.ts.map +1 -0
  286. package/dist/esm/models/components/appendagenttrialsrequest.js +96 -0
  287. package/dist/esm/models/components/appendagenttrialsrequest.js.map +1 -0
  288. package/dist/esm/models/components/evalleaderboardrow.d.ts +85 -0
  289. package/dist/esm/models/components/evalleaderboardrow.d.ts.map +1 -0
  290. package/dist/esm/models/components/evalleaderboardrow.js +89 -0
  291. package/dist/esm/models/components/evalleaderboardrow.js.map +1 -0
  292. package/dist/esm/models/components/evalrunoverview.d.ts +101 -0
  293. package/dist/esm/models/components/evalrunoverview.d.ts.map +1 -0
  294. package/dist/esm/models/components/evalrunoverview.js +108 -0
  295. package/dist/esm/models/components/evalrunoverview.js.map +1 -0
  296. package/dist/esm/models/components/evalsampleobject.d.ts +32 -0
  297. package/dist/esm/models/components/evalsampleobject.d.ts.map +1 -1
  298. package/dist/esm/models/components/evalsampleobject.js +20 -0
  299. package/dist/esm/models/components/evalsampleobject.js.map +1 -1
  300. package/dist/esm/models/components/evalsampletrajectoryuploadedobject.d.ts +32 -0
  301. package/dist/esm/models/components/evalsampletrajectoryuploadedobject.d.ts.map +1 -0
  302. package/dist/esm/models/components/evalsampletrajectoryuploadedobject.js +49 -0
  303. package/dist/esm/models/components/evalsampletrajectoryuploadedobject.js.map +1 -0
  304. package/dist/esm/models/components/evalsuiteleaderboard.d.ts +35 -0
  305. package/dist/esm/models/components/evalsuiteleaderboard.d.ts.map +1 -0
  306. package/dist/esm/models/components/evalsuiteleaderboard.js +40 -0
  307. package/dist/esm/models/components/evalsuiteleaderboard.js.map +1 -0
  308. package/dist/esm/models/components/importagentrunrequest.d.ts +164 -0
  309. package/dist/esm/models/components/importagentrunrequest.d.ts.map +1 -0
  310. package/dist/esm/models/components/importagentrunrequest.js +142 -0
  311. package/dist/esm/models/components/importagentrunrequest.js.map +1 -0
  312. package/dist/esm/models/components/index.d.ts +8 -0
  313. package/dist/esm/models/components/index.d.ts.map +1 -1
  314. package/dist/esm/models/components/index.js +8 -0
  315. package/dist/esm/models/components/index.js.map +1 -1
  316. package/dist/esm/models/components/responsescreaterequest.d.ts +432 -0
  317. package/dist/esm/models/components/responsescreaterequest.d.ts.map +1 -0
  318. package/dist/esm/models/components/responsescreaterequest.js +403 -0
  319. package/dist/esm/models/components/responsescreaterequest.js.map +1 -0
  320. package/dist/esm/models/components/uploadevalsampletrajectoryrequest.d.ts +34 -0
  321. package/dist/esm/models/components/uploadevalsampletrajectoryrequest.d.ts.map +1 -0
  322. package/dist/esm/models/components/uploadevalsampletrajectoryrequest.js +33 -0
  323. package/dist/esm/models/components/uploadevalsampletrajectoryrequest.js.map +1 -0
  324. package/dist/esm/models/operations/appendagentevaltrials.d.ts +264 -0
  325. package/dist/esm/models/operations/appendagentevaltrials.d.ts.map +1 -0
  326. package/dist/esm/models/operations/appendagentevaltrials.js +323 -0
  327. package/dist/esm/models/operations/appendagentevaltrials.js.map +1 -0
  328. package/dist/esm/models/operations/cancelevalrun.d.ts +20 -0
  329. package/dist/esm/models/operations/cancelevalrun.d.ts.map +1 -1
  330. package/dist/esm/models/operations/cancelevalrun.js +16 -0
  331. package/dist/esm/models/operations/cancelevalrun.js.map +1 -1
  332. package/dist/esm/models/operations/createevalrun.d.ts +20 -0
  333. package/dist/esm/models/operations/createevalrun.d.ts.map +1 -1
  334. package/dist/esm/models/operations/createevalrun.js +16 -0
  335. package/dist/esm/models/operations/createevalrun.js.map +1 -1
  336. package/dist/esm/models/operations/createresponse.d.ts +64 -0
  337. package/dist/esm/models/operations/createresponse.d.ts.map +1 -0
  338. package/dist/esm/models/operations/createresponse.js +86 -0
  339. package/dist/esm/models/operations/createresponse.js.map +1 -0
  340. package/dist/esm/models/operations/getevalrun.d.ts +20 -0
  341. package/dist/esm/models/operations/getevalrun.d.ts.map +1 -1
  342. package/dist/esm/models/operations/getevalrun.js +16 -0
  343. package/dist/esm/models/operations/getevalrun.js.map +1 -1
  344. package/dist/esm/models/operations/getevalrunoverview.d.ts +24 -0
  345. package/dist/esm/models/operations/getevalrunoverview.d.ts.map +1 -0
  346. package/dist/esm/models/operations/getevalrunoverview.js +35 -0
  347. package/dist/esm/models/operations/getevalrunoverview.js.map +1 -0
  348. package/dist/esm/models/operations/getevalsampletrajectory.d.ts +75 -0
  349. package/dist/esm/models/operations/getevalsampletrajectory.d.ts.map +1 -0
  350. package/dist/esm/models/operations/getevalsampletrajectory.js +93 -0
  351. package/dist/esm/models/operations/getevalsampletrajectory.js.map +1 -0
  352. package/dist/esm/models/operations/getevalsuiteleaderboard.d.ts +32 -0
  353. package/dist/esm/models/operations/getevalsuiteleaderboard.d.ts.map +1 -0
  354. package/dist/esm/models/operations/getevalsuiteleaderboard.js +45 -0
  355. package/dist/esm/models/operations/getevalsuiteleaderboard.js.map +1 -0
  356. package/dist/esm/models/operations/importagentevalrun.d.ts +262 -0
  357. package/dist/esm/models/operations/importagentevalrun.d.ts.map +1 -0
  358. package/dist/esm/models/operations/importagentevalrun.js +319 -0
  359. package/dist/esm/models/operations/importagentevalrun.js.map +1 -0
  360. package/dist/esm/models/operations/index.d.ts +7 -0
  361. package/dist/esm/models/operations/index.d.ts.map +1 -1
  362. package/dist/esm/models/operations/index.js +7 -0
  363. package/dist/esm/models/operations/index.js.map +1 -1
  364. package/dist/esm/models/operations/listevalruns.d.ts +20 -0
  365. package/dist/esm/models/operations/listevalruns.d.ts.map +1 -1
  366. package/dist/esm/models/operations/listevalruns.js +16 -0
  367. package/dist/esm/models/operations/listevalruns.js.map +1 -1
  368. package/dist/esm/models/operations/listevalscheduleruns.d.ts +20 -0
  369. package/dist/esm/models/operations/listevalscheduleruns.d.ts.map +1 -1
  370. package/dist/esm/models/operations/listevalscheduleruns.js +16 -0
  371. package/dist/esm/models/operations/listevalscheduleruns.js.map +1 -1
  372. package/dist/esm/models/operations/listevalsuiteleaderboardruns.d.ts +20 -0
  373. package/dist/esm/models/operations/listevalsuiteleaderboardruns.d.ts.map +1 -1
  374. package/dist/esm/models/operations/listevalsuiteleaderboardruns.js +16 -0
  375. package/dist/esm/models/operations/listevalsuiteleaderboardruns.js.map +1 -1
  376. package/dist/esm/models/operations/rerunfailedevalsamples.d.ts +20 -0
  377. package/dist/esm/models/operations/rerunfailedevalsamples.d.ts.map +1 -1
  378. package/dist/esm/models/operations/rerunfailedevalsamples.js +16 -0
  379. package/dist/esm/models/operations/rerunfailedevalsamples.js.map +1 -1
  380. package/dist/esm/models/operations/retryfailedevalrun.d.ts +20 -0
  381. package/dist/esm/models/operations/retryfailedevalrun.d.ts.map +1 -1
  382. package/dist/esm/models/operations/retryfailedevalrun.js +16 -0
  383. package/dist/esm/models/operations/retryfailedevalrun.js.map +1 -1
  384. package/dist/esm/models/operations/uploadevalsampletrajectory.d.ts +31 -0
  385. package/dist/esm/models/operations/uploadevalsampletrajectory.d.ts.map +1 -0
  386. package/dist/esm/models/operations/uploadevalsampletrajectory.js +48 -0
  387. package/dist/esm/models/operations/uploadevalsampletrajectory.js.map +1 -0
  388. package/dist/esm/react-query/index.d.ts +7 -0
  389. package/dist/esm/react-query/index.d.ts.map +1 -1
  390. package/dist/esm/react-query/index.js +7 -0
  391. package/dist/esm/react-query/index.js.map +1 -1
  392. package/dist/esm/react-query/llmEvalsAppendAgentTrials.d.ts +28 -0
  393. package/dist/esm/react-query/llmEvalsAppendAgentTrials.d.ts.map +1 -0
  394. package/dist/esm/react-query/llmEvalsAppendAgentTrials.js +42 -0
  395. package/dist/esm/react-query/llmEvalsAppendAgentTrials.js.map +1 -0
  396. package/dist/esm/react-query/llmEvalsGetRunOverview.core.d.ts +16 -0
  397. package/dist/esm/react-query/llmEvalsGetRunOverview.core.d.ts.map +1 -0
  398. package/dist/esm/react-query/llmEvalsGetRunOverview.core.js +32 -0
  399. package/dist/esm/react-query/llmEvalsGetRunOverview.core.js.map +1 -0
  400. package/dist/esm/react-query/llmEvalsGetRunOverview.d.ts +40 -0
  401. package/dist/esm/react-query/llmEvalsGetRunOverview.d.ts.map +1 -0
  402. package/dist/esm/react-query/llmEvalsGetRunOverview.js +50 -0
  403. package/dist/esm/react-query/llmEvalsGetRunOverview.js.map +1 -0
  404. package/dist/esm/react-query/llmEvalsGetSampleTrajectory.core.d.ts +16 -0
  405. package/dist/esm/react-query/llmEvalsGetSampleTrajectory.core.d.ts.map +1 -0
  406. package/dist/esm/react-query/llmEvalsGetSampleTrajectory.core.js +40 -0
  407. package/dist/esm/react-query/llmEvalsGetSampleTrajectory.core.js.map +1 -0
  408. package/dist/esm/react-query/llmEvalsGetSampleTrajectory.d.ts +44 -0
  409. package/dist/esm/react-query/llmEvalsGetSampleTrajectory.d.ts.map +1 -0
  410. package/dist/esm/react-query/llmEvalsGetSampleTrajectory.js +55 -0
  411. package/dist/esm/react-query/llmEvalsGetSampleTrajectory.js.map +1 -0
  412. package/dist/esm/react-query/llmEvalsGetSuiteLeaderboard.core.d.ts +20 -0
  413. package/dist/esm/react-query/llmEvalsGetSuiteLeaderboard.core.d.ts.map +1 -0
  414. package/dist/esm/react-query/llmEvalsGetSuiteLeaderboard.core.js +36 -0
  415. package/dist/esm/react-query/llmEvalsGetSuiteLeaderboard.core.js.map +1 -0
  416. package/dist/esm/react-query/llmEvalsGetSuiteLeaderboard.d.ts +48 -0
  417. package/dist/esm/react-query/llmEvalsGetSuiteLeaderboard.d.ts.map +1 -0
  418. package/dist/esm/react-query/llmEvalsGetSuiteLeaderboard.js +55 -0
  419. package/dist/esm/react-query/llmEvalsGetSuiteLeaderboard.js.map +1 -0
  420. package/dist/esm/react-query/llmEvalsImportAgentRun.d.ts +28 -0
  421. package/dist/esm/react-query/llmEvalsImportAgentRun.d.ts.map +1 -0
  422. package/dist/esm/react-query/llmEvalsImportAgentRun.js +42 -0
  423. package/dist/esm/react-query/llmEvalsImportAgentRun.js.map +1 -0
  424. package/dist/esm/react-query/llmEvalsUploadSampleTrajectory.d.ts +29 -0
  425. package/dist/esm/react-query/llmEvalsUploadSampleTrajectory.d.ts.map +1 -0
  426. package/dist/esm/react-query/llmEvalsUploadSampleTrajectory.js +42 -0
  427. package/dist/esm/react-query/llmEvalsUploadSampleTrajectory.js.map +1 -0
  428. package/dist/esm/react-query/llmResponsesCreate.d.ts +34 -0
  429. package/dist/esm/react-query/llmResponsesCreate.d.ts.map +1 -0
  430. package/dist/esm/react-query/llmResponsesCreate.js +43 -0
  431. package/dist/esm/react-query/llmResponsesCreate.js.map +1 -0
  432. package/dist/esm/sdk/evals.d.ts +42 -0
  433. package/dist/esm/sdk/evals.d.ts.map +1 -1
  434. package/dist/esm/sdk/evals.js +60 -0
  435. package/dist/esm/sdk/evals.js.map +1 -1
  436. package/dist/esm/sdk/responses.d.ts +11 -0
  437. package/dist/esm/sdk/responses.d.ts.map +1 -1
  438. package/dist/esm/sdk/responses.js +11 -0
  439. package/dist/esm/sdk/responses.js.map +1 -1
  440. package/jsr.json +1 -1
  441. package/package.json +1 -1
  442. package/src/funcs/llmEvalsAppendAgentTrials.ts +188 -0
  443. package/src/funcs/llmEvalsGetRunOverview.ts +192 -0
  444. package/src/funcs/llmEvalsGetSampleTrajectory.ts +197 -0
  445. package/src/funcs/llmEvalsGetSuiteLeaderboard.ts +197 -0
  446. package/src/funcs/llmEvalsImportAgentRun.ts +181 -0
  447. package/src/funcs/llmEvalsUploadSampleTrajectory.ts +201 -0
  448. package/src/funcs/llmResponsesCreate.ts +188 -0
  449. package/src/lib/config.ts +2 -2
  450. package/src/mcp-server/mcp-server.ts +1 -1
  451. package/src/mcp-server/server.ts +15 -1
  452. package/src/mcp-server/tools/llmEvalsAppendAgentTrials.ts +37 -0
  453. package/src/mcp-server/tools/llmEvalsGetRunOverview.ts +37 -0
  454. package/src/mcp-server/tools/llmEvalsGetSampleTrajectory.ts +37 -0
  455. package/src/mcp-server/tools/llmEvalsGetSuiteLeaderboard.ts +37 -0
  456. package/src/mcp-server/tools/llmEvalsImportAgentRun.ts +37 -0
  457. package/src/mcp-server/tools/llmEvalsUploadSampleTrajectory.ts +38 -0
  458. package/src/mcp-server/tools/llmResponsesCreate.ts +37 -0
  459. package/src/models/components/appendagenttrialsrequest.ts +224 -0
  460. package/src/models/components/evalleaderboardrow.ts +200 -0
  461. package/src/models/components/evalrunoverview.ts +240 -0
  462. package/src/models/components/evalsampleobject.ts +48 -0
  463. package/src/models/components/evalsampletrajectoryuploadedobject.ts +98 -0
  464. package/src/models/components/evalsuiteleaderboard.ts +93 -0
  465. package/src/models/components/importagentrunrequest.ts +319 -0
  466. package/src/models/components/index.ts +8 -0
  467. package/src/models/components/responsescreaterequest.ts +1076 -0
  468. package/src/models/components/uploadevalsampletrajectoryrequest.ts +79 -0
  469. package/src/models/operations/appendagentevaltrials.ts +678 -0
  470. package/src/models/operations/cancelevalrun.ts +36 -0
  471. package/src/models/operations/createevalrun.ts +36 -0
  472. package/src/models/operations/createresponse.ts +198 -0
  473. package/src/models/operations/getevalrun.ts +36 -0
  474. package/src/models/operations/getevalrunoverview.ts +73 -0
  475. package/src/models/operations/getevalsampletrajectory.ts +199 -0
  476. package/src/models/operations/getevalsuiteleaderboard.ts +93 -0
  477. package/src/models/operations/importagentevalrun.ts +665 -0
  478. package/src/models/operations/index.ts +7 -0
  479. package/src/models/operations/listevalruns.ts +36 -0
  480. package/src/models/operations/listevalscheduleruns.ts +36 -0
  481. package/src/models/operations/listevalsuiteleaderboardruns.ts +36 -0
  482. package/src/models/operations/rerunfailedevalsamples.ts +36 -0
  483. package/src/models/operations/retryfailedevalrun.ts +36 -0
  484. package/src/models/operations/uploadevalsampletrajectory.ts +98 -0
  485. package/src/react-query/index.ts +7 -0
  486. package/src/react-query/llmEvalsAppendAgentTrials.ts +109 -0
  487. package/src/react-query/llmEvalsGetRunOverview.core.ts +77 -0
  488. package/src/react-query/llmEvalsGetRunOverview.ts +150 -0
  489. package/src/react-query/llmEvalsGetSampleTrajectory.core.ts +94 -0
  490. package/src/react-query/llmEvalsGetSampleTrajectory.ts +159 -0
  491. package/src/react-query/llmEvalsGetSuiteLeaderboard.core.ts +89 -0
  492. package/src/react-query/llmEvalsGetSuiteLeaderboard.ts +163 -0
  493. package/src/react-query/llmEvalsImportAgentRun.ts +109 -0
  494. package/src/react-query/llmEvalsUploadSampleTrajectory.ts +110 -0
  495. package/src/react-query/llmResponsesCreate.ts +113 -0
  496. package/src/sdk/evals.ts +108 -0
  497. package/src/sdk/responses.ts +23 -0
@@ -0,0 +1 @@
1
+ {"version":3,"file":"llmEvalsImportAgentRun.js","sourceRoot":"","sources":["../../../../src/mcp-server/tools/llmEvalsImportAgentRun.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,sBAAsB,EAAE,MAAM,uCAAuC,CAAC;AAC/E,OAAO,KAAK,UAAU,MAAM,kCAAkC,CAAC;AAC/D,OAAO,EAAE,YAAY,EAAkB,MAAM,aAAa,CAAC;AAE3D,MAAM,IAAI,GAAG;IACX,OAAO,EAAE,UAAU,CAAC,uCAAuC;CAC5D,CAAC;AAEF,MAAM,CAAC,MAAM,2BAA2B,GAAgC;IACtE,IAAI,EAAE,4BAA4B;IAClC,WAAW,EAAE;;o8BAEq7B;IACl8B,IAAI;IACJ,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE;QAChC,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,MAAM,sBAAsB,CACpD,MAAM,EACN,IAAI,CAAC,OAAO,EACZ,EAAE,YAAY,EAAE,EAAE,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,EAAE,CACzC,CAAC,QAAQ,EAAE,CAAC;QAEb,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;YACf,OAAO;gBACL,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;gBACvD,OAAO,EAAE,IAAI;aACd,CAAC;QACJ,CAAC;QAED,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC;QAE3B,OAAO,YAAY,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC;IACtC,CAAC;CACF,CAAC"}
@@ -0,0 +1,8 @@
1
+ import * as operations from "../../models/operations/index.js";
2
+ import { ToolDefinition } from "../tools.js";
3
+ declare const args: {
4
+ request: import("zod").ZodType<operations.UploadEvalSampleTrajectoryRequest, import("zod").ZodTypeDef, unknown>;
5
+ };
6
+ export declare const tool$llmEvalsUploadSampleTrajectory: ToolDefinition<typeof args>;
7
+ export {};
8
+ //# sourceMappingURL=llmEvalsUploadSampleTrajectory.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"llmEvalsUploadSampleTrajectory.d.ts","sourceRoot":"","sources":["../../../../src/mcp-server/tools/llmEvalsUploadSampleTrajectory.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,UAAU,MAAM,kCAAkC,CAAC;AAC/D,OAAO,EAAgB,cAAc,EAAE,MAAM,aAAa,CAAC;AAE3D,QAAA,MAAM,IAAI;;CAET,CAAC;AAEF,eAAO,MAAM,mCAAmC,EAAE,cAAc,CAAC,OAAO,IAAI,CAyBzE,CAAC"}
@@ -0,0 +1,28 @@
1
+ /*
2
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
3
+ */
4
+ import { llmEvalsUploadSampleTrajectory } from "../../funcs/llmEvalsUploadSampleTrajectory.js";
5
+ import * as operations from "../../models/operations/index.js";
6
+ import { formatResult } from "../tools.js";
7
+ const args = {
8
+ request: operations.UploadEvalSampleTrajectoryRequest$inboundSchema,
9
+ };
10
+ export const tool$llmEvalsUploadSampleTrajectory = {
11
+ name: "llm-evals-upload-sample-trajectory",
12
+ description: `Upload an eval sample trajectory
13
+
14
+ Stores one coding-agent trial's session, in whatever interchange format the harness produced (record it in \`schema_version\`, e.g. \`ATIF-v1.7\`). \`steps\` is stored verbatim and never validated, so a harness format change needs no gateway release. Separate from the run import because a trajectory averages ~92 KB and a whole run's worth would not fit in one request. Upserts, so re-sending after a timeout replaces the trajectory rather than failing. Sample indices are unique per (task, model), not per run, so pass \`task_id\` and/or \`model\` when the run covers more than one — an ambiguous index returns 400 rather than writing to the wrong trial. Rejects a payload whose \`steps\` and \`logs\` together exceed 16 MiB with 413.`,
15
+ args,
16
+ tool: async (client, args, ctx) => {
17
+ const [result, apiCall] = await llmEvalsUploadSampleTrajectory(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
18
+ if (!result.ok) {
19
+ return {
20
+ content: [{ type: "text", text: result.error.message }],
21
+ isError: true,
22
+ };
23
+ }
24
+ const value = result.value;
25
+ return formatResult(value, apiCall);
26
+ },
27
+ };
28
+ //# sourceMappingURL=llmEvalsUploadSampleTrajectory.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"llmEvalsUploadSampleTrajectory.js","sourceRoot":"","sources":["../../../../src/mcp-server/tools/llmEvalsUploadSampleTrajectory.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,8BAA8B,EAAE,MAAM,+CAA+C,CAAC;AAC/F,OAAO,KAAK,UAAU,MAAM,kCAAkC,CAAC;AAC/D,OAAO,EAAE,YAAY,EAAkB,MAAM,aAAa,CAAC;AAE3D,MAAM,IAAI,GAAG;IACX,OAAO,EAAE,UAAU,CAAC,+CAA+C;CACpE,CAAC;AAEF,MAAM,CAAC,MAAM,mCAAmC,GAC9C;IACE,IAAI,EAAE,oCAAoC;IAC1C,WAAW,EAAE;;iuBAEgtB;IAC7tB,IAAI;IACJ,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE;QAChC,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,MAAM,8BAA8B,CAC5D,MAAM,EACN,IAAI,CAAC,OAAO,EACZ,EAAE,YAAY,EAAE,EAAE,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,EAAE,CACzC,CAAC,QAAQ,EAAE,CAAC;QAEb,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;YACf,OAAO;gBACL,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;gBACvD,OAAO,EAAE,IAAI;aACd,CAAC;QACJ,CAAC;QAED,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC;QAE3B,OAAO,YAAY,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC;IACtC,CAAC;CACF,CAAC"}
@@ -0,0 +1,8 @@
1
+ import * as operations from "../../models/operations/index.js";
2
+ import { ToolDefinition } from "../tools.js";
3
+ declare const args: {
4
+ request: import("zod").ZodType<operations.CreateResponseRequest, import("zod").ZodTypeDef, unknown>;
5
+ };
6
+ export declare const tool$llmResponsesCreate: ToolDefinition<typeof args>;
7
+ export {};
8
+ //# sourceMappingURL=llmResponsesCreate.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"llmResponsesCreate.d.ts","sourceRoot":"","sources":["../../../../src/mcp-server/tools/llmResponsesCreate.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,UAAU,MAAM,kCAAkC,CAAC;AAC/D,OAAO,EAAgB,cAAc,EAAE,MAAM,aAAa,CAAC;AAE3D,QAAA,MAAM,IAAI;;CAET,CAAC;AAEF,eAAO,MAAM,uBAAuB,EAAE,cAAc,CAAC,OAAO,IAAI,CAwB/D,CAAC"}
@@ -0,0 +1,28 @@
1
+ /*
2
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
3
+ */
4
+ import { llmResponsesCreate } from "../../funcs/llmResponsesCreate.js";
5
+ import * as operations from "../../models/operations/index.js";
6
+ import { formatResult } from "../tools.js";
7
+ const args = {
8
+ request: operations.CreateResponseRequest$inboundSchema,
9
+ };
10
+ export const tool$llmResponsesCreate = {
11
+ name: "llm-responses-create",
12
+ description: `Create an agent-powered response with tool support
13
+
14
+ Creates a new AI agent response using advanced language models with autonomous tool usage capabilities.`,
15
+ args,
16
+ tool: async (client, args, ctx) => {
17
+ const [result, apiCall] = await llmResponsesCreate(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
18
+ if (!result.ok) {
19
+ return {
20
+ content: [{ type: "text", text: result.error.message }],
21
+ isError: true,
22
+ };
23
+ }
24
+ const value = result.value;
25
+ return formatResult(value, apiCall);
26
+ },
27
+ };
28
+ //# sourceMappingURL=llmResponsesCreate.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"llmResponsesCreate.js","sourceRoot":"","sources":["../../../../src/mcp-server/tools/llmResponsesCreate.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,kBAAkB,EAAE,MAAM,mCAAmC,CAAC;AACvE,OAAO,KAAK,UAAU,MAAM,kCAAkC,CAAC;AAC/D,OAAO,EAAE,YAAY,EAAkB,MAAM,aAAa,CAAC;AAE3D,MAAM,IAAI,GAAG;IACX,OAAO,EAAE,UAAU,CAAC,mCAAmC;CACxD,CAAC;AAEF,MAAM,CAAC,MAAM,uBAAuB,GAAgC;IAClE,IAAI,EAAE,sBAAsB;IAC5B,WAAW,EAAE;;wGAEyF;IACtG,IAAI;IACJ,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE;QAChC,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,MAAM,kBAAkB,CAChD,MAAM,EACN,IAAI,CAAC,OAAO,EACZ,EAAE,YAAY,EAAE,EAAE,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,EAAE,CACzC,CAAC,QAAQ,EAAE,CAAC;QAEb,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;YACf,OAAO;gBACL,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;gBACvD,OAAO,EAAE,IAAI;aACd,CAAC;QACJ,CAAC;QAED,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC;QAE3B,OAAO,YAAY,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC;IACtC,CAAC;CACF,CAAC"}
@@ -0,0 +1,111 @@
1
+ import * as z from "zod/v3";
2
+ import { ClosedEnum } from "../../types/enums.js";
3
+ import { Result as SafeParseResult } from "../../types/fp.js";
4
+ import { SDKValidationError } from "../errors/sdkvalidationerror.js";
5
+ /**
6
+ * `failed` means no score was produced at all. An agent timeout is NOT that — the harness still runs the verifier and scores the trial, so it stays `completed` and carries its exception in `error`.
7
+ */
8
+ export declare const AppendAgentTrialsRequestStatus: {
9
+ readonly Completed: "completed";
10
+ readonly Failed: "failed";
11
+ };
12
+ /**
13
+ * `failed` means no score was produced at all. An agent timeout is NOT that — the harness still runs the verifier and scores the trial, so it stays `completed` and carries its exception in `error`.
14
+ */
15
+ export type AppendAgentTrialsRequestStatus = ClosedEnum<typeof AppendAgentTrialsRequestStatus>;
16
+ export type AppendAgentTrialsRequestTrials = {
17
+ /**
18
+ * Must already exist in the target suite version's manifest; unknown ids are rejected rather than invented.
19
+ */
20
+ taskId: string;
21
+ sampleIndex: number;
22
+ /**
23
+ * `failed` means no score was produced at all. An agent timeout is NOT that — the harness still runs the verifier and scores the trial, so it stays `completed` and carries its exception in `error`.
24
+ */
25
+ status?: AppendAgentTrialsRequestStatus | undefined;
26
+ /**
27
+ * The task instruction.
28
+ */
29
+ prompt?: string | undefined;
30
+ target?: string | null | undefined;
31
+ /**
32
+ * What decided the score — the verifier's output. The agent's session goes to the trajectory route.
33
+ */
34
+ outputText?: string | null | undefined;
35
+ /**
36
+ * Metric values for this trial, e.g. `{ "reward": 1, "resolved": 1 }`. The board averages `reward` by default.
37
+ */
38
+ scores?: {
39
+ [k: string]: number;
40
+ } | undefined;
41
+ /**
42
+ * The exception this trial recorded, if any. Does not by itself make the trial a failure.
43
+ */
44
+ error?: {
45
+ [k: string]: any;
46
+ } | null | undefined;
47
+ costUsd?: number | null | undefined;
48
+ inputTokens?: number | null | undefined;
49
+ outputTokens?: number | null | undefined;
50
+ /**
51
+ * The share of `input_tokens` served from the provider's prompt cache.
52
+ */
53
+ cachedTokens?: number | null | undefined;
54
+ phaseTimings?: {
55
+ [k: string]: any;
56
+ } | null | undefined;
57
+ };
58
+ export type AppendAgentTrialsRequest = {
59
+ /**
60
+ * Each trial pins its own `sample_index` (unique per task and model): re-sending a batch replaces the same rows instead of duplicating them, so a retried batch is safe.
61
+ */
62
+ trials?: Array<AppendAgentTrialsRequestTrials> | undefined;
63
+ /**
64
+ * Marks the run completed once these trials land — the last call of a streaming job sends this, usually with an empty `trials`.
65
+ */
66
+ finalize?: boolean | undefined;
67
+ };
68
+ /** @internal */
69
+ export declare const AppendAgentTrialsRequestStatus$inboundSchema: z.ZodNativeEnum<typeof AppendAgentTrialsRequestStatus>;
70
+ /** @internal */
71
+ export declare const AppendAgentTrialsRequestStatus$outboundSchema: z.ZodNativeEnum<typeof AppendAgentTrialsRequestStatus>;
72
+ /** @internal */
73
+ export declare const AppendAgentTrialsRequestTrials$inboundSchema: z.ZodType<AppendAgentTrialsRequestTrials, z.ZodTypeDef, unknown>;
74
+ /** @internal */
75
+ export type AppendAgentTrialsRequestTrials$Outbound = {
76
+ task_id: string;
77
+ sample_index: number;
78
+ status: string;
79
+ prompt: string;
80
+ target?: string | null | undefined;
81
+ output_text?: string | null | undefined;
82
+ scores?: {
83
+ [k: string]: number;
84
+ } | undefined;
85
+ error?: {
86
+ [k: string]: any;
87
+ } | null | undefined;
88
+ cost_usd?: number | null | undefined;
89
+ input_tokens?: number | null | undefined;
90
+ output_tokens?: number | null | undefined;
91
+ cached_tokens?: number | null | undefined;
92
+ phase_timings?: {
93
+ [k: string]: any;
94
+ } | null | undefined;
95
+ };
96
+ /** @internal */
97
+ export declare const AppendAgentTrialsRequestTrials$outboundSchema: z.ZodType<AppendAgentTrialsRequestTrials$Outbound, z.ZodTypeDef, AppendAgentTrialsRequestTrials>;
98
+ export declare function appendAgentTrialsRequestTrialsToJSON(appendAgentTrialsRequestTrials: AppendAgentTrialsRequestTrials): string;
99
+ export declare function appendAgentTrialsRequestTrialsFromJSON(jsonString: string): SafeParseResult<AppendAgentTrialsRequestTrials, SDKValidationError>;
100
+ /** @internal */
101
+ export declare const AppendAgentTrialsRequest$inboundSchema: z.ZodType<AppendAgentTrialsRequest, z.ZodTypeDef, unknown>;
102
+ /** @internal */
103
+ export type AppendAgentTrialsRequest$Outbound = {
104
+ trials?: Array<AppendAgentTrialsRequestTrials$Outbound> | undefined;
105
+ finalize: boolean;
106
+ };
107
+ /** @internal */
108
+ export declare const AppendAgentTrialsRequest$outboundSchema: z.ZodType<AppendAgentTrialsRequest$Outbound, z.ZodTypeDef, AppendAgentTrialsRequest>;
109
+ export declare function appendAgentTrialsRequestToJSON(appendAgentTrialsRequest: AppendAgentTrialsRequest): string;
110
+ export declare function appendAgentTrialsRequestFromJSON(jsonString: string): SafeParseResult<AppendAgentTrialsRequest, SDKValidationError>;
111
+ //# sourceMappingURL=appendagenttrialsrequest.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"appendagenttrialsrequest.d.ts","sourceRoot":"","sources":["../../../../src/models/components/appendagenttrialsrequest.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,QAAQ,CAAC;AAG5B,OAAO,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAC;AAClD,OAAO,EAAE,MAAM,IAAI,eAAe,EAAE,MAAM,mBAAmB,CAAC;AAC9D,OAAO,EAAE,kBAAkB,EAAE,MAAM,iCAAiC,CAAC;AAErE;;GAEG;AACH,eAAO,MAAM,8BAA8B;;;CAGjC,CAAC;AACX;;GAEG;AACH,MAAM,MAAM,8BAA8B,GAAG,UAAU,CACrD,OAAO,8BAA8B,CACtC,CAAC;AAEF,MAAM,MAAM,8BAA8B,GAAG;IAC3C;;OAEG;IACH,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,EAAE,MAAM,CAAC;IACpB;;OAEG;IACH,MAAM,CAAC,EAAE,8BAA8B,GAAG,SAAS,CAAC;IACpD;;OAEG;IACH,MAAM,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC5B,MAAM,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACnC;;OAEG;IACH,UAAU,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACvC;;OAEG;IACH,MAAM,CAAC,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,MAAM,CAAA;KAAE,GAAG,SAAS,CAAC;IAC7C;;OAEG;IACH,KAAK,CAAC,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,GAAG,IAAI,GAAG,SAAS,CAAC;IAChD,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACpC,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACxC,YAAY,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACzC;;OAEG;IACH,YAAY,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACzC,YAAY,CAAC,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,GAAG,IAAI,GAAG,SAAS,CAAC;CACxD,CAAC;AAEF,MAAM,MAAM,wBAAwB,GAAG;IACrC;;OAEG;IACH,MAAM,CAAC,EAAE,KAAK,CAAC,8BAA8B,CAAC,GAAG,SAAS,CAAC;IAC3D;;OAEG;IACH,QAAQ,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;CAChC,CAAC;AAEF,gBAAgB;AAChB,eAAO,MAAM,4CAA4C,EAAE,CAAC,CAAC,aAAa,CACxE,OAAO,8BAA8B,CACS,CAAC;AACjD,gBAAgB;AAChB,eAAO,MAAM,6CAA6C,EAAE,CAAC,CAAC,aAAa,CACzE,OAAO,8BAA8B,CACS,CAAC;AAEjD,gBAAgB;AAChB,eAAO,MAAM,4CAA4C,EAAE,CAAC,CAAC,OAAO,CAClE,8BAA8B,EAC9B,CAAC,CAAC,UAAU,EACZ,OAAO,CA0BP,CAAC;AACH,gBAAgB;AAChB,MAAM,MAAM,uCAAuC,GAAG;IACpD,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,MAAM,CAAC;IACrB,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACnC,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACxC,MAAM,CAAC,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,MAAM,CAAA;KAAE,GAAG,SAAS,CAAC;IAC7C,KAAK,CAAC,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,GAAG,IAAI,GAAG,SAAS,CAAC;IAChD,QAAQ,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACrC,YAAY,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACzC,aAAa,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IAC1C,aAAa,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IAC1C,aAAa,CAAC,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,GAAG,IAAI,GAAG,SAAS,CAAC;CACzD,CAAC;AAEF,gBAAgB;AAChB,eAAO,MAAM,6CAA6C,EAAE,CAAC,CAAC,OAAO,CACnE,uCAAuC,EACvC,CAAC,CAAC,UAAU,EACZ,8BAA8B,CA0B9B,CAAC;AAEH,wBAAgB,oCAAoC,CAClD,8BAA8B,EAAE,8BAA8B,GAC7D,MAAM,CAMR;AACD,wBAAgB,sCAAsC,CACpD,UAAU,EAAE,MAAM,GACjB,eAAe,CAAC,8BAA8B,EAAE,kBAAkB,CAAC,CAMrE;AAED,gBAAgB;AAChB,eAAO,MAAM,sCAAsC,EAAE,CAAC,CAAC,OAAO,CAC5D,wBAAwB,EACxB,CAAC,CAAC,UAAU,EACZ,OAAO,CAKP,CAAC;AACH,gBAAgB;AAChB,MAAM,MAAM,iCAAiC,GAAG;IAC9C,MAAM,CAAC,EAAE,KAAK,CAAC,uCAAuC,CAAC,GAAG,SAAS,CAAC;IACpE,QAAQ,EAAE,OAAO,CAAC;CACnB,CAAC;AAEF,gBAAgB;AAChB,eAAO,MAAM,uCAAuC,EAAE,CAAC,CAAC,OAAO,CAC7D,iCAAiC,EACjC,CAAC,CAAC,UAAU,EACZ,wBAAwB,CAKxB,CAAC;AAEH,wBAAgB,8BAA8B,CAC5C,wBAAwB,EAAE,wBAAwB,GACjD,MAAM,CAIR;AACD,wBAAgB,gCAAgC,CAC9C,UAAU,EAAE,MAAM,GACjB,eAAe,CAAC,wBAAwB,EAAE,kBAAkB,CAAC,CAM/D"}
@@ -0,0 +1,96 @@
1
+ /*
2
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
3
+ */
4
+ import * as z from "zod/v3";
5
+ import { remap as remap$ } from "../../lib/primitives.js";
6
+ import { safeParse } from "../../lib/schemas.js";
7
+ /**
8
+ * `failed` means no score was produced at all. An agent timeout is NOT that — the harness still runs the verifier and scores the trial, so it stays `completed` and carries its exception in `error`.
9
+ */
10
+ export const AppendAgentTrialsRequestStatus = {
11
+ Completed: "completed",
12
+ Failed: "failed",
13
+ };
14
+ /** @internal */
15
+ export const AppendAgentTrialsRequestStatus$inboundSchema = z.nativeEnum(AppendAgentTrialsRequestStatus);
16
+ /** @internal */
17
+ export const AppendAgentTrialsRequestStatus$outboundSchema = AppendAgentTrialsRequestStatus$inboundSchema;
18
+ /** @internal */
19
+ export const AppendAgentTrialsRequestTrials$inboundSchema = z.object({
20
+ task_id: z.string(),
21
+ sample_index: z.number().int(),
22
+ status: AppendAgentTrialsRequestStatus$inboundSchema.default("completed"),
23
+ prompt: z.string().default(""),
24
+ target: z.nullable(z.string()).optional(),
25
+ output_text: z.nullable(z.string()).optional(),
26
+ scores: z.record(z.number()).optional(),
27
+ error: z.nullable(z.record(z.any())).optional(),
28
+ cost_usd: z.nullable(z.number()).optional(),
29
+ input_tokens: z.nullable(z.number().int()).optional(),
30
+ output_tokens: z.nullable(z.number().int()).optional(),
31
+ cached_tokens: z.nullable(z.number().int()).optional(),
32
+ phase_timings: z.nullable(z.record(z.any())).optional(),
33
+ }).transform((v) => {
34
+ return remap$(v, {
35
+ "task_id": "taskId",
36
+ "sample_index": "sampleIndex",
37
+ "output_text": "outputText",
38
+ "cost_usd": "costUsd",
39
+ "input_tokens": "inputTokens",
40
+ "output_tokens": "outputTokens",
41
+ "cached_tokens": "cachedTokens",
42
+ "phase_timings": "phaseTimings",
43
+ });
44
+ });
45
+ /** @internal */
46
+ export const AppendAgentTrialsRequestTrials$outboundSchema = z.object({
47
+ taskId: z.string(),
48
+ sampleIndex: z.number().int(),
49
+ status: AppendAgentTrialsRequestStatus$outboundSchema.default("completed"),
50
+ prompt: z.string().default(""),
51
+ target: z.nullable(z.string()).optional(),
52
+ outputText: z.nullable(z.string()).optional(),
53
+ scores: z.record(z.number()).optional(),
54
+ error: z.nullable(z.record(z.any())).optional(),
55
+ costUsd: z.nullable(z.number()).optional(),
56
+ inputTokens: z.nullable(z.number().int()).optional(),
57
+ outputTokens: z.nullable(z.number().int()).optional(),
58
+ cachedTokens: z.nullable(z.number().int()).optional(),
59
+ phaseTimings: z.nullable(z.record(z.any())).optional(),
60
+ }).transform((v) => {
61
+ return remap$(v, {
62
+ taskId: "task_id",
63
+ sampleIndex: "sample_index",
64
+ outputText: "output_text",
65
+ costUsd: "cost_usd",
66
+ inputTokens: "input_tokens",
67
+ outputTokens: "output_tokens",
68
+ cachedTokens: "cached_tokens",
69
+ phaseTimings: "phase_timings",
70
+ });
71
+ });
72
+ export function appendAgentTrialsRequestTrialsToJSON(appendAgentTrialsRequestTrials) {
73
+ return JSON.stringify(AppendAgentTrialsRequestTrials$outboundSchema.parse(appendAgentTrialsRequestTrials));
74
+ }
75
+ export function appendAgentTrialsRequestTrialsFromJSON(jsonString) {
76
+ return safeParse(jsonString, (x) => AppendAgentTrialsRequestTrials$inboundSchema.parse(JSON.parse(x)), `Failed to parse 'AppendAgentTrialsRequestTrials' from JSON`);
77
+ }
78
+ /** @internal */
79
+ export const AppendAgentTrialsRequest$inboundSchema = z.object({
80
+ trials: z.array(z.lazy(() => AppendAgentTrialsRequestTrials$inboundSchema))
81
+ .optional(),
82
+ finalize: z.boolean().default(false),
83
+ });
84
+ /** @internal */
85
+ export const AppendAgentTrialsRequest$outboundSchema = z.object({
86
+ trials: z.array(z.lazy(() => AppendAgentTrialsRequestTrials$outboundSchema))
87
+ .optional(),
88
+ finalize: z.boolean().default(false),
89
+ });
90
+ export function appendAgentTrialsRequestToJSON(appendAgentTrialsRequest) {
91
+ return JSON.stringify(AppendAgentTrialsRequest$outboundSchema.parse(appendAgentTrialsRequest));
92
+ }
93
+ export function appendAgentTrialsRequestFromJSON(jsonString) {
94
+ return safeParse(jsonString, (x) => AppendAgentTrialsRequest$inboundSchema.parse(JSON.parse(x)), `Failed to parse 'AppendAgentTrialsRequest' from JSON`);
95
+ }
96
+ //# sourceMappingURL=appendagenttrialsrequest.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"appendagenttrialsrequest.js","sourceRoot":"","sources":["../../../../src/models/components/appendagenttrialsrequest.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,QAAQ,CAAC;AAC5B,OAAO,EAAE,KAAK,IAAI,MAAM,EAAE,MAAM,yBAAyB,CAAC;AAC1D,OAAO,EAAE,SAAS,EAAE,MAAM,sBAAsB,CAAC;AAKjD;;GAEG;AACH,MAAM,CAAC,MAAM,8BAA8B,GAAG;IAC5C,SAAS,EAAE,WAAW;IACtB,MAAM,EAAE,QAAQ;CACR,CAAC;AAwDX,gBAAgB;AAChB,MAAM,CAAC,MAAM,4CAA4C,GAErD,CAAC,CAAC,UAAU,CAAC,8BAA8B,CAAC,CAAC;AACjD,gBAAgB;AAChB,MAAM,CAAC,MAAM,6CAA6C,GAEtD,4CAA4C,CAAC;AAEjD,gBAAgB;AAChB,MAAM,CAAC,MAAM,4CAA4C,GAIrD,CAAC,CAAC,MAAM,CAAC;IACX,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE;IACnB,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE;IAC9B,MAAM,EAAE,4CAA4C,CAAC,OAAO,CAAC,WAAW,CAAC;IACzE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,EAAE,CAAC;IAC9B,MAAM,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE;IACzC,WAAW,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE;IAC9C,MAAM,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE;IACvC,KAAK,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,EAAE;IAC/C,QAAQ,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE;IAC3C,YAAY,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE;IACrD,aAAa,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE;IACtD,aAAa,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE;IACtD,aAAa,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,EAAE;CACxD,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,EAAE;IACjB,OAAO,MAAM,CAAC,CAAC,EAAE;QACf,SAAS,EAAE,QAAQ;QACnB,cAAc,EAAE,aAAa;QAC7B,aAAa,EAAE,YAAY;QAC3B,UAAU,EAAE,SAAS;QACrB,cAAc,EAAE,aAAa;QAC7B,eAAe,EAAE,cAAc;QAC/B,eAAe,EAAE,cAAc;QAC/B,eAAe,EAAE,cAAc;KAChC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAkBH,gBAAgB;AAChB,MAAM,CAAC,MAAM,6CAA6C,GAItD,CAAC,CAAC,MAAM,CAAC;IACX,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE;IAClB,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE;IAC7B,MAAM,EAAE,6CAA6C,CAAC,OAAO,CAAC,WAAW,CAAC;IAC1E,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,EAAE,CAAC;IAC9B,MAAM,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE;IACzC,UAAU,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE;IAC7C,MAAM,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE;IACvC,KAAK,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,EAAE;IAC/C,OAAO,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE;IAC1C,WAAW,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE;IACpD,YAAY,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE;IACrD,YAAY,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE;IACrD,YAAY,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,EAAE;CACvD,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,EAAE;IACjB,OAAO,MAAM,CAAC,CAAC,EAAE;QACf,MAAM,EAAE,SAAS;QACjB,WAAW,EAAE,cAAc;QAC3B,UAAU,EAAE,aAAa;QACzB,OAAO,EAAE,UAAU;QACnB,WAAW,EAAE,cAAc;QAC3B,YAAY,EAAE,eAAe;QAC7B,YAAY,EAAE,eAAe;QAC7B,YAAY,EAAE,eAAe;KAC9B,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,MAAM,UAAU,oCAAoC,CAClD,8BAA8D;IAE9D,OAAO,IAAI,CAAC,SAAS,CACnB,6CAA6C,CAAC,KAAK,CACjD,8BAA8B,CAC/B,CACF,CAAC;AACJ,CAAC;AACD,MAAM,UAAU,sCAAsC,CACpD,UAAkB;IAElB,OAAO,SAAS,CACd,UAAU,EACV,CAAC,CAAC,EAAE,EAAE,CAAC,4CAA4C,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,EACxE,4DAA4D,CAC7D,CAAC;AACJ,CAAC;AAED,gBAAgB;AAChB,MAAM,CAAC,MAAM,sCAAsC,GAI/C,CAAC,CAAC,MAAM,CAAC;IACX,MAAM,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC,4CAA4C,CAAC,CAAC;SACxE,QAAQ,EAAE;IACb,QAAQ,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC;CACrC,CAAC,CAAC;AAOH,gBAAgB;AAChB,MAAM,CAAC,MAAM,uCAAuC,GAIhD,CAAC,CAAC,MAAM,CAAC;IACX,MAAM,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC,6CAA6C,CAAC,CAAC;SACzE,QAAQ,EAAE;IACb,QAAQ,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC;CACrC,CAAC,CAAC;AAEH,MAAM,UAAU,8BAA8B,CAC5C,wBAAkD;IAElD,OAAO,IAAI,CAAC,SAAS,CACnB,uCAAuC,CAAC,KAAK,CAAC,wBAAwB,CAAC,CACxE,CAAC;AACJ,CAAC;AACD,MAAM,UAAU,gCAAgC,CAC9C,UAAkB;IAElB,OAAO,SAAS,CACd,UAAU,EACV,CAAC,CAAC,EAAE,EAAE,CAAC,sCAAsC,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,EAClE,sDAAsD,CACvD,CAAC;AACJ,CAAC"}
@@ -0,0 +1,85 @@
1
+ import * as z from "zod/v3";
2
+ import { Result as SafeParseResult } from "../../types/fp.js";
3
+ import { SDKValidationError } from "../errors/sdkvalidationerror.js";
4
+ export type AccuracyCi = {
5
+ low: number;
6
+ high: number;
7
+ margin: number;
8
+ };
9
+ export type EvalLeaderboardRow = {
10
+ agentName: string;
11
+ /**
12
+ * Version from the row's newest run. Not part of the row identity, so older runs in this row may have reported a different one.
13
+ */
14
+ agentVersion: string | null;
15
+ agentEffort: string | null;
16
+ model: string;
17
+ /**
18
+ * Mean of `metric` over the row's scored trials. Null when no trial carried a numeric value for that metric — distinct from 0, which means every trial scored zero.
19
+ */
20
+ accuracy: number | null;
21
+ /**
22
+ * 95% Wald interval, p ± 1.96·sqrt(p(1-p)/n), clamped to [0, 1]. Derived per request rather than stored, so it cannot drift when a failed trial is rerun. It assumes `metric` is a 0/1 outcome (the harness `reward`); on a continuous metric the mean is still right but this interval is not, so read it only for pass/fail metrics. Null when `accuracy` is.
23
+ */
24
+ accuracyCi: AccuracyCi | null;
25
+ /**
26
+ * Every sample in the row's runs, whatever its status.
27
+ */
28
+ trials: number;
29
+ /**
30
+ * Trials carrying a numeric `metric` — the denominator behind `accuracy` and `accuracy_ci`.
31
+ */
32
+ scoredTrials: number;
33
+ /**
34
+ * Trials that ended in the `failed` status, i.e. produced no score at all. An agent timeout is NOT one of these: the harness still verifies and scores that trial.
35
+ */
36
+ failedTrials: number;
37
+ /**
38
+ * Summed trial spend, falling back to the runs' own totals when trials carry no cost. Null when neither reported any.
39
+ */
40
+ costUsd: number | null;
41
+ runCount: number;
42
+ /**
43
+ * The runs behind this row, largest (most trials) first, ties newest first — element 0 is the run to open when the reader drills into this row, so a full benchmark job outranks a newer one-task smoke run.
44
+ */
45
+ runIds: Array<string>;
46
+ /**
47
+ * Creation time of the newest run in the row, unix seconds.
48
+ */
49
+ lastRunAt: number;
50
+ };
51
+ /** @internal */
52
+ export declare const AccuracyCi$inboundSchema: z.ZodType<AccuracyCi, z.ZodTypeDef, unknown>;
53
+ /** @internal */
54
+ export type AccuracyCi$Outbound = {
55
+ low: number;
56
+ high: number;
57
+ margin: number;
58
+ };
59
+ /** @internal */
60
+ export declare const AccuracyCi$outboundSchema: z.ZodType<AccuracyCi$Outbound, z.ZodTypeDef, AccuracyCi>;
61
+ export declare function accuracyCiToJSON(accuracyCi: AccuracyCi): string;
62
+ export declare function accuracyCiFromJSON(jsonString: string): SafeParseResult<AccuracyCi, SDKValidationError>;
63
+ /** @internal */
64
+ export declare const EvalLeaderboardRow$inboundSchema: z.ZodType<EvalLeaderboardRow, z.ZodTypeDef, unknown>;
65
+ /** @internal */
66
+ export type EvalLeaderboardRow$Outbound = {
67
+ agent_name: string;
68
+ agent_version: string | null;
69
+ agent_effort: string | null;
70
+ model: string;
71
+ accuracy: number | null;
72
+ accuracy_ci: AccuracyCi$Outbound | null;
73
+ trials: number;
74
+ scored_trials: number;
75
+ failed_trials: number;
76
+ cost_usd: number | null;
77
+ run_count: number;
78
+ run_ids: Array<string>;
79
+ last_run_at: number;
80
+ };
81
+ /** @internal */
82
+ export declare const EvalLeaderboardRow$outboundSchema: z.ZodType<EvalLeaderboardRow$Outbound, z.ZodTypeDef, EvalLeaderboardRow>;
83
+ export declare function evalLeaderboardRowToJSON(evalLeaderboardRow: EvalLeaderboardRow): string;
84
+ export declare function evalLeaderboardRowFromJSON(jsonString: string): SafeParseResult<EvalLeaderboardRow, SDKValidationError>;
85
+ //# sourceMappingURL=evalleaderboardrow.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"evalleaderboardrow.d.ts","sourceRoot":"","sources":["../../../../src/models/components/evalleaderboardrow.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,QAAQ,CAAC;AAG5B,OAAO,EAAE,MAAM,IAAI,eAAe,EAAE,MAAM,mBAAmB,CAAC;AAC9D,OAAO,EAAE,kBAAkB,EAAE,MAAM,iCAAiC,CAAC;AAErE,MAAM,MAAM,UAAU,GAAG;IACvB,GAAG,EAAE,MAAM,CAAC;IACZ,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,MAAM,CAAC;CAChB,CAAC;AAEF,MAAM,MAAM,kBAAkB,GAAG;IAC/B,SAAS,EAAE,MAAM,CAAC;IAClB;;OAEG;IACH,YAAY,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B,KAAK,EAAE,MAAM,CAAC;IACd;;OAEG;IACH,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB;;OAEG;IACH,UAAU,EAAE,UAAU,GAAG,IAAI,CAAC;IAC9B;;OAEG;IACH,MAAM,EAAE,MAAM,CAAC;IACf;;OAEG;IACH,YAAY,EAAE,MAAM,CAAC;IACrB;;OAEG;IACH,YAAY,EAAE,MAAM,CAAC;IACrB;;OAEG;IACH,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IACvB,QAAQ,EAAE,MAAM,CAAC;IACjB;;OAEG;IACH,MAAM,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;IACtB;;OAEG;IACH,SAAS,EAAE,MAAM,CAAC;CACnB,CAAC;AAEF,gBAAgB;AAChB,eAAO,MAAM,wBAAwB,EAAE,CAAC,CAAC,OAAO,CAC9C,UAAU,EACV,CAAC,CAAC,UAAU,EACZ,OAAO,CAKP,CAAC;AACH,gBAAgB;AAChB,MAAM,MAAM,mBAAmB,GAAG;IAChC,GAAG,EAAE,MAAM,CAAC;IACZ,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,MAAM,CAAC;CAChB,CAAC;AAEF,gBAAgB;AAChB,eAAO,MAAM,yBAAyB,EAAE,CAAC,CAAC,OAAO,CAC/C,mBAAmB,EACnB,CAAC,CAAC,UAAU,EACZ,UAAU,CAKV,CAAC;AAEH,wBAAgB,gBAAgB,CAAC,UAAU,EAAE,UAAU,GAAG,MAAM,CAE/D;AACD,wBAAgB,kBAAkB,CAChC,UAAU,EAAE,MAAM,GACjB,eAAe,CAAC,UAAU,EAAE,kBAAkB,CAAC,CAMjD;AAED,gBAAgB;AAChB,eAAO,MAAM,gCAAgC,EAAE,CAAC,CAAC,OAAO,CACtD,kBAAkB,EAClB,CAAC,CAAC,UAAU,EACZ,OAAO,CA4BP,CAAC;AACH,gBAAgB;AAChB,MAAM,MAAM,2BAA2B,GAAG;IACxC,UAAU,EAAE,MAAM,CAAC;IACnB,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,YAAY,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB,WAAW,EAAE,mBAAmB,GAAG,IAAI,CAAC;IACxC,MAAM,EAAE,MAAM,CAAC;IACf,aAAa,EAAE,MAAM,CAAC;IACtB,aAAa,EAAE,MAAM,CAAC;IACtB,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;IACvB,WAAW,EAAE,MAAM,CAAC;CACrB,CAAC;AAEF,gBAAgB;AAChB,eAAO,MAAM,iCAAiC,EAAE,CAAC,CAAC,OAAO,CACvD,2BAA2B,EAC3B,CAAC,CAAC,UAAU,EACZ,kBAAkB,CA4BlB,CAAC;AAEH,wBAAgB,wBAAwB,CACtC,kBAAkB,EAAE,kBAAkB,GACrC,MAAM,CAIR;AACD,wBAAgB,0BAA0B,CACxC,UAAU,EAAE,MAAM,GACjB,eAAe,CAAC,kBAAkB,EAAE,kBAAkB,CAAC,CAMzD"}
@@ -0,0 +1,89 @@
1
+ /*
2
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
3
+ */
4
+ import * as z from "zod/v3";
5
+ import { remap as remap$ } from "../../lib/primitives.js";
6
+ import { safeParse } from "../../lib/schemas.js";
7
+ /** @internal */
8
+ export const AccuracyCi$inboundSchema = z.object({
9
+ low: z.number(),
10
+ high: z.number(),
11
+ margin: z.number(),
12
+ });
13
+ /** @internal */
14
+ export const AccuracyCi$outboundSchema = z.object({
15
+ low: z.number(),
16
+ high: z.number(),
17
+ margin: z.number(),
18
+ });
19
+ export function accuracyCiToJSON(accuracyCi) {
20
+ return JSON.stringify(AccuracyCi$outboundSchema.parse(accuracyCi));
21
+ }
22
+ export function accuracyCiFromJSON(jsonString) {
23
+ return safeParse(jsonString, (x) => AccuracyCi$inboundSchema.parse(JSON.parse(x)), `Failed to parse 'AccuracyCi' from JSON`);
24
+ }
25
+ /** @internal */
26
+ export const EvalLeaderboardRow$inboundSchema = z.object({
27
+ agent_name: z.string(),
28
+ agent_version: z.nullable(z.string()),
29
+ agent_effort: z.nullable(z.string()),
30
+ model: z.string(),
31
+ accuracy: z.nullable(z.number()),
32
+ accuracy_ci: z.nullable(z.lazy(() => AccuracyCi$inboundSchema)),
33
+ trials: z.number().int(),
34
+ scored_trials: z.number().int(),
35
+ failed_trials: z.number().int(),
36
+ cost_usd: z.nullable(z.number()),
37
+ run_count: z.number().int(),
38
+ run_ids: z.array(z.string()),
39
+ last_run_at: z.number().int(),
40
+ }).transform((v) => {
41
+ return remap$(v, {
42
+ "agent_name": "agentName",
43
+ "agent_version": "agentVersion",
44
+ "agent_effort": "agentEffort",
45
+ "accuracy_ci": "accuracyCi",
46
+ "scored_trials": "scoredTrials",
47
+ "failed_trials": "failedTrials",
48
+ "cost_usd": "costUsd",
49
+ "run_count": "runCount",
50
+ "run_ids": "runIds",
51
+ "last_run_at": "lastRunAt",
52
+ });
53
+ });
54
+ /** @internal */
55
+ export const EvalLeaderboardRow$outboundSchema = z.object({
56
+ agentName: z.string(),
57
+ agentVersion: z.nullable(z.string()),
58
+ agentEffort: z.nullable(z.string()),
59
+ model: z.string(),
60
+ accuracy: z.nullable(z.number()),
61
+ accuracyCi: z.nullable(z.lazy(() => AccuracyCi$outboundSchema)),
62
+ trials: z.number().int(),
63
+ scoredTrials: z.number().int(),
64
+ failedTrials: z.number().int(),
65
+ costUsd: z.nullable(z.number()),
66
+ runCount: z.number().int(),
67
+ runIds: z.array(z.string()),
68
+ lastRunAt: z.number().int(),
69
+ }).transform((v) => {
70
+ return remap$(v, {
71
+ agentName: "agent_name",
72
+ agentVersion: "agent_version",
73
+ agentEffort: "agent_effort",
74
+ accuracyCi: "accuracy_ci",
75
+ scoredTrials: "scored_trials",
76
+ failedTrials: "failed_trials",
77
+ costUsd: "cost_usd",
78
+ runCount: "run_count",
79
+ runIds: "run_ids",
80
+ lastRunAt: "last_run_at",
81
+ });
82
+ });
83
+ export function evalLeaderboardRowToJSON(evalLeaderboardRow) {
84
+ return JSON.stringify(EvalLeaderboardRow$outboundSchema.parse(evalLeaderboardRow));
85
+ }
86
+ export function evalLeaderboardRowFromJSON(jsonString) {
87
+ return safeParse(jsonString, (x) => EvalLeaderboardRow$inboundSchema.parse(JSON.parse(x)), `Failed to parse 'EvalLeaderboardRow' from JSON`);
88
+ }
89
+ //# sourceMappingURL=evalleaderboardrow.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"evalleaderboardrow.js","sourceRoot":"","sources":["../../../../src/models/components/evalleaderboardrow.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,QAAQ,CAAC;AAC5B,OAAO,EAAE,KAAK,IAAI,MAAM,EAAE,MAAM,yBAAyB,CAAC;AAC1D,OAAO,EAAE,SAAS,EAAE,MAAM,sBAAsB,CAAC;AAqDjD,gBAAgB;AAChB,MAAM,CAAC,MAAM,wBAAwB,GAIjC,CAAC,CAAC,MAAM,CAAC;IACX,GAAG,EAAE,CAAC,CAAC,MAAM,EAAE;IACf,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE;IAChB,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE;CACnB,CAAC,CAAC;AAQH,gBAAgB;AAChB,MAAM,CAAC,MAAM,yBAAyB,GAIlC,CAAC,CAAC,MAAM,CAAC;IACX,GAAG,EAAE,CAAC,CAAC,MAAM,EAAE;IACf,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE;IAChB,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE;CACnB,CAAC,CAAC;AAEH,MAAM,UAAU,gBAAgB,CAAC,UAAsB;IACrD,OAAO,IAAI,CAAC,SAAS,CAAC,yBAAyB,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC,CAAC;AACrE,CAAC;AACD,MAAM,UAAU,kBAAkB,CAChC,UAAkB;IAElB,OAAO,SAAS,CACd,UAAU,EACV,CAAC,CAAC,EAAE,EAAE,CAAC,wBAAwB,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,EACpD,wCAAwC,CACzC,CAAC;AACJ,CAAC;AAED,gBAAgB;AAChB,MAAM,CAAC,MAAM,gCAAgC,GAIzC,CAAC,CAAC,MAAM,CAAC;IACX,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE;IACtB,aAAa,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC;IACrC,YAAY,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC;IACpC,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE;IACjB,QAAQ,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC;IAChC,WAAW,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC,wBAAwB,CAAC,CAAC;IAC/D,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE;IACxB,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE;IAC/B,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE;IAC/B,QAAQ,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC;IAChC,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE;IAC3B,OAAO,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC;IAC5B,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE;CAC9B,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,EAAE;IACjB,OAAO,MAAM,CAAC,CAAC,EAAE;QACf,YAAY,EAAE,WAAW;QACzB,eAAe,EAAE,cAAc;QAC/B,cAAc,EAAE,aAAa;QAC7B,aAAa,EAAE,YAAY;QAC3B,eAAe,EAAE,cAAc;QAC/B,eAAe,EAAE,cAAc;QAC/B,UAAU,EAAE,SAAS;QACrB,WAAW,EAAE,UAAU;QACvB,SAAS,EAAE,QAAQ;QACnB,aAAa,EAAE,WAAW;KAC3B,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAkBH,gBAAgB;AAChB,MAAM,CAAC,MAAM,iCAAiC,GAI1C,CAAC,CAAC,MAAM,CAAC;IACX,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE;IACrB,YAAY,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC;IACpC,WAAW,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC;IACnC,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE;IACjB,QAAQ,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC;IAChC,UAAU,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC,yBAAyB,CAAC,CAAC;IAC/D,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE;IACxB,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE;IAC9B,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE;IAC9B,OAAO,EAAE,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC;IAC/B,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE;IAC1B,MAAM,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC;IAC3B,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE;CAC5B,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,EAAE;IACjB,OAAO,MAAM,CAAC,CAAC,EAAE;QACf,SAAS,EAAE,YAAY;QACvB,YAAY,EAAE,eAAe;QAC7B,WAAW,EAAE,cAAc;QAC3B,UAAU,EAAE,aAAa;QACzB,YAAY,EAAE,eAAe;QAC7B,YAAY,EAAE,eAAe;QAC7B,OAAO,EAAE,UAAU;QACnB,QAAQ,EAAE,WAAW;QACrB,MAAM,EAAE,SAAS;QACjB,SAAS,EAAE,aAAa;KACzB,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,MAAM,UAAU,wBAAwB,CACtC,kBAAsC;IAEtC,OAAO,IAAI,CAAC,SAAS,CACnB,iCAAiC,CAAC,KAAK,CAAC,kBAAkB,CAAC,CAC5D,CAAC;AACJ,CAAC;AACD,MAAM,UAAU,0BAA0B,CACxC,UAAkB;IAElB,OAAO,SAAS,CACd,UAAU,EACV,CAAC,CAAC,EAAE,EAAE,CAAC,gCAAgC,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,EAC5D,gDAAgD,CACjD,CAAC;AACJ,CAAC"}
@@ -0,0 +1,101 @@
1
+ import * as z from "zod/v3";
2
+ import { Result as SafeParseResult } from "../../types/fp.js";
3
+ import { SDKValidationError } from "../errors/sdkvalidationerror.js";
4
+ import { EvalRunStatus } from "./evalrunstatus.js";
5
+ export type Trials = {
6
+ total: number;
7
+ completed: number;
8
+ /**
9
+ * Trials in the `failed` status — no score produced.
10
+ */
11
+ failed: number;
12
+ /**
13
+ * Trials with a non-null `error`. Overlaps `failed` rather than partitioning it: a harness trial can carry an agent timeout and still be completed and scored.
14
+ */
15
+ errored: number;
16
+ /**
17
+ * Trials that were rerun through rerun-failed, counted from the pre-rerun snapshot. 0 when the run was never rerun.
18
+ */
19
+ retried: number;
20
+ };
21
+ /**
22
+ * Summed over the run's trials. Each field is null when no trial reported it.
23
+ */
24
+ export type Tokens = {
25
+ input: number | null;
26
+ output: number | null;
27
+ /**
28
+ * Share of `input` served from the provider's prompt cache.
29
+ */
30
+ cached: number | null;
31
+ };
32
+ export type EvalRunOverview = {
33
+ object: "eval.run.overview";
34
+ runId: string;
35
+ suiteId: string;
36
+ status: EvalRunStatus;
37
+ agentName: string | null;
38
+ agentVersion: string | null;
39
+ agentEffort: string | null;
40
+ models: Array<string>;
41
+ metric: string;
42
+ /**
43
+ * Mean of `metric` over trials that carry it numerically. Null when none do.
44
+ */
45
+ avgReward: number | null;
46
+ trials: Trials;
47
+ costUsd: number | null;
48
+ /**
49
+ * Summed over the run's trials. Each field is null when no trial reported it.
50
+ */
51
+ tokens: Tokens;
52
+ };
53
+ /** @internal */
54
+ export declare const Trials$inboundSchema: z.ZodType<Trials, z.ZodTypeDef, unknown>;
55
+ /** @internal */
56
+ export type Trials$Outbound = {
57
+ total: number;
58
+ completed: number;
59
+ failed: number;
60
+ errored: number;
61
+ retried: number;
62
+ };
63
+ /** @internal */
64
+ export declare const Trials$outboundSchema: z.ZodType<Trials$Outbound, z.ZodTypeDef, Trials>;
65
+ export declare function trialsToJSON(trials: Trials): string;
66
+ export declare function trialsFromJSON(jsonString: string): SafeParseResult<Trials, SDKValidationError>;
67
+ /** @internal */
68
+ export declare const Tokens$inboundSchema: z.ZodType<Tokens, z.ZodTypeDef, unknown>;
69
+ /** @internal */
70
+ export type Tokens$Outbound = {
71
+ input: number | null;
72
+ output: number | null;
73
+ cached: number | null;
74
+ };
75
+ /** @internal */
76
+ export declare const Tokens$outboundSchema: z.ZodType<Tokens$Outbound, z.ZodTypeDef, Tokens>;
77
+ export declare function tokensToJSON(tokens: Tokens): string;
78
+ export declare function tokensFromJSON(jsonString: string): SafeParseResult<Tokens, SDKValidationError>;
79
+ /** @internal */
80
+ export declare const EvalRunOverview$inboundSchema: z.ZodType<EvalRunOverview, z.ZodTypeDef, unknown>;
81
+ /** @internal */
82
+ export type EvalRunOverview$Outbound = {
83
+ object: "eval.run.overview";
84
+ run_id: string;
85
+ suite_id: string;
86
+ status: string;
87
+ agent_name: string | null;
88
+ agent_version: string | null;
89
+ agent_effort: string | null;
90
+ models: Array<string>;
91
+ metric: string;
92
+ avg_reward: number | null;
93
+ trials: Trials$Outbound;
94
+ cost_usd: number | null;
95
+ tokens: Tokens$Outbound;
96
+ };
97
+ /** @internal */
98
+ export declare const EvalRunOverview$outboundSchema: z.ZodType<EvalRunOverview$Outbound, z.ZodTypeDef, EvalRunOverview>;
99
+ export declare function evalRunOverviewToJSON(evalRunOverview: EvalRunOverview): string;
100
+ export declare function evalRunOverviewFromJSON(jsonString: string): SafeParseResult<EvalRunOverview, SDKValidationError>;
101
+ //# sourceMappingURL=evalrunoverview.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"evalrunoverview.d.ts","sourceRoot":"","sources":["../../../../src/models/components/evalrunoverview.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,QAAQ,CAAC;AAG5B,OAAO,EAAE,MAAM,IAAI,eAAe,EAAE,MAAM,mBAAmB,CAAC;AAC9D,OAAO,EAAE,kBAAkB,EAAE,MAAM,iCAAiC,CAAC;AACrE,OAAO,EACL,aAAa,EAGd,MAAM,oBAAoB,CAAC;AAE5B,MAAM,MAAM,MAAM,GAAG;IACnB,KAAK,EAAE,MAAM,CAAC;IACd,SAAS,EAAE,MAAM,CAAC;IAClB;;OAEG;IACH,MAAM,EAAE,MAAM,CAAC;IACf;;OAEG;IACH,OAAO,EAAE,MAAM,CAAC;IAChB;;OAEG;IACH,OAAO,EAAE,MAAM,CAAC;CACjB,CAAC;AAEF;;GAEG;AACH,MAAM,MAAM,MAAM,GAAG;IACnB,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB;;OAEG;IACH,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;CACvB,CAAC;AAEF,MAAM,MAAM,eAAe,GAAG;IAC5B,MAAM,EAAE,mBAAmB,CAAC;IAC5B,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,aAAa,CAAC;IACtB,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,YAAY,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B,MAAM,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;IACtB,MAAM,EAAE,MAAM,CAAC;IACf;;OAEG;IACH,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IACvB;;OAEG;IACH,MAAM,EAAE,MAAM,CAAC;CAChB,CAAC;AAEF,gBAAgB;AAChB,eAAO,MAAM,oBAAoB,EAAE,CAAC,CAAC,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,UAAU,EAAE,OAAO,CAOtE,CAAC;AACL,gBAAgB;AAChB,MAAM,MAAM,eAAe,GAAG;IAC5B,KAAK,EAAE,MAAM,CAAC;IACd,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,CAAC;IAChB,OAAO,EAAE,MAAM,CAAC;CACjB,CAAC;AAEF,gBAAgB;AAChB,eAAO,MAAM,qBAAqB,EAAE,CAAC,CAAC,OAAO,CAC3C,eAAe,EACf,CAAC,CAAC,UAAU,EACZ,MAAM,CAON,CAAC;AAEH,wBAAgB,YAAY,CAAC,MAAM,EAAE,MAAM,GAAG,MAAM,CAEnD;AACD,wBAAgB,cAAc,CAC5B,UAAU,EAAE,MAAM,GACjB,eAAe,CAAC,MAAM,EAAE,kBAAkB,CAAC,CAM7C;AAED,gBAAgB;AAChB,eAAO,MAAM,oBAAoB,EAAE,CAAC,CAAC,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,UAAU,EAAE,OAAO,CAKtE,CAAC;AACL,gBAAgB;AAChB,MAAM,MAAM,eAAe,GAAG;IAC5B,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;CACvB,CAAC;AAEF,gBAAgB;AAChB,eAAO,MAAM,qBAAqB,EAAE,CAAC,CAAC,OAAO,CAC3C,eAAe,EACf,CAAC,CAAC,UAAU,EACZ,MAAM,CAKN,CAAC;AAEH,wBAAgB,YAAY,CAAC,MAAM,EAAE,MAAM,GAAG,MAAM,CAEnD;AACD,wBAAgB,cAAc,CAC5B,UAAU,EAAE,MAAM,GACjB,eAAe,CAAC,MAAM,EAAE,kBAAkB,CAAC,CAM7C;AAED,gBAAgB;AAChB,eAAO,MAAM,6BAA6B,EAAE,CAAC,CAAC,OAAO,CACnD,eAAe,EACf,CAAC,CAAC,UAAU,EACZ,OAAO,CAyBP,CAAC;AACH,gBAAgB;AAChB,MAAM,MAAM,wBAAwB,GAAG;IACrC,MAAM,EAAE,mBAAmB,CAAC;IAC5B,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,CAAC;IACf,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,YAAY,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,MAAM,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;IACtB,MAAM,EAAE,MAAM,CAAC;IACf,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,MAAM,EAAE,eAAe,CAAC;IACxB,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB,MAAM,EAAE,eAAe,CAAC;CACzB,CAAC;AAEF,gBAAgB;AAChB,eAAO,MAAM,8BAA8B,EAAE,CAAC,CAAC,OAAO,CACpD,wBAAwB,EACxB,CAAC,CAAC,UAAU,EACZ,eAAe,CAyBf,CAAC;AAEH,wBAAgB,qBAAqB,CACnC,eAAe,EAAE,eAAe,GAC/B,MAAM,CAER;AACD,wBAAgB,uBAAuB,CACrC,UAAU,EAAE,MAAM,GACjB,eAAe,CAAC,eAAe,EAAE,kBAAkB,CAAC,CAMtD"}