@lightsage/mcp 0.3.1 → 0.4.1

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 (402) hide show
  1. package/README.md +2 -2
  2. package/api/well-known-openai-apps-challenge.js +25 -0
  3. package/bin/mcp-server.js +1878 -1023
  4. package/bin/mcp-server.js.map +77 -58
  5. package/esm/funcs/evalTemplatesV2EvalTemplatesCreate.d.ts +16 -0
  6. package/esm/funcs/evalTemplatesV2EvalTemplatesCreate.d.ts.map +1 -0
  7. package/esm/funcs/evalTemplatesV2EvalTemplatesCreate.js +83 -0
  8. package/esm/funcs/evalTemplatesV2EvalTemplatesCreate.js.map +1 -0
  9. package/esm/funcs/evalTemplatesV2EvalTemplatesDelete.d.ts +16 -0
  10. package/esm/funcs/evalTemplatesV2EvalTemplatesDelete.d.ts.map +1 -0
  11. package/esm/funcs/{evalResultsV2EvalResultsList.js → evalTemplatesV2EvalTemplatesDelete.js} +17 -16
  12. package/esm/funcs/evalTemplatesV2EvalTemplatesDelete.js.map +1 -0
  13. package/esm/funcs/evalTemplatesV2EvalTemplatesList.d.ts +15 -0
  14. package/esm/funcs/evalTemplatesV2EvalTemplatesList.d.ts.map +1 -0
  15. package/esm/funcs/evalTemplatesV2EvalTemplatesList.js +72 -0
  16. package/esm/funcs/evalTemplatesV2EvalTemplatesList.js.map +1 -0
  17. package/esm/funcs/evalTemplatesV2EvalTemplatesRetrieve.d.ts +16 -0
  18. package/esm/funcs/evalTemplatesV2EvalTemplatesRetrieve.d.ts.map +1 -0
  19. package/esm/funcs/{evalResultsV2EvalResultsRetrieve.js → evalTemplatesV2EvalTemplatesRetrieve.js} +10 -10
  20. package/esm/funcs/evalTemplatesV2EvalTemplatesRetrieve.js.map +1 -0
  21. package/esm/funcs/evalTemplatesV2EvalTemplatesUpdate.d.ts +16 -0
  22. package/esm/funcs/evalTemplatesV2EvalTemplatesUpdate.d.ts.map +1 -0
  23. package/esm/funcs/evalTemplatesV2EvalTemplatesUpdate.js +89 -0
  24. package/esm/funcs/evalTemplatesV2EvalTemplatesUpdate.js.map +1 -0
  25. package/esm/funcs/evalsCancel.d.ts +16 -0
  26. package/esm/funcs/evalsCancel.d.ts.map +1 -0
  27. package/esm/funcs/{evalsAnalysis.js → evalsCancel.js} +10 -10
  28. package/esm/funcs/evalsCancel.js.map +1 -0
  29. package/esm/funcs/evalsGet.d.ts +16 -0
  30. package/esm/funcs/evalsGet.d.ts.map +1 -0
  31. package/esm/funcs/{evalRunsV2EvalRunsRetrieve.js → evalsGet.js} +5 -5
  32. package/esm/funcs/evalsGet.js.map +1 -0
  33. package/esm/funcs/evalsRun.d.ts +6 -4
  34. package/esm/funcs/evalsRun.d.ts.map +1 -1
  35. package/esm/funcs/evalsRun.js +6 -4
  36. package/esm/funcs/evalsRun.js.map +1 -1
  37. package/esm/funcs/evalsV2EvalsBatchCreate.d.ts +16 -0
  38. package/esm/funcs/evalsV2EvalsBatchCreate.d.ts.map +1 -0
  39. package/esm/funcs/evalsV2EvalsBatchCreate.js +83 -0
  40. package/esm/funcs/evalsV2EvalsBatchCreate.js.map +1 -0
  41. package/esm/funcs/evalsV2EvalsCreate.d.ts +1 -1
  42. package/esm/funcs/evalsV2EvalsCreate.js +1 -1
  43. package/esm/funcs/evalsV2EvalsUpdate.d.ts +1 -1
  44. package/esm/funcs/evalsV2EvalsUpdate.js +1 -1
  45. package/esm/funcs/promptsV2PromptRunsList.d.ts +1 -1
  46. package/esm/funcs/promptsV2PromptRunsList.js +1 -1
  47. package/esm/funcs/promptsV2PromptRunsRetrieve.d.ts +1 -1
  48. package/esm/funcs/promptsV2PromptRunsRetrieve.js +1 -1
  49. package/esm/landing-page.js +1 -1
  50. package/esm/lib/config.d.ts +3 -3
  51. package/esm/lib/config.js +3 -3
  52. package/esm/lib/config.js.map +1 -1
  53. package/esm/mcp-server/mcp-server.js +1 -1
  54. package/esm/mcp-server/server.d.ts.map +1 -1
  55. package/esm/mcp-server/server.js +17 -9
  56. package/esm/mcp-server/server.js.map +1 -1
  57. package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesCreate.d.ts +7 -0
  58. package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesCreate.d.ts.map +1 -0
  59. package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesCreate.js +35 -0
  60. package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesCreate.js.map +1 -0
  61. package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesDelete.d.ts +7 -0
  62. package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesDelete.d.ts.map +1 -0
  63. package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesDelete.js +35 -0
  64. package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesDelete.js.map +1 -0
  65. package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesList.d.ts +3 -0
  66. package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesList.d.ts.map +1 -0
  67. package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesList.js +30 -0
  68. package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesList.js.map +1 -0
  69. package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesRetrieve.d.ts +7 -0
  70. package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesRetrieve.d.ts.map +1 -0
  71. package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesRetrieve.js +35 -0
  72. package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesRetrieve.js.map +1 -0
  73. package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesUpdate.d.ts +7 -0
  74. package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesUpdate.d.ts.map +1 -0
  75. package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesUpdate.js +35 -0
  76. package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesUpdate.js.map +1 -0
  77. package/esm/mcp-server/tools/evalsCancel.d.ts +7 -0
  78. package/esm/mcp-server/tools/evalsCancel.d.ts.map +1 -0
  79. package/esm/mcp-server/tools/evalsCancel.js +35 -0
  80. package/esm/mcp-server/tools/evalsCancel.js.map +1 -0
  81. package/esm/mcp-server/tools/{evalRunsV2EvalRunsRetrieve.d.ts → evalsGet.d.ts} +2 -2
  82. package/esm/mcp-server/tools/evalsGet.d.ts.map +1 -0
  83. package/esm/mcp-server/tools/{evalRunsV2EvalRunsRetrieve.js → evalsGet.js} +8 -8
  84. package/esm/mcp-server/tools/evalsGet.js.map +1 -0
  85. package/esm/mcp-server/tools/evalsRun.d.ts.map +1 -1
  86. package/esm/mcp-server/tools/evalsRun.js +6 -4
  87. package/esm/mcp-server/tools/evalsRun.js.map +1 -1
  88. package/esm/mcp-server/tools/evalsV2EvalsBatchCreate.d.ts +7 -0
  89. package/esm/mcp-server/tools/evalsV2EvalsBatchCreate.d.ts.map +1 -0
  90. package/esm/mcp-server/tools/evalsV2EvalsBatchCreate.js +35 -0
  91. package/esm/mcp-server/tools/evalsV2EvalsBatchCreate.js.map +1 -0
  92. package/esm/mcp-server/tools/evalsV2EvalsCreate.js +1 -1
  93. package/esm/mcp-server/tools/evalsV2EvalsCreate.js.map +1 -1
  94. package/esm/mcp-server/tools/evalsV2EvalsUpdate.js +1 -1
  95. package/esm/mcp-server/tools/evalsV2EvalsUpdate.js.map +1 -1
  96. package/esm/mcp-server/tools/promptsV2PromptRunsList.js +1 -1
  97. package/esm/mcp-server/tools/promptsV2PromptRunsList.js.map +1 -1
  98. package/esm/mcp-server/tools/promptsV2PromptRunsRetrieve.js +1 -1
  99. package/esm/mcp-server/tools/promptsV2PromptRunsRetrieve.js.map +1 -1
  100. package/esm/models/createevaltemplatebatchrequest.d.ts +12 -0
  101. package/esm/models/createevaltemplatebatchrequest.d.ts.map +1 -0
  102. package/esm/models/createevaltemplatebatchrequest.js +12 -0
  103. package/esm/models/createevaltemplatebatchrequest.js.map +1 -0
  104. package/esm/models/createevaltemplaterequest.d.ts +79 -0
  105. package/esm/models/createevaltemplaterequest.d.ts.map +1 -0
  106. package/esm/models/createevaltemplaterequest.js +74 -0
  107. package/esm/models/createevaltemplaterequest.js.map +1 -0
  108. package/esm/models/createv2evalconfigurationrequest.d.ts +2 -0
  109. package/esm/models/createv2evalconfigurationrequest.d.ts.map +1 -1
  110. package/esm/models/createv2evalconfigurationrequest.js +2 -0
  111. package/esm/models/createv2evalconfigurationrequest.js.map +1 -1
  112. package/esm/models/createv2evalrequest.d.ts +5 -5
  113. package/esm/models/createv2evalrequest.d.ts.map +1 -1
  114. package/esm/models/createv2evalrequest.js +5 -10
  115. package/esm/models/createv2evalrequest.js.map +1 -1
  116. package/esm/models/createv2evalrunrequest.d.ts +5 -5
  117. package/esm/models/createv2evalrunrequest.d.ts.map +1 -1
  118. package/esm/models/createv2evalrunrequest.js +6 -10
  119. package/esm/models/createv2evalrunrequest.js.map +1 -1
  120. package/esm/models/evaltemplate.d.ts +83 -0
  121. package/esm/models/evaltemplate.d.ts.map +1 -0
  122. package/esm/models/evaltemplate.js +81 -0
  123. package/esm/models/evaltemplate.js.map +1 -0
  124. package/esm/models/evaltemplatebatchcreatedeval.d.ts +11 -0
  125. package/esm/models/evaltemplatebatchcreatedeval.d.ts.map +1 -0
  126. package/esm/models/evaltemplatebatchcreatedeval.js +14 -0
  127. package/esm/models/evaltemplatebatchcreatedeval.js.map +1 -0
  128. package/esm/models/evaltemplatebatchitem.d.ts +32 -0
  129. package/esm/models/evaltemplatebatchitem.d.ts.map +1 -0
  130. package/esm/models/evaltemplatebatchitem.js +27 -0
  131. package/esm/models/evaltemplatebatchitem.js.map +1 -0
  132. package/esm/models/evaltemplatebatchresponse.d.ts +12 -0
  133. package/esm/models/evaltemplatebatchresponse.d.ts.map +1 -0
  134. package/esm/models/evaltemplatebatchresponse.js +15 -0
  135. package/esm/models/evaltemplatebatchresponse.js.map +1 -0
  136. package/esm/models/evaltemplatebatchresult.d.ts +10 -0
  137. package/esm/models/evaltemplatebatchresult.d.ts.map +1 -0
  138. package/esm/models/evaltemplatebatchresult.js +13 -0
  139. package/esm/models/evaltemplatebatchresult.js.map +1 -0
  140. package/esm/models/evaltemplateruntime.d.ts +20 -0
  141. package/esm/models/evaltemplateruntime.d.ts.map +1 -0
  142. package/esm/models/evaltemplateruntime.js +21 -0
  143. package/esm/models/evaltemplateruntime.js.map +1 -0
  144. package/esm/models/evaltemplateschedule.d.ts +28 -0
  145. package/esm/models/evaltemplateschedule.d.ts.map +1 -0
  146. package/esm/models/evaltemplateschedule.js +24 -0
  147. package/esm/models/evaltemplateschedule.js.map +1 -0
  148. package/esm/models/evaltemplatesresponse.d.ts +8 -0
  149. package/esm/models/evaltemplatesresponse.d.ts.map +1 -0
  150. package/esm/models/evaltemplatesresponse.js +11 -0
  151. package/esm/models/evaltemplatesresponse.js.map +1 -0
  152. package/esm/models/graderdefinition.d.ts +14 -0
  153. package/esm/models/graderdefinition.d.ts.map +1 -0
  154. package/esm/models/graderdefinition.js +17 -0
  155. package/esm/models/graderdefinition.js.map +1 -0
  156. package/esm/models/publiccompetitorrow.js +2 -2
  157. package/esm/models/publiccompetitorrow.js.map +1 -1
  158. package/esm/models/publiccriterionverdict.d.ts +26 -0
  159. package/esm/models/publiccriterionverdict.d.ts.map +1 -0
  160. package/esm/models/publiccriterionverdict.js +24 -0
  161. package/esm/models/publiccriterionverdict.js.map +1 -0
  162. package/esm/models/publiccurrentvisibilityresponse.js +2 -2
  163. package/esm/models/publiccurrentvisibilityresponse.js.map +1 -1
  164. package/esm/models/publicevalrunenvironment.d.ts +15 -0
  165. package/esm/models/publicevalrunenvironment.d.ts.map +1 -0
  166. package/esm/models/publicevalrunenvironment.js +15 -0
  167. package/esm/models/publicevalrunenvironment.js.map +1 -0
  168. package/esm/models/publicevalrunprogress.d.ts +1 -0
  169. package/esm/models/publicevalrunprogress.d.ts.map +1 -1
  170. package/esm/models/publicevalrunprogress.js +3 -2
  171. package/esm/models/publicevalrunprogress.js.map +1 -1
  172. package/esm/models/publicgraderresult.d.ts +13 -0
  173. package/esm/models/publicgraderresult.d.ts.map +1 -0
  174. package/esm/models/publicgraderresult.js +18 -0
  175. package/esm/models/publicgraderresult.js.map +1 -0
  176. package/esm/models/publicpromptvisibilitymetrics.js +2 -2
  177. package/esm/models/publicpromptvisibilitymetrics.js.map +1 -1
  178. package/esm/models/publicvisibilitysnapshot.js +2 -2
  179. package/esm/models/publicvisibilitysnapshot.js.map +1 -1
  180. package/esm/models/searchgroundingmode.d.ts +20 -0
  181. package/esm/models/searchgroundingmode.d.ts.map +1 -0
  182. package/esm/models/searchgroundingmode.js +19 -0
  183. package/esm/models/searchgroundingmode.js.map +1 -0
  184. package/esm/models/updateevaltemplaterequest.d.ts +71 -0
  185. package/esm/models/updateevaltemplaterequest.d.ts.map +1 -0
  186. package/esm/models/updateevaltemplaterequest.js +71 -0
  187. package/esm/models/updateevaltemplaterequest.js.map +1 -0
  188. package/esm/models/updatev2evalconfigurationrequest.d.ts +2 -0
  189. package/esm/models/updatev2evalconfigurationrequest.d.ts.map +1 -1
  190. package/esm/models/updatev2evalconfigurationrequest.js +3 -0
  191. package/esm/models/updatev2evalconfigurationrequest.js.map +1 -1
  192. package/esm/models/updatev2evalrequest.d.ts +5 -5
  193. package/esm/models/updatev2evalrequest.d.ts.map +1 -1
  194. package/esm/models/updatev2evalrequest.js +6 -10
  195. package/esm/models/updatev2evalrequest.js.map +1 -1
  196. package/esm/models/v2eval.d.ts +6 -2
  197. package/esm/models/v2eval.d.ts.map +1 -1
  198. package/esm/models/v2eval.js +5 -1
  199. package/esm/models/v2eval.js.map +1 -1
  200. package/esm/models/v2evalconfiguration.d.ts +2 -0
  201. package/esm/models/v2evalconfiguration.d.ts.map +1 -1
  202. package/esm/models/v2evalconfiguration.js +2 -0
  203. package/esm/models/v2evalconfiguration.js.map +1 -1
  204. package/esm/models/v2evalresultsummary.d.ts +10 -0
  205. package/esm/models/v2evalresultsummary.d.ts.map +1 -1
  206. package/esm/models/v2evalresultsummary.js +12 -1
  207. package/esm/models/v2evalresultsummary.js.map +1 -1
  208. package/esm/models/v2evalrun.d.ts.map +1 -1
  209. package/esm/models/v2evalrun.js +1 -1
  210. package/esm/models/v2evalrun.js.map +1 -1
  211. package/esm/models/v2evalrundetail.d.ts +30 -0
  212. package/esm/models/v2evalrundetail.d.ts.map +1 -0
  213. package/esm/models/v2evalrundetail.js +33 -0
  214. package/esm/models/v2evalrundetail.js.map +1 -0
  215. package/esm/models/v2evalrunscancelop.d.ts +31 -0
  216. package/esm/models/v2evalrunscancelop.d.ts.map +1 -0
  217. package/esm/models/v2evalrunscancelop.js +27 -0
  218. package/esm/models/v2evalrunscancelop.js.map +1 -0
  219. package/esm/models/v2evalrunscreateop.d.ts +23 -3
  220. package/esm/models/v2evalrunscreateop.d.ts.map +1 -1
  221. package/esm/models/v2evalrunscreateop.js +15 -2
  222. package/esm/models/v2evalrunscreateop.js.map +1 -1
  223. package/esm/models/v2evalrunsretrieveop.d.ts +9 -2
  224. package/esm/models/v2evalrunsretrieveop.d.ts.map +1 -1
  225. package/esm/models/v2evalrunsretrieveop.js +6 -2
  226. package/esm/models/v2evalrunsretrieveop.js.map +1 -1
  227. package/esm/models/v2evalruntraceevent.d.ts.map +1 -1
  228. package/esm/models/v2evalruntraceevent.js +1 -1
  229. package/esm/models/v2evalruntraceevent.js.map +1 -1
  230. package/esm/models/v2evalsbatchcreateop.d.ts +27 -0
  231. package/esm/models/v2evalsbatchcreateop.d.ts.map +1 -0
  232. package/esm/models/v2evalsbatchcreateop.js +24 -0
  233. package/esm/models/v2evalsbatchcreateop.js.map +1 -0
  234. package/esm/models/v2evaltemplatescreateop.d.ts +13 -0
  235. package/esm/models/v2evaltemplatescreateop.d.ts.map +1 -0
  236. package/esm/models/v2evaltemplatescreateop.js +16 -0
  237. package/esm/models/v2evaltemplatescreateop.js.map +1 -0
  238. package/esm/models/v2evaltemplatesdeleteop.d.ts +24 -0
  239. package/esm/models/v2evaltemplatesdeleteop.d.ts.map +1 -0
  240. package/esm/models/v2evaltemplatesdeleteop.js +23 -0
  241. package/esm/models/v2evaltemplatesdeleteop.js.map +1 -0
  242. package/esm/models/v2evaltemplatesretrieveop.d.ts +17 -0
  243. package/esm/models/v2evaltemplatesretrieveop.d.ts.map +1 -0
  244. package/esm/models/v2evaltemplatesretrieveop.js +19 -0
  245. package/esm/models/v2evaltemplatesretrieveop.js.map +1 -0
  246. package/esm/models/v2evaltemplatesupdateop.d.ts +33 -0
  247. package/esm/models/v2evaltemplatesupdateop.d.ts.map +1 -0
  248. package/esm/models/v2evaltemplatesupdateop.js +29 -0
  249. package/esm/models/v2evaltemplatesupdateop.js.map +1 -0
  250. package/esm/models/v2inlinemcpdefinition.d.ts +4 -4
  251. package/esm/models/v2inlinemcpdefinition.d.ts.map +1 -1
  252. package/esm/models/v2inlinemcpdefinition.js +3 -3
  253. package/esm/models/v2inlinemcpdefinition.js.map +1 -1
  254. package/esm/models/v2promptrundetail.d.ts +6 -0
  255. package/esm/models/v2promptrundetail.d.ts.map +1 -1
  256. package/esm/models/v2promptrundetail.js +4 -1
  257. package/esm/models/v2promptrundetail.js.map +1 -1
  258. package/esm/models/v2promptrunsummary.d.ts.map +1 -1
  259. package/esm/models/v2promptrunsummary.js +1 -1
  260. package/esm/models/v2promptrunsummary.js.map +1 -1
  261. package/esm/models/v2publictask.d.ts +2 -2
  262. package/esm/models/v2publictask.d.ts.map +1 -1
  263. package/esm/models/v2publictask.js +4 -4
  264. package/esm/models/v2publictask.js.map +1 -1
  265. package/esm/tool-names.d.ts.map +1 -1
  266. package/esm/tool-names.js +31 -15
  267. package/esm/tool-names.js.map +1 -1
  268. package/package.json +1 -1
  269. package/src/funcs/evalTemplatesV2EvalTemplatesCreate.ts +145 -0
  270. package/src/funcs/{evalResultsV2EvalResultsList.ts → evalTemplatesV2EvalTemplatesDelete.ts} +23 -19
  271. package/src/funcs/evalTemplatesV2EvalTemplatesList.ts +124 -0
  272. package/src/funcs/{evalResultsV2EvalResultsRetrieve.ts → evalTemplatesV2EvalTemplatesRetrieve.ts} +13 -13
  273. package/src/funcs/evalTemplatesV2EvalTemplatesUpdate.ts +154 -0
  274. package/src/funcs/{evalsAnalysis.ts → evalsCancel.ts} +13 -13
  275. package/src/funcs/{evalRunsV2EvalRunsRetrieve.ts → evalsGet.ts} +4 -4
  276. package/src/funcs/evalsRun.ts +6 -4
  277. package/src/funcs/evalsV2EvalsBatchCreate.ts +145 -0
  278. package/src/funcs/evalsV2EvalsCreate.ts +1 -1
  279. package/src/funcs/evalsV2EvalsUpdate.ts +1 -1
  280. package/src/funcs/promptsV2PromptRunsList.ts +1 -1
  281. package/src/funcs/promptsV2PromptRunsRetrieve.ts +1 -1
  282. package/src/landing-page.ts +1 -1
  283. package/src/lib/config.ts +3 -3
  284. package/src/mcp-server/mcp-server.ts +1 -1
  285. package/src/mcp-server/server.ts +17 -9
  286. package/src/mcp-server/tools/evalTemplatesV2EvalTemplatesCreate.ts +45 -0
  287. package/src/mcp-server/tools/evalTemplatesV2EvalTemplatesDelete.ts +45 -0
  288. package/src/mcp-server/tools/evalTemplatesV2EvalTemplatesList.ts +36 -0
  289. package/src/mcp-server/tools/{evalResultsV2EvalResultsRetrieve.ts → evalTemplatesV2EvalTemplatesRetrieve.ts} +9 -9
  290. package/src/mcp-server/tools/evalTemplatesV2EvalTemplatesUpdate.ts +45 -0
  291. package/src/mcp-server/tools/evalsCancel.ts +43 -0
  292. package/src/mcp-server/tools/{evalRunsV2EvalRunsRetrieve.ts → evalsGet.ts} +7 -7
  293. package/src/mcp-server/tools/evalsRun.ts +6 -4
  294. package/src/mcp-server/tools/evalsV2EvalsBatchCreate.ts +43 -0
  295. package/src/mcp-server/tools/evalsV2EvalsCreate.ts +1 -1
  296. package/src/mcp-server/tools/evalsV2EvalsUpdate.ts +1 -1
  297. package/src/mcp-server/tools/promptsV2PromptRunsList.ts +1 -1
  298. package/src/mcp-server/tools/promptsV2PromptRunsRetrieve.ts +1 -1
  299. package/src/models/createevaltemplatebatchrequest.ts +35 -0
  300. package/src/models/createevaltemplaterequest.ts +151 -0
  301. package/src/models/createv2evalconfigurationrequest.ts +8 -0
  302. package/src/models/createv2evalrequest.ts +18 -22
  303. package/src/models/createv2evalrunrequest.ts +19 -24
  304. package/src/models/evaltemplate.ts +154 -0
  305. package/src/models/evaltemplatebatchcreatedeval.ts +26 -0
  306. package/src/models/evaltemplatebatchitem.ts +54 -0
  307. package/src/models/evaltemplatebatchresponse.ts +30 -0
  308. package/src/models/evaltemplatebatchresult.ts +26 -0
  309. package/src/models/evaltemplateruntime.ts +64 -0
  310. package/src/models/evaltemplateschedule.ts +44 -0
  311. package/src/models/evaltemplatesresponse.ts +18 -0
  312. package/src/models/graderdefinition.ts +34 -0
  313. package/src/models/publiccompetitorrow.ts +2 -2
  314. package/src/models/publiccriterionverdict.ts +48 -0
  315. package/src/models/publiccurrentvisibilityresponse.ts +2 -2
  316. package/src/models/publicevalrunenvironment.ts +47 -0
  317. package/src/models/publicevalrunprogress.ts +8 -2
  318. package/src/models/publicgraderresult.ts +44 -0
  319. package/src/models/publicpromptvisibilitymetrics.ts +2 -2
  320. package/src/models/publicvisibilitysnapshot.ts +2 -2
  321. package/src/models/searchgroundingmode.ts +26 -0
  322. package/src/models/updateevaltemplaterequest.ts +116 -0
  323. package/src/models/updatev2evalconfigurationrequest.ts +9 -0
  324. package/src/models/updatev2evalrequest.ts +20 -21
  325. package/src/models/v2eval.ts +14 -2
  326. package/src/models/v2evalconfiguration.ts +8 -0
  327. package/src/models/v2evalresultsummary.ts +36 -1
  328. package/src/models/v2evalrun.ts +3 -1
  329. package/src/models/v2evalrundetail.ts +90 -0
  330. package/src/models/v2evalrunscancelop.ts +70 -0
  331. package/src/models/v2evalrunscreateop.ts +43 -5
  332. package/src/models/v2evalrunsretrieveop.ts +23 -3
  333. package/src/models/v2evalruntraceevent.ts +3 -1
  334. package/src/models/v2evalsbatchcreateop.ts +66 -0
  335. package/src/models/v2evaltemplatescreateop.ts +37 -0
  336. package/src/models/v2evaltemplatesdeleteop.ts +56 -0
  337. package/src/models/v2evaltemplatesretrieveop.ts +43 -0
  338. package/src/models/v2evaltemplatesupdateop.ts +79 -0
  339. package/src/models/v2inlinemcpdefinition.ts +7 -5
  340. package/src/models/v2promptrundetail.ts +12 -1
  341. package/src/models/v2promptrunsummary.ts +3 -1
  342. package/src/models/v2publictask.ts +9 -7
  343. package/src/tool-names.ts +31 -15
  344. package/test/simulations.test.js +30 -3
  345. package/esm/funcs/evalResultsV2EvalResultsList.d.ts +0 -16
  346. package/esm/funcs/evalResultsV2EvalResultsList.d.ts.map +0 -1
  347. package/esm/funcs/evalResultsV2EvalResultsList.js.map +0 -1
  348. package/esm/funcs/evalResultsV2EvalResultsRetrieve.d.ts +0 -16
  349. package/esm/funcs/evalResultsV2EvalResultsRetrieve.d.ts.map +0 -1
  350. package/esm/funcs/evalResultsV2EvalResultsRetrieve.js.map +0 -1
  351. package/esm/funcs/evalRunsV2EvalRunsRetrieve.d.ts +0 -16
  352. package/esm/funcs/evalRunsV2EvalRunsRetrieve.d.ts.map +0 -1
  353. package/esm/funcs/evalRunsV2EvalRunsRetrieve.js.map +0 -1
  354. package/esm/funcs/evalsAnalysis.d.ts +0 -16
  355. package/esm/funcs/evalsAnalysis.d.ts.map +0 -1
  356. package/esm/funcs/evalsAnalysis.js.map +0 -1
  357. package/esm/mcp-server/tools/evalResultsV2EvalResultsList.d.ts +0 -7
  358. package/esm/mcp-server/tools/evalResultsV2EvalResultsList.d.ts.map +0 -1
  359. package/esm/mcp-server/tools/evalResultsV2EvalResultsList.js +0 -35
  360. package/esm/mcp-server/tools/evalResultsV2EvalResultsList.js.map +0 -1
  361. package/esm/mcp-server/tools/evalResultsV2EvalResultsRetrieve.d.ts +0 -7
  362. package/esm/mcp-server/tools/evalResultsV2EvalResultsRetrieve.d.ts.map +0 -1
  363. package/esm/mcp-server/tools/evalResultsV2EvalResultsRetrieve.js +0 -35
  364. package/esm/mcp-server/tools/evalResultsV2EvalResultsRetrieve.js.map +0 -1
  365. package/esm/mcp-server/tools/evalRunsV2EvalRunsRetrieve.d.ts.map +0 -1
  366. package/esm/mcp-server/tools/evalRunsV2EvalRunsRetrieve.js.map +0 -1
  367. package/esm/mcp-server/tools/evalsAnalysis.d.ts +0 -7
  368. package/esm/mcp-server/tools/evalsAnalysis.d.ts.map +0 -1
  369. package/esm/mcp-server/tools/evalsAnalysis.js +0 -35
  370. package/esm/mcp-server/tools/evalsAnalysis.js.map +0 -1
  371. package/esm/models/v2evalresult.d.ts +0 -23
  372. package/esm/models/v2evalresult.d.ts.map +0 -1
  373. package/esm/models/v2evalresult.js +0 -24
  374. package/esm/models/v2evalresult.js.map +0 -1
  375. package/esm/models/v2evalresultslistop.d.ts +0 -11
  376. package/esm/models/v2evalresultslistop.d.ts.map +0 -1
  377. package/esm/models/v2evalresultslistop.js +0 -16
  378. package/esm/models/v2evalresultslistop.js.map +0 -1
  379. package/esm/models/v2evalresultsresponse.d.ts +0 -9
  380. package/esm/models/v2evalresultsresponse.d.ts.map +0 -1
  381. package/esm/models/v2evalresultsresponse.js +0 -12
  382. package/esm/models/v2evalresultsresponse.js.map +0 -1
  383. package/esm/models/v2evalresultsretrieveop.d.ts +0 -10
  384. package/esm/models/v2evalresultsretrieveop.d.ts.map +0 -1
  385. package/esm/models/v2evalresultsretrieveop.js +0 -15
  386. package/esm/models/v2evalresultsretrieveop.js.map +0 -1
  387. package/esm/models/v2evalrunanalysisresponse.d.ts +0 -12
  388. package/esm/models/v2evalrunanalysisresponse.d.ts.map +0 -1
  389. package/esm/models/v2evalrunanalysisresponse.js +0 -15
  390. package/esm/models/v2evalrunanalysisresponse.js.map +0 -1
  391. package/esm/models/v2evalrunsanalysisop.d.ts +0 -17
  392. package/esm/models/v2evalrunsanalysisop.d.ts.map +0 -1
  393. package/esm/models/v2evalrunsanalysisop.js +0 -19
  394. package/esm/models/v2evalrunsanalysisop.js.map +0 -1
  395. package/src/mcp-server/tools/evalResultsV2EvalResultsList.ts +0 -43
  396. package/src/mcp-server/tools/evalsAnalysis.ts +0 -43
  397. package/src/models/v2evalresult.ts +0 -46
  398. package/src/models/v2evalresultslistop.ts +0 -39
  399. package/src/models/v2evalresultsresponse.ts +0 -23
  400. package/src/models/v2evalresultsretrieveop.ts +0 -28
  401. package/src/models/v2evalrunanalysisresponse.ts +0 -32
  402. package/src/models/v2evalrunsanalysisop.ts +0 -48
package/bin/mcp-server.js CHANGED
@@ -28171,6 +28171,9 @@ var require_proxy_addr = __commonJS(function(exports, module) {
28171
28171
  if (!isip(addr))
28172
28172
  return false;
28173
28173
  var ip = parseip(addr);
28174
+ if (ip.kind() === "ipv6" && ip.isIPv4MappedAddress()) {
28175
+ ip = ip.toIPv4Address();
28176
+ }
28174
28177
  var ipconv;
28175
28178
  var kind = ip.kind();
28176
28179
  for (var i = 0;i < subnets.length; i++) {
@@ -28183,10 +28186,15 @@ var require_proxy_addr = __commonJS(function(exports, module) {
28183
28186
  if (subnetkind === "ipv4" && !ip.isIPv4MappedAddress()) {
28184
28187
  continue;
28185
28188
  }
28189
+ if (subnetkind !== "ipv4" && !(subnetrange >= 96 && subnetip.isIPv4MappedAddress())) {
28190
+ continue;
28191
+ }
28186
28192
  if (!ipconv) {
28187
28193
  ipconv = subnetkind === "ipv4" ? ip.toIPv4Address() : ip.toIPv4MappedAddress();
28188
28194
  }
28189
28195
  trusted = ipconv;
28196
+ } else if (kind === "ipv6" && subnetip.isIPv4MappedAddress()) {
28197
+ continue;
28190
28198
  }
28191
28199
  if (trusted.match(subnetip, subnetrange)) {
28192
28200
  return true;
@@ -28204,12 +28212,20 @@ var require_proxy_addr = __commonJS(function(exports, module) {
28204
28212
  if (!isip(addr))
28205
28213
  return false;
28206
28214
  var ip = parseip(addr);
28215
+ if (ip.kind() === "ipv6" && ip.isIPv4MappedAddress()) {
28216
+ ip = ip.toIPv4Address();
28217
+ }
28207
28218
  var kind = ip.kind();
28208
28219
  if (kind !== subnetkind) {
28209
28220
  if (subnetisipv4 && !ip.isIPv4MappedAddress()) {
28210
28221
  return false;
28211
28222
  }
28223
+ if (!subnetisipv4 && !(subnetrange >= 96 && subnetip.isIPv4MappedAddress())) {
28224
+ return false;
28225
+ }
28212
28226
  ip = subnetisipv4 ? ip.toIPv4Address() : ip.toIPv4MappedAddress();
28227
+ } else if (kind === "ipv6" && subnetip.isIPv4MappedAddress()) {
28228
+ return false;
28213
28229
  }
28214
28230
  return ip.match(subnetip, subnetrange);
28215
28231
  };
@@ -42217,6 +42233,22 @@ var require_utils4 = __commonJS(function(exports, module) {
42217
42233
  }
42218
42234
  return output;
42219
42235
  }
42236
+ function recomposeZonedIPv6Host(component) {
42237
+ const zone = component.ipv6Zone;
42238
+ if (zone === undefined)
42239
+ return;
42240
+ const host = component.host;
42241
+ const separator = host.indexOf("%");
42242
+ if (separator === -1)
42243
+ return;
42244
+ const address = host.slice(0, separator);
42245
+ const escaped = address + "%25" + zone;
42246
+ const derived = normalizeIPv6("[" + escaped + "]");
42247
+ if (derived.isIPV6 === true && derived.host === host) {
42248
+ return "[" + escaped + "]";
42249
+ }
42250
+ return;
42251
+ }
42220
42252
  function recomposeAuthority(component) {
42221
42253
  const uriTokens = [];
42222
42254
  if (component.userinfo !== undefined) {
@@ -42226,15 +42258,20 @@ var require_utils4 = __commonJS(function(exports, module) {
42226
42258
  if (component.host !== undefined) {
42227
42259
  let host = component.host;
42228
42260
  if (!isIPv4(host)) {
42229
- let ipV6res = normalizeIPv6(host);
42230
- if (ipV6res.isIPV6 !== true && ipV6res.isIPVFuture !== true) {
42231
- host = normalizePercentEncoding(host, true);
42232
- ipV6res = normalizeIPv6(host);
42233
- }
42234
- if (ipV6res.isIPV6 === true || ipV6res.isIPVFuture === true) {
42235
- host = `[${ipV6res.escapedHost}]`;
42261
+ const zonedHost = recomposeZonedIPv6Host(component);
42262
+ if (zonedHost !== undefined) {
42263
+ host = zonedHost;
42236
42264
  } else {
42237
- host = reescapeHostDelimiters(host, false);
42265
+ let ipV6res = normalizeIPv6(host);
42266
+ if (ipV6res.isIPV6 !== true && ipV6res.isIPVFuture !== true) {
42267
+ host = normalizePercentEncoding(host, true);
42268
+ ipV6res = normalizeIPv6(host);
42269
+ }
42270
+ if (ipV6res.isIPV6 === true || ipV6res.isIPVFuture === true) {
42271
+ host = `[${ipV6res.escapedHost}]`;
42272
+ } else {
42273
+ host = reescapeHostDelimiters(host, false);
42274
+ }
42238
42275
  }
42239
42276
  }
42240
42277
  uriTokens.push(host);
@@ -42447,6 +42484,7 @@ var require_schemes = __commonJS(function(exports, module) {
42447
42484
  var HEX_PAIR = /^[\da-f]{2}$/iu;
42448
42485
  var MAILTO_DOMAIN_LITERAL = /^\[[\x21-\x5A\x5E-\x7E]*\]$/u;
42449
42486
  var MAILTO_DOMAIN_ERROR = "URI mailto has an invalid recipient domain.";
42487
+ var MAILTO_AUTHORITY_ERROR = "URI mailto must not have an authority component.";
42450
42488
  var HAS_SURROGATE = /[\uD800-\uDFFF]/u;
42451
42489
  function decodeHex(str) {
42452
42490
  if (typeof str !== "string" || str.indexOf("%") === -1) {
@@ -42561,30 +42599,37 @@ var require_schemes = __commonJS(function(exports, module) {
42561
42599
  mailtoComponent.error = mailtoComponent.error || "URI mailto has malformed header fields.";
42562
42600
  continue;
42563
42601
  }
42564
- const name = eqIdx === -1 ? token : token.slice(0, eqIdx);
42602
+ const name = decodeHex(eqIdx === -1 ? token : token.slice(0, eqIdx));
42603
+ const normalizedName = name.toLowerCase();
42565
42604
  const value = eqIdx === -1 ? "" : token.slice(eqIdx + 1);
42566
- if (name === "to") {
42605
+ if (normalizedName === "to") {
42567
42606
  const addrs = value.split(",");
42568
42607
  for (let j = 0;j < addrs.length; j++)
42569
42608
  to.push(addrs[j]);
42570
42609
  continue;
42571
42610
  }
42572
- if (name === "subject") {
42611
+ if (normalizedName === "subject") {
42573
42612
  mailtoComponent.subject = decodeHex(value);
42574
42613
  continue;
42575
42614
  }
42576
- if (name === "body") {
42615
+ if (normalizedName === "body") {
42577
42616
  mailtoComponent.body = decodeHex(value);
42578
42617
  continue;
42579
42618
  }
42580
42619
  if (headers === null)
42581
42620
  headers = Object.create(null);
42582
- headers[decodeHex(name)] = decodeHex(value);
42621
+ headers[name] = decodeHex(value);
42583
42622
  }
42584
42623
  if (headers !== null)
42585
42624
  mailtoComponent.headers = headers;
42586
42625
  }
42587
42626
  mailtoComponent.query = undefined;
42627
+ if (to.length > 0 && (mailtoComponent.userinfo !== undefined || mailtoComponent.host !== undefined || mailtoComponent.port !== undefined)) {
42628
+ mailtoComponent.userinfo = undefined;
42629
+ mailtoComponent.host = undefined;
42630
+ mailtoComponent.port = undefined;
42631
+ mailtoComponent.error = mailtoComponent.error || MAILTO_AUTHORITY_ERROR;
42632
+ }
42588
42633
  for (let i = 0;i < to.length; i++) {
42589
42634
  const rawAddr = to[i];
42590
42635
  const atIdx = rawAddr.lastIndexOf("@");
@@ -42601,10 +42646,35 @@ var require_schemes = __commonJS(function(exports, module) {
42601
42646
  mailtoComponent.to = to;
42602
42647
  return mailtoComponent;
42603
42648
  }
42649
+ function mailtoEncodeHeaderName(name) {
42650
+ return encodeWithAllow(name.replace(/%/gu, "%25"), HFNAME);
42651
+ }
42604
42652
  function mailtoSerialize(component, options) {
42605
42653
  const mailtoComponent = component;
42606
42654
  const to = Array.isArray(mailtoComponent.to) ? mailtoComponent.to.slice() : [];
42655
+ const sourceHeaders = mailtoComponent.headers && typeof mailtoComponent.headers === "object" ? Object.assign(Object.create(null), mailtoComponent.headers) : Object.create(null);
42656
+ const headers = Object.create(null);
42657
+ for (const name in sourceHeaders) {
42658
+ const normalizedName = name.toLowerCase();
42659
+ if (normalizedName === "to") {
42660
+ const addrs = String(sourceHeaders[name]).split(",");
42661
+ for (let i = 0;i < addrs.length; i++)
42662
+ to.push(addrs[i]);
42663
+ } else if (normalizedName === "subject" || normalizedName === "body") {
42664
+ headers[normalizedName] = sourceHeaders[name];
42665
+ } else {
42666
+ headers[name] = sourceHeaders[name];
42667
+ }
42668
+ }
42669
+ if (mailtoComponent.subject !== undefined)
42670
+ headers.subject = mailtoComponent.subject;
42671
+ if (mailtoComponent.body !== undefined)
42672
+ headers.body = mailtoComponent.body;
42673
+ mailtoComponent.headers = headers;
42607
42674
  if (to.length) {
42675
+ mailtoComponent.userinfo = undefined;
42676
+ mailtoComponent.host = undefined;
42677
+ mailtoComponent.port = undefined;
42608
42678
  for (let i = 0;i < to.length; i++) {
42609
42679
  const addr = String(to[i]);
42610
42680
  const atIdx = addr.lastIndexOf("@");
@@ -42620,18 +42690,12 @@ var require_schemes = __commonJS(function(exports, module) {
42620
42690
  } else {
42621
42691
  mailtoComponent.path = undefined;
42622
42692
  }
42623
- const headers = mailtoComponent.headers && typeof mailtoComponent.headers === "object" ? Object.assign(Object.create(null), mailtoComponent.headers) : Object.create(null);
42624
- if (mailtoComponent.subject)
42625
- headers.subject = mailtoComponent.subject;
42626
- if (mailtoComponent.body)
42627
- headers.body = mailtoComponent.body;
42628
- mailtoComponent.headers = headers;
42629
42693
  let query = "";
42630
42694
  let count = 0;
42631
42695
  for (const name in headers) {
42632
42696
  if (count++ !== 0)
42633
42697
  query += "&";
42634
- query += encodeWithAllow(name, HFNAME) + "=" + encodeWithAllow(String(headers[name]), HFNAME);
42698
+ query += mailtoEncodeHeaderName(name) + "=" + encodeWithAllow(String(headers[name]), HFNAME);
42635
42699
  }
42636
42700
  if (count !== 0) {
42637
42701
  mailtoComponent.query = query;
@@ -42724,6 +42788,12 @@ var require_fast_uri = __commonJS(function(exports, module) {
42724
42788
  schemelessOptions.skipEscape = true;
42725
42789
  return serialize(resolved, schemelessOptions);
42726
42790
  }
42791
+ function copyHost(target, source) {
42792
+ target.host = source.host;
42793
+ if (source.ipv6Zone !== undefined) {
42794
+ target.ipv6Zone = source.ipv6Zone;
42795
+ }
42796
+ }
42727
42797
  function resolveComponent(base, relative, options, skipNormalization) {
42728
42798
  const target = {};
42729
42799
  if (!skipNormalization) {
@@ -42734,14 +42804,14 @@ var require_fast_uri = __commonJS(function(exports, module) {
42734
42804
  if (!options.tolerant && relative.scheme) {
42735
42805
  target.scheme = relative.scheme;
42736
42806
  target.userinfo = relative.userinfo;
42737
- target.host = relative.host;
42807
+ copyHost(target, relative);
42738
42808
  target.port = relative.port;
42739
42809
  target.path = removeDotSegments(relative.path || "");
42740
42810
  target.query = relative.query;
42741
42811
  } else {
42742
42812
  if (relative.userinfo !== undefined || relative.host !== undefined || relative.port !== undefined) {
42743
42813
  target.userinfo = relative.userinfo;
42744
- target.host = relative.host;
42814
+ copyHost(target, relative);
42745
42815
  target.port = relative.port;
42746
42816
  target.path = removeDotSegments(relative.path || "");
42747
42817
  target.query = relative.query;
@@ -42769,7 +42839,7 @@ var require_fast_uri = __commonJS(function(exports, module) {
42769
42839
  target.query = relative.query;
42770
42840
  }
42771
42841
  target.userinfo = base.userinfo;
42772
- target.host = base.host;
42842
+ copyHost(target, base);
42773
42843
  target.port = base.port;
42774
42844
  }
42775
42845
  target.scheme = base.scheme;
@@ -42785,6 +42855,7 @@ var require_fast_uri = __commonJS(function(exports, module) {
42785
42855
  function serialize(cmpts, opts) {
42786
42856
  const component = {
42787
42857
  host: cmpts.host,
42858
+ ipv6Zone: cmpts.ipv6Zone,
42788
42859
  scheme: cmpts.scheme,
42789
42860
  userinfo: cmpts.userinfo,
42790
42861
  port: cmpts.port,
@@ -42981,6 +43052,9 @@ var require_fast_uri = __commonJS(function(exports, module) {
42981
43052
  isIP = ipv6result.isIPV6 || ipv6result.isIPVFuture === true;
42982
43053
  malformedIPLiteral = hasIPLiteralBracket && (!bracketedIPLiteral || ipv6result.error === true);
42983
43054
  parsed.host = isIP ? ipv6result.host : ipv6result.host.toLowerCase();
43055
+ if (isIP && ipv6result.isIPV6 === true && parsed.host.indexOf("%") !== -1) {
43056
+ parsed.ipv6Zone = parsed.host.slice(parsed.host.indexOf("%") + 1);
43057
+ }
42984
43058
  if (malformedIPLiteral) {
42985
43059
  parsed.error = parsed.error || "URI host is malformed.";
42986
43060
  malformedAuthorityOrPort = true;
@@ -43005,13 +43079,14 @@ var require_fast_uri = __commonJS(function(exports, module) {
43005
43079
  if (!malformedIPLiteral) {
43006
43080
  malformedHost = canonicalizeHost(parsed, options, schemeHandler, isIP);
43007
43081
  }
43008
- if (!schemeHandler || schemeHandler && !schemeHandler.skipNormalize) {
43009
- if (uri.indexOf("%") !== -1) {
43010
- if (parsed.host !== undefined && !malformedIPLiteral) {
43011
- const host = isIP ? parsed.host : normalizePercentEncoding(parsed.host, true);
43012
- parsed.host = reescapeHostDelimiters(host, isIP);
43013
- }
43082
+ if (uri.indexOf("%") !== -1 && parsed.host !== undefined && !malformedIPLiteral) {
43083
+ let host = isIP ? parsed.host : normalizePercentEncoding(parsed.host, true);
43084
+ if (!isIP) {
43085
+ host = normalizePercentEncoding(host.toLowerCase());
43014
43086
  }
43087
+ parsed.host = reescapeHostDelimiters(host, isIP);
43088
+ }
43089
+ if (!schemeHandler || schemeHandler && !schemeHandler.skipNormalize) {
43015
43090
  if (parsed.path) {
43016
43091
  parsed.path = normalizePathEncoding(parsed.path);
43017
43092
  }
@@ -52769,9 +52844,9 @@ var init_config = __esm(() => {
52769
52844
  SDK_METADATA = {
52770
52845
  language: "typescript",
52771
52846
  openapiDocVersion: "1.0.0",
52772
- sdkVersion: "0.3.1",
52773
- genVersion: "2.937.18",
52774
- userAgent: "speakeasy-sdk/mcp-typescript 0.3.1 2.937.18 1.0.0 @lightsage/mcp"
52847
+ sdkVersion: "0.4.1",
52848
+ genVersion: "2.941.0",
52849
+ userAgent: "speakeasy-sdk/mcp-typescript 0.4.1 2.941.0 1.0.0 @lightsage/mcp"
52775
52850
  };
52776
52851
  });
52777
52852
 
@@ -54953,68 +55028,95 @@ Returns reusable environments without exposing secret values.`,
54953
55028
  };
54954
55029
  });
54955
55030
 
54956
- // src/models/v2evalresultsummary.ts
54957
- var V2EvalResultSummary$zodSchema;
54958
- var init_v2evalresultsummary = __esm(() => {
55031
+ // src/models/publicevalrunprogress.ts
55032
+ var PublicEvalRunProgress$zodSchema;
55033
+ var init_publicevalrunprogress = __esm(() => {
54959
55034
  init_zod();
54960
- V2EvalResultSummary$zodSchema = object({
54961
- agent: string2().nullable().optional().describe("Runnable agent combining the harness and model."),
55035
+ PublicEvalRunProgress$zodSchema = object({
55036
+ cancelled: int().default(0),
55037
+ completed: int().default(0),
55038
+ failed: int().default(0),
55039
+ interrupted: int().default(0).describe("Deprecated compatibility alias for `retrying`; use `retrying`."),
55040
+ percent: number2().default(0).describe("The percentage of child jobs in a final state (`completed`, `failed`, or `cancelled`). It can reach `100` while the parent job is still summarizing, so use `status` to detect completion."),
55041
+ queued: int().default(0),
55042
+ retrying: int().default(0).describe("Recoverable child jobs waiting to be retried. These remain active and are excluded from `percent`."),
55043
+ running: int().default(0),
55044
+ total: int().default(0)
55045
+ });
55046
+ });
55047
+
55048
+ // src/models/v2evalrun.ts
55049
+ var V2EvalRun$zodSchema;
55050
+ var init_v2evalrun = __esm(() => {
55051
+ init_zod();
55052
+ init_publicevalrunprogress();
55053
+ V2EvalRun$zodSchema = object({
55054
+ agent: array(string2()).optional().describe("Runnable agents selected for this simulation."),
55055
+ attempt_count: int().default(1).describe("Total concrete attempts after all fan-out dimensions."),
55056
+ completed_at: string2().nullable().optional(),
55057
+ configuration_id: string2().nullable().optional(),
54962
55058
  created_at: string2().nullable().optional(),
54963
- eval_id: string2().nullable().optional(),
54964
- failure_reasons: array(string2()).nullable().optional(),
54965
- id: string2().describe("Result ID, not an eval job ID."),
54966
- latency_ms: int().nullable().optional(),
54967
- passed: boolean2().nullable().optional(),
54968
- prompt: string2().nullable().optional(),
54969
- prompt_index: int().default(0),
54970
- repeat_index: int().default(0),
54971
- score: number2().nullable().optional().describe("Score from 0 to 1.")
55059
+ dispatch_id: string2().nullable().optional(),
55060
+ error: string2().nullable().optional(),
55061
+ estimated_cost: number2().nullable().optional(),
55062
+ eval_id: string2().nullable().optional().describe("Saved eval definition ID, including one created by a direct run."),
55063
+ eval_name: string2().nullable().optional().describe("Display name captured for the eval when the run was created."),
55064
+ id: string2().describe("Run ID to use when polling eval-run status."),
55065
+ job_id: string2().describe("Alias of `id` for the durable parent job."),
55066
+ persona: string2().nullable().optional().describe("Perspective applied to every attempt in this simulation."),
55067
+ progress: PublicEvalRunProgress$zodSchema.optional(),
55068
+ prompt_count: int().default(1).describe("Number of prompts included in the simulation."),
55069
+ runs: int().default(1).describe("Independent attempts requested for every prompt-agent pair."),
55070
+ stage: string2().nullable().optional(),
55071
+ started_at: string2().nullable().optional(),
55072
+ status: string2().describe("Terminal values: `completed`, `failed`, and `cancelled`. Recoverable interruptions are reported as `running` with stage `retrying`.")
54972
55073
  });
54973
55074
  });
54974
55075
 
54975
- // src/models/v2evalresultsresponse.ts
54976
- var V2EvalResultsResponse$zodSchema;
54977
- var init_v2evalresultsresponse = __esm(() => {
55076
+ // src/models/v2evalrunsresponse.ts
55077
+ var V2EvalRunsResponse$zodSchema;
55078
+ var init_v2evalrunsresponse = __esm(() => {
54978
55079
  init_zod();
54979
- init_v2evalresultsummary();
54980
- V2EvalResultsResponse$zodSchema = object({
55080
+ init_v2evalrun();
55081
+ V2EvalRunsResponse$zodSchema = object({
54981
55082
  count: int(),
54982
- data: array(V2EvalResultSummary$zodSchema).optional(),
54983
- limit: int()
55083
+ data: array(V2EvalRun$zodSchema).optional(),
55084
+ limit: int(),
55085
+ offset: int()
54984
55086
  });
54985
55087
  });
54986
55088
 
54987
- // src/models/v2evalresultslistop.ts
54988
- var V2EvalResultsListRequest$zodSchema, V2EvalResultsListResponse$zodSchema;
54989
- var init_v2evalresultslistop = __esm(() => {
55089
+ // src/models/v2evalrunslistop.ts
55090
+ var V2EvalRunsListRequest$zodSchema, V2EvalRunsListResponse$zodSchema;
55091
+ var init_v2evalrunslistop = __esm(() => {
54990
55092
  init_zod();
54991
55093
  init_httpvalidationerror();
54992
- init_v2evalresultsresponse();
54993
- V2EvalResultsListRequest$zodSchema = object({
54994
- limit: int().default(20),
54995
- since: string2().describe("Relative window such as `7d` or an ISO timestamp.").optional()
55094
+ init_v2evalrunsresponse();
55095
+ V2EvalRunsListRequest$zodSchema = object({
55096
+ limit: int().default(25),
55097
+ offset: int().default(0)
54996
55098
  });
54997
- V2EvalResultsListResponse$zodSchema = union([
54998
- V2EvalResultsResponse$zodSchema,
55099
+ V2EvalRunsListResponse$zodSchema = union([
55100
+ V2EvalRunsResponse$zodSchema,
54999
55101
  HTTPValidationError$zodSchema
55000
55102
  ]);
55001
55103
  });
55002
55104
 
55003
- // src/funcs/evalResultsV2EvalResultsList.ts
55004
- function evalResultsV2EvalResultsList(client$, request, options) {
55105
+ // src/funcs/evalRunsV2EvalRunsList.ts
55106
+ function evalRunsV2EvalRunsList(client$, request, options) {
55005
55107
  return new APIPromise($do7(client$, request, options));
55006
55108
  }
55007
55109
  async function $do7(client$, request, options) {
55008
- const parsed$ = safeParse4(request, (value$) => V2EvalResultsListRequest$zodSchema.optional().parse(value$), "Input validation failed");
55110
+ const parsed$ = safeParse4(request, (value$) => V2EvalRunsListRequest$zodSchema.optional().parse(value$), "Input validation failed");
55009
55111
  if (!parsed$.ok) {
55010
55112
  return [parsed$, { status: "invalid" }];
55011
55113
  }
55012
55114
  const payload$ = parsed$.value;
55013
55115
  const body$ = null;
55014
- const path$ = pathToFunc("/v2/eval-results")();
55116
+ const path$ = pathToFunc("/v2/eval-runs")();
55015
55117
  const query$ = encodeFormQuery({
55016
55118
  limit: payload$?.limit,
55017
- since: payload$?.since
55119
+ offset: payload$?.offset
55018
55120
  });
55019
55121
  const headers$ = new Headers(compactMap({
55020
55122
  Accept: "application/json"
@@ -55024,7 +55126,7 @@ async function $do7(client$, request, options) {
55024
55126
  const context = {
55025
55127
  options: client$._options,
55026
55128
  baseURL: options?.serverURL ?? client$._baseURL ?? "",
55027
- operationID: "v2_eval_results_list",
55129
+ operationID: "v2_eval_runs_list",
55028
55130
  oAuth2Scopes: null,
55029
55131
  resolvedSecurity: requestSecurity,
55030
55132
  securitySource: client$._options.security,
@@ -55067,30 +55169,30 @@ async function $do7(client$, request, options) {
55067
55169
  response: doResult.value
55068
55170
  }];
55069
55171
  }
55070
- var init_evalResultsV2EvalResultsList = __esm(() => {
55172
+ var init_evalRunsV2EvalRunsList = __esm(() => {
55071
55173
  init_encodings();
55072
55174
  init_primitives();
55073
55175
  init_schemas4();
55074
55176
  init_security();
55075
55177
  init_url();
55076
- init_v2evalresultslistop();
55178
+ init_v2evalrunslistop();
55077
55179
  init_async();
55078
55180
  });
55079
55181
 
55080
- // src/mcp-server/tools/evalResultsV2EvalResultsList.ts
55081
- var args4, tool$evalResultsV2EvalResultsList;
55082
- var init_evalResultsV2EvalResultsList2 = __esm(() => {
55083
- init_evalResultsV2EvalResultsList();
55084
- init_v2evalresultslistop();
55182
+ // src/mcp-server/tools/evalRunsV2EvalRunsList.ts
55183
+ var args4, tool$evalRunsV2EvalRunsList;
55184
+ var init_evalRunsV2EvalRunsList2 = __esm(() => {
55185
+ init_evalRunsV2EvalRunsList();
55186
+ init_v2evalrunslistop();
55085
55187
  init_tools();
55086
55188
  args4 = {
55087
- request: V2EvalResultsListRequest$zodSchema.optional()
55189
+ request: V2EvalRunsListRequest$zodSchema.optional()
55088
55190
  };
55089
- tool$evalResultsV2EvalResultsList = {
55090
- name: "eval-results-v2-eval-results-list",
55091
- description: `List Results
55191
+ tool$evalRunsV2EvalRunsList = {
55192
+ name: "eval-runs-v2-eval-runs-list",
55193
+ description: `List Runs
55092
55194
 
55093
- Returns completed custom-eval results from all supported run flows, ordered newest first. Result IDs are different from eval run IDs.`,
55195
+ Returns jobs created through \`POST /v2/eval-runs\`, ordered newest first, with lifecycle state and progress.`,
55094
55196
  annotations: {
55095
55197
  title: "",
55096
55198
  destructiveHint: false,
@@ -55100,7 +55202,7 @@ Returns completed custom-eval results from all supported run flows, ordered newe
55100
55202
  },
55101
55203
  args: args4,
55102
55204
  tool: async (client, args, ctx) => {
55103
- const [result] = await evalResultsV2EvalResultsList(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
55205
+ const [result] = await evalRunsV2EvalRunsList(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
55104
55206
  if (!result.ok) {
55105
55207
  return {
55106
55208
  content: [{ type: "text", text: result.error.message }],
@@ -55112,63 +55214,51 @@ Returns completed custom-eval results from all supported run flows, ordered newe
55112
55214
  };
55113
55215
  });
55114
55216
 
55115
- // src/models/v2evalresult.ts
55116
- var V2EvalResult$zodSchema;
55117
- var init_v2evalresult = __esm(() => {
55118
- init_zod();
55119
- V2EvalResult$zodSchema = object({
55120
- agent: string2().nullable().optional().describe("Runnable agent combining the harness and model."),
55121
- created_at: string2().nullable().optional(),
55122
- eval_id: string2().nullable().optional(),
55123
- execution_stderr: string2().nullable().optional(),
55124
- execution_stdout: string2().nullable().optional(),
55125
- exit_code: int().nullable().optional(),
55126
- failure_reasons: array(string2()).nullable().optional(),
55127
- id: string2().describe("Result ID, not an eval job ID."),
55128
- latency_ms: int().nullable().optional(),
55129
- output: string2().nullable().optional(),
55130
- passed: boolean2().nullable().optional(),
55131
- prompt: string2().nullable().optional(),
55132
- prompt_index: int().default(0),
55133
- repeat_index: int().default(0),
55134
- score: number2().nullable().optional().describe("Score from 0 to 1."),
55135
- tool_calls: array(record(string2(), any())).nullable().optional()
55136
- });
55137
- });
55138
-
55139
- // src/models/v2evalresultsretrieveop.ts
55140
- var V2EvalResultsRetrieveRequest$zodSchema, V2EvalResultsRetrieveResponse$zodSchema;
55141
- var init_v2evalresultsretrieveop = __esm(() => {
55217
+ // src/models/v2evalrunscancelop.ts
55218
+ var V2EvalRunsCancelRequest$zodSchema, V2EvalRunsCancelServiceUnavailableResponseBody$zodSchema, V2EvalRunsCancelConflictResponseBody$zodSchema, V2EvalRunsCancelNotFoundResponseBody$zodSchema, V2EvalRunsCancelResponse$zodSchema;
55219
+ var init_v2evalrunscancelop = __esm(() => {
55142
55220
  init_zod();
55143
55221
  init_httpvalidationerror();
55144
- init_v2evalresult();
55145
- V2EvalResultsRetrieveRequest$zodSchema = object({
55146
- result_id: string2()
55222
+ init_v2evalrun();
55223
+ V2EvalRunsCancelRequest$zodSchema = object({
55224
+ run_id: string2()
55147
55225
  });
55148
- V2EvalResultsRetrieveResponse$zodSchema = union([
55149
- V2EvalResult$zodSchema,
55226
+ V2EvalRunsCancelServiceUnavailableResponseBody$zodSchema = object({
55227
+ detail: string2()
55228
+ }).describe("Eval run cancellation is temporarily unavailable.");
55229
+ V2EvalRunsCancelConflictResponseBody$zodSchema = object({
55230
+ detail: string2()
55231
+ }).describe("The eval run has already completed or failed.");
55232
+ V2EvalRunsCancelNotFoundResponseBody$zodSchema = object({
55233
+ detail: string2()
55234
+ }).describe("Eval run not found. Use the run ID returned when you started the eval.");
55235
+ V2EvalRunsCancelResponse$zodSchema = union([
55236
+ V2EvalRun$zodSchema,
55237
+ lazy(() => V2EvalRunsCancelNotFoundResponseBody$zodSchema),
55238
+ lazy(() => V2EvalRunsCancelConflictResponseBody$zodSchema),
55239
+ lazy(() => V2EvalRunsCancelServiceUnavailableResponseBody$zodSchema),
55150
55240
  HTTPValidationError$zodSchema
55151
55241
  ]);
55152
55242
  });
55153
55243
 
55154
- // src/funcs/evalResultsV2EvalResultsRetrieve.ts
55155
- function evalResultsV2EvalResultsRetrieve(client$, request, options) {
55244
+ // src/funcs/evalsCancel.ts
55245
+ function evalsCancel(client$, request, options) {
55156
55246
  return new APIPromise($do8(client$, request, options));
55157
55247
  }
55158
55248
  async function $do8(client$, request, options) {
55159
- const parsed$ = safeParse4(request, (value$) => V2EvalResultsRetrieveRequest$zodSchema.parse(value$), "Input validation failed");
55249
+ const parsed$ = safeParse4(request, (value$) => V2EvalRunsCancelRequest$zodSchema.parse(value$), "Input validation failed");
55160
55250
  if (!parsed$.ok) {
55161
55251
  return [parsed$, { status: "invalid" }];
55162
55252
  }
55163
55253
  const payload$ = parsed$.value;
55164
55254
  const body$ = null;
55165
55255
  const pathParams$ = {
55166
- result_id: encodeSimple("result_id", payload$.result_id, {
55256
+ run_id: encodeSimple("run_id", payload$.run_id, {
55167
55257
  explode: false,
55168
55258
  charEncoding: "percent"
55169
55259
  })
55170
55260
  };
55171
- const path$ = pathToFunc("/v2/eval-results/{result_id}")(pathParams$);
55261
+ const path$ = pathToFunc("/v2/eval-runs/{run_id}/cancel")(pathParams$);
55172
55262
  const headers$ = new Headers(compactMap({
55173
55263
  Accept: "application/json"
55174
55264
  }));
@@ -55177,7 +55267,7 @@ async function $do8(client$, request, options) {
55177
55267
  const context = {
55178
55268
  options: client$._options,
55179
55269
  baseURL: options?.serverURL ?? client$._baseURL ?? "",
55180
- operationID: "v2_eval_results_retrieve",
55270
+ operationID: "v2_eval_runs_cancel",
55181
55271
  oAuth2Scopes: null,
55182
55272
  resolvedSecurity: requestSecurity,
55183
55273
  securitySource: client$._options.security,
@@ -55192,7 +55282,7 @@ async function $do8(client$, request, options) {
55192
55282
  };
55193
55283
  const requestRes = client$._createRequest(context, {
55194
55284
  security: requestSecurity,
55195
- method: "GET",
55285
+ method: "POST",
55196
55286
  baseURL: options?.serverURL,
55197
55287
  path: path$,
55198
55288
  headers: headers$,
@@ -55219,40 +55309,40 @@ async function $do8(client$, request, options) {
55219
55309
  response: doResult.value
55220
55310
  }];
55221
55311
  }
55222
- var init_evalResultsV2EvalResultsRetrieve = __esm(() => {
55312
+ var init_evalsCancel = __esm(() => {
55223
55313
  init_encodings();
55224
55314
  init_primitives();
55225
55315
  init_schemas4();
55226
55316
  init_security();
55227
55317
  init_url();
55228
- init_v2evalresultsretrieveop();
55318
+ init_v2evalrunscancelop();
55229
55319
  init_async();
55230
55320
  });
55231
55321
 
55232
- // src/mcp-server/tools/evalResultsV2EvalResultsRetrieve.ts
55233
- var args5, tool$evalResultsV2EvalResultsRetrieve;
55234
- var init_evalResultsV2EvalResultsRetrieve2 = __esm(() => {
55235
- init_evalResultsV2EvalResultsRetrieve();
55236
- init_v2evalresultsretrieveop();
55322
+ // src/mcp-server/tools/evalsCancel.ts
55323
+ var args5, tool$evalsCancel;
55324
+ var init_evalsCancel2 = __esm(() => {
55325
+ init_evalsCancel();
55326
+ init_v2evalrunscancelop();
55237
55327
  init_tools();
55238
55328
  args5 = {
55239
- request: V2EvalResultsRetrieveRequest$zodSchema
55329
+ request: V2EvalRunsCancelRequest$zodSchema
55240
55330
  };
55241
- tool$evalResultsV2EvalResultsRetrieve = {
55242
- name: "eval-results-v2-eval-results-retrieve",
55243
- description: `Retrieve Result
55331
+ tool$evalsCancel = {
55332
+ name: "evals-cancel",
55333
+ description: `Cancel an eval run
55244
55334
 
55245
- Returns one completed custom-eval result with its output and diagnostic details.`,
55335
+ Marks an active eval run as cancelled and prevents queued attempts from starting. Running sandbox attempts are stopped asynchronously. Requests already sent to an external provider cannot be recalled, but any later response does not change the run from cancelled. Repeating this request for an already-cancelled run is safe.`,
55246
55336
  annotations: {
55247
55337
  title: "",
55248
55338
  destructiveHint: false,
55249
55339
  idempotentHint: false,
55250
55340
  openWorldHint: false,
55251
- readOnlyHint: true
55341
+ readOnlyHint: false
55252
55342
  },
55253
55343
  args: args5,
55254
55344
  tool: async (client, args, ctx) => {
55255
- const [result] = await evalResultsV2EvalResultsRetrieve(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
55345
+ const [result] = await evalsCancel(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
55256
55346
  if (!result.ok) {
55257
55347
  return {
55258
55348
  content: [{ type: "text", text: result.error.message }],
@@ -55264,29 +55354,124 @@ Returns one completed custom-eval result with its output and diagnostic details.
55264
55354
  };
55265
55355
  });
55266
55356
 
55267
- // src/models/publicevalrunprogress.ts
55268
- var PublicEvalRunProgress$zodSchema;
55269
- var init_publicevalrunprogress = __esm(() => {
55357
+ // src/models/publicevalrunenvironment.ts
55358
+ var PublicEvalRunEnvironment$zodSchema;
55359
+ var init_publicevalrunenvironment = __esm(() => {
55270
55360
  init_zod();
55271
- PublicEvalRunProgress$zodSchema = object({
55272
- cancelled: int().default(0),
55273
- completed: int().default(0),
55274
- failed: int().default(0),
55275
- interrupted: int().default(0),
55276
- percent: number2().default(0).describe("The percentage of child jobs in a terminal state. It can reach `100` while the parent job is still summarizing, so use `status` to detect completion."),
55277
- queued: int().default(0),
55278
- running: int().default(0),
55279
- total: int().default(0)
55361
+ PublicEvalRunEnvironment$zodSchema = object({
55362
+ changed_since_queued: boolean2().nullable().optional().describe("True when the saved environment has been edited or deleted since the run was queued, so a rerun would use different values. Null when unknown or for inline environments."),
55363
+ id: string2().nullable().optional().describe("Environment ID, as listed by the environments operation."),
55364
+ inline: boolean2().default(false).describe("True when variables were supplied inline with the run request instead of from a saved environment."),
55365
+ name: string2().nullable().optional().describe("Environment name when the run was queued."),
55366
+ queued_at: string2().nullable().optional().describe("When the run was queued with this environment."),
55367
+ updated_at: string2().nullable().optional().describe("Environment version the run used: when its name or variables last changed before the run was queued. Null if the environment had not changed since versioning was introduced."),
55368
+ variable_names: array(string2()).optional().describe("Variable names supplied to the run. Values are never returned.")
55369
+ }).describe("Non-secret summary of the environment an eval run was queued with.");
55370
+ });
55371
+
55372
+ // src/models/publiccriterionverdict.ts
55373
+ var Verdict$zodSchema, PublicCriterionVerdict$zodSchema;
55374
+ var init_publiccriterionverdict = __esm(() => {
55375
+ init_zod();
55376
+ Verdict$zodSchema = _enum([
55377
+ "PASS",
55378
+ "FAIL",
55379
+ "NOT_VERIFIED"
55380
+ ]).describe("`PASS` when the evidence shows the requirement was met, `FAIL` when it shows it was not, and `NOT_VERIFIED` when the evidence cannot confirm it. Any `FAIL` or `NOT_VERIFIED` fails the run, whatever the score.");
55381
+ PublicCriterionVerdict$zodSchema = object({
55382
+ criterion: string2().describe("One mandatory requirement from the judge criteria."),
55383
+ evidence: string2().nullable().optional().describe("Short evidence the judge gave for the verdict."),
55384
+ verdict: Verdict$zodSchema.describe("`PASS` when the evidence shows the requirement was met, `FAIL` when it shows it was not, and `NOT_VERIFIED` when the evidence cannot confirm it. Any `FAIL` or `NOT_VERIFIED` fails the run, whatever the score.")
55280
55385
  });
55281
55386
  });
55282
55387
 
55283
- // src/models/v2evalrun.ts
55284
- var V2EvalRun$zodSchema;
55285
- var init_v2evalrun = __esm(() => {
55388
+ // src/models/publicgraderresult.ts
55389
+ var PublicGraderResult$zodSchema;
55390
+ var init_publicgraderresult = __esm(() => {
55391
+ init_zod();
55392
+ init_publiccriterionverdict();
55393
+ PublicGraderResult$zodSchema = object({
55394
+ criteria_verdicts: array(PublicCriterionVerdict$zodSchema).optional().describe("Judges only: a verdict for each mandatory requirement in the judge criteria. Empty for other checks and for runs graded before per-requirement verdicts were recorded."),
55395
+ deductions: array(string2()).optional().describe("Specific shortcomings the check recorded, even when it passed. A judge can pass with a score below 1 and still list deductions here; those checks do not appear in `failure_reasons`."),
55396
+ details: string2().nullable().optional().describe("Reasoning or check output explaining the verdict."),
55397
+ label: string2().nullable().optional().describe("Display name for the check, when the grader sets one."),
55398
+ name: string2().describe("Grader name as configured on the eval."),
55399
+ passed: boolean2().nullable().optional().describe("Whether this check passed."),
55400
+ score: number2().nullable().optional().describe("Score from 0 to 1 for checks that score partially, such as judges; null for pass/fail checks.")
55401
+ });
55402
+ });
55403
+
55404
+ // src/models/searchgroundingmode.ts
55405
+ var SearchGroundingMode$zodSchema;
55406
+ var init_searchgroundingmode = __esm(() => {
55407
+ init_zod();
55408
+ SearchGroundingMode$zodSchema = _enum([
55409
+ "disabled",
55410
+ "enabled",
55411
+ "default"
55412
+ ]).describe("Requested web-retrieval behavior for an eval run.");
55413
+ });
55414
+
55415
+ // src/models/v2evalresultsummary.ts
55416
+ var V2EvalResultSummary$zodSchema;
55417
+ var init_v2evalresultsummary = __esm(() => {
55418
+ init_zod();
55419
+ init_publicevalrunenvironment();
55420
+ init_publicgraderresult();
55421
+ init_searchgroundingmode();
55422
+ V2EvalResultSummary$zodSchema = object({
55423
+ agent: string2().nullable().optional().describe("Runnable agent combining the harness and model."),
55424
+ created_at: string2().nullable().optional(),
55425
+ environment: PublicEvalRunEnvironment$zodSchema.nullable().optional().describe("The environment this attempt used, without secret values. Null when the attempt had no environment."),
55426
+ eval_id: string2().nullable().optional(),
55427
+ failure_reasons: array(string2()).nullable().optional().describe("Checks that failed, listed even when the result passed overall. Deductions from checks that passed are in `grader_results[].deductions`, not here."),
55428
+ grader_results: array(PublicGraderResult$zodSchema).optional().describe("Per-check verdicts behind `score`, including deductions recorded by passing checks."),
55429
+ grounding_qualification_reason: string2().nullable().optional(),
55430
+ grounding_qualified: boolean2().nullable().optional().describe("Whether the attempt satisfied the requested grounding experiment. Null means the default mode was used or isolation could not be proven."),
55431
+ id: string2().describe("Result ID, not an eval job ID."),
55432
+ latency_ms: int().nullable().optional(),
55433
+ passed: boolean2().nullable().optional(),
55434
+ prompt: string2().nullable().optional(),
55435
+ prompt_index: int().default(0),
55436
+ repeat_index: int().default(0),
55437
+ score: number2().nullable().optional().describe("Score from 0 to 1."),
55438
+ search_grounding: SearchGroundingMode$zodSchema.nullable().optional(),
55439
+ search_observed: boolean2().nullable().optional(),
55440
+ web_retrieval_observed: boolean2().nullable().optional()
55441
+ });
55442
+ });
55443
+
55444
+ // src/models/v2evalrunanalysis.ts
55445
+ var V2EvalRunAnalysis$zodSchema;
55446
+ var init_v2evalrunanalysis = __esm(() => {
55447
+ init_zod();
55448
+ V2EvalRunAnalysis$zodSchema = object({
55449
+ agent: string2().nullable().optional().describe("Runnable agent used for this concrete attempt."),
55450
+ analysis: record(string2(), any()).optional().describe("Persisted eval_run_analysis payload, including verdict, outcome, findings, bottleneck, and analysis sections."),
55451
+ analysis_version: string2().nullable().optional().describe("Version of the analysis pipeline that produced this payload."),
55452
+ attempt_id: string2().describe("Child eval-attempt ID used to correlate the analysis with trace events."),
55453
+ created_at: string2().nullable().optional(),
55454
+ eval_id: string2().nullable().optional().describe("Public eval definition ID, not the backing executable eval ID."),
55455
+ id: string2().describe("Persisted eval analysis ID."),
55456
+ prompt: string2().nullable().optional().describe("Exact task prompt used for this concrete attempt."),
55457
+ prompt_index: int().default(0).describe("Zero-based prompt index within the submitted simulation."),
55458
+ repeat_index: int().default(0).describe("Zero-based repeat index within the submitted run."),
55459
+ result_id: string2().describe("Concrete eval result ID analyzed for this attempt."),
55460
+ updated_at: string2().nullable().optional(),
55461
+ variant_key: string2().default("default").describe("Runtime variant that produced this result.")
55462
+ });
55463
+ });
55464
+
55465
+ // src/models/v2evalrundetail.ts
55466
+ var V2EvalRunDetail$zodSchema;
55467
+ var init_v2evalrundetail = __esm(() => {
55286
55468
  init_zod();
55287
55469
  init_publicevalrunprogress();
55288
- V2EvalRun$zodSchema = object({
55470
+ init_v2evalresultsummary();
55471
+ init_v2evalrunanalysis();
55472
+ V2EvalRunDetail$zodSchema = object({
55289
55473
  agent: array(string2()).optional().describe("Runnable agents selected for this simulation."),
55474
+ analysis: array(V2EvalRunAnalysis$zodSchema).optional().describe("Per-attempt analysis for this run. Empty until an attempt has been analyzed."),
55290
55475
  attempt_count: int().default(1).describe("Total concrete attempts after all fan-out dimensions."),
55291
55476
  completed_at: string2().nullable().optional(),
55292
55477
  configuration_id: string2().nullable().optional(),
@@ -55301,58 +55486,52 @@ var init_v2evalrun = __esm(() => {
55301
55486
  persona: string2().nullable().optional().describe("Perspective applied to every attempt in this simulation."),
55302
55487
  progress: PublicEvalRunProgress$zodSchema.optional(),
55303
55488
  prompt_count: int().default(1).describe("Number of prompts included in the simulation."),
55489
+ results: array(V2EvalResultSummary$zodSchema).optional().describe("Per-attempt outcomes for this run. Empty while the run is active; populated as attempts finish."),
55490
+ results_truncated: boolean2().default(false).describe("True when the run produced more results than the inline cap. The remaining results are not included in this response."),
55304
55491
  runs: int().default(1).describe("Independent attempts requested for every prompt-agent pair."),
55305
55492
  stage: string2().nullable().optional(),
55306
55493
  started_at: string2().nullable().optional(),
55307
- status: string2()
55494
+ status: string2().describe("Terminal values: `completed`, `failed`, and `cancelled`. Recoverable interruptions are reported as `running` with stage `retrying`.")
55308
55495
  });
55309
55496
  });
55310
55497
 
55311
- // src/models/v2evalrunsresponse.ts
55312
- var V2EvalRunsResponse$zodSchema;
55313
- var init_v2evalrunsresponse = __esm(() => {
55314
- init_zod();
55315
- init_v2evalrun();
55316
- V2EvalRunsResponse$zodSchema = object({
55317
- count: int(),
55318
- data: array(V2EvalRun$zodSchema).optional(),
55319
- limit: int(),
55320
- offset: int()
55321
- });
55322
- });
55323
-
55324
- // src/models/v2evalrunslistop.ts
55325
- var V2EvalRunsListRequest$zodSchema, V2EvalRunsListResponse$zodSchema;
55326
- var init_v2evalrunslistop = __esm(() => {
55498
+ // src/models/v2evalrunsretrieveop.ts
55499
+ var V2EvalRunsRetrieveRequest$zodSchema, V2EvalRunsRetrieveResponseBody$zodSchema, V2EvalRunsRetrieveResponse$zodSchema;
55500
+ var init_v2evalrunsretrieveop = __esm(() => {
55327
55501
  init_zod();
55328
55502
  init_httpvalidationerror();
55329
- init_v2evalrunsresponse();
55330
- V2EvalRunsListRequest$zodSchema = object({
55331
- limit: int().default(25),
55332
- offset: int().default(0)
55503
+ init_v2evalrundetail();
55504
+ V2EvalRunsRetrieveRequest$zodSchema = object({
55505
+ run_id: string2()
55333
55506
  });
55334
- V2EvalRunsListResponse$zodSchema = union([
55335
- V2EvalRunsResponse$zodSchema,
55507
+ V2EvalRunsRetrieveResponseBody$zodSchema = object({
55508
+ detail: string2()
55509
+ }).describe("Eval run not found. Use the run ID returned when you started the eval.");
55510
+ V2EvalRunsRetrieveResponse$zodSchema = union([
55511
+ V2EvalRunDetail$zodSchema,
55512
+ lazy(() => V2EvalRunsRetrieveResponseBody$zodSchema),
55336
55513
  HTTPValidationError$zodSchema
55337
55514
  ]);
55338
55515
  });
55339
55516
 
55340
- // src/funcs/evalRunsV2EvalRunsList.ts
55341
- function evalRunsV2EvalRunsList(client$, request, options) {
55517
+ // src/funcs/evalsGet.ts
55518
+ function evalsGet(client$, request, options) {
55342
55519
  return new APIPromise($do9(client$, request, options));
55343
55520
  }
55344
55521
  async function $do9(client$, request, options) {
55345
- const parsed$ = safeParse4(request, (value$) => V2EvalRunsListRequest$zodSchema.optional().parse(value$), "Input validation failed");
55522
+ const parsed$ = safeParse4(request, (value$) => V2EvalRunsRetrieveRequest$zodSchema.parse(value$), "Input validation failed");
55346
55523
  if (!parsed$.ok) {
55347
55524
  return [parsed$, { status: "invalid" }];
55348
55525
  }
55349
55526
  const payload$ = parsed$.value;
55350
55527
  const body$ = null;
55351
- const path$ = pathToFunc("/v2/eval-runs")();
55352
- const query$ = encodeFormQuery({
55353
- limit: payload$?.limit,
55354
- offset: payload$?.offset
55355
- });
55528
+ const pathParams$ = {
55529
+ run_id: encodeSimple("run_id", payload$.run_id, {
55530
+ explode: false,
55531
+ charEncoding: "percent"
55532
+ })
55533
+ };
55534
+ const path$ = pathToFunc("/v2/eval-runs/{run_id}")(pathParams$);
55356
55535
  const headers$ = new Headers(compactMap({
55357
55536
  Accept: "application/json"
55358
55537
  }));
@@ -55361,7 +55540,7 @@ async function $do9(client$, request, options) {
55361
55540
  const context = {
55362
55541
  options: client$._options,
55363
55542
  baseURL: options?.serverURL ?? client$._baseURL ?? "",
55364
- operationID: "v2_eval_runs_list",
55543
+ operationID: "v2_eval_runs_retrieve",
55365
55544
  oAuth2Scopes: null,
55366
55545
  resolvedSecurity: requestSecurity,
55367
55546
  securitySource: client$._options.security,
@@ -55380,7 +55559,6 @@ async function $do9(client$, request, options) {
55380
55559
  baseURL: options?.serverURL,
55381
55560
  path: path$,
55382
55561
  headers: headers$,
55383
- query: query$,
55384
55562
  body: body$,
55385
55563
  userAgent: client$._options.userAgent,
55386
55564
  timeoutMs: options?.timeoutMs || client$._options.timeoutMs || -1
@@ -55404,135 +55582,7 @@ async function $do9(client$, request, options) {
55404
55582
  response: doResult.value
55405
55583
  }];
55406
55584
  }
55407
- var init_evalRunsV2EvalRunsList = __esm(() => {
55408
- init_encodings();
55409
- init_primitives();
55410
- init_schemas4();
55411
- init_security();
55412
- init_url();
55413
- init_v2evalrunslistop();
55414
- init_async();
55415
- });
55416
-
55417
- // src/mcp-server/tools/evalRunsV2EvalRunsList.ts
55418
- var args6, tool$evalRunsV2EvalRunsList;
55419
- var init_evalRunsV2EvalRunsList2 = __esm(() => {
55420
- init_evalRunsV2EvalRunsList();
55421
- init_v2evalrunslistop();
55422
- init_tools();
55423
- args6 = {
55424
- request: V2EvalRunsListRequest$zodSchema.optional()
55425
- };
55426
- tool$evalRunsV2EvalRunsList = {
55427
- name: "eval-runs-v2-eval-runs-list",
55428
- description: `List Runs
55429
-
55430
- Returns jobs created through \`POST /v2/eval-runs\`, ordered newest first, with lifecycle state and progress.`,
55431
- annotations: {
55432
- title: "",
55433
- destructiveHint: false,
55434
- idempotentHint: false,
55435
- openWorldHint: false,
55436
- readOnlyHint: true
55437
- },
55438
- args: args6,
55439
- tool: async (client, args, ctx) => {
55440
- const [result] = await evalRunsV2EvalRunsList(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
55441
- if (!result.ok) {
55442
- return {
55443
- content: [{ type: "text", text: result.error.message }],
55444
- isError: true
55445
- };
55446
- }
55447
- return formatResult(result.value);
55448
- }
55449
- };
55450
- });
55451
-
55452
- // src/models/v2evalrunsretrieveop.ts
55453
- var V2EvalRunsRetrieveRequest$zodSchema, V2EvalRunsRetrieveResponse$zodSchema;
55454
- var init_v2evalrunsretrieveop = __esm(() => {
55455
- init_zod();
55456
- init_httpvalidationerror();
55457
- init_v2evalrun();
55458
- V2EvalRunsRetrieveRequest$zodSchema = object({
55459
- run_id: string2()
55460
- });
55461
- V2EvalRunsRetrieveResponse$zodSchema = union([
55462
- V2EvalRun$zodSchema,
55463
- HTTPValidationError$zodSchema
55464
- ]);
55465
- });
55466
-
55467
- // src/funcs/evalRunsV2EvalRunsRetrieve.ts
55468
- function evalRunsV2EvalRunsRetrieve(client$, request, options) {
55469
- return new APIPromise($do10(client$, request, options));
55470
- }
55471
- async function $do10(client$, request, options) {
55472
- const parsed$ = safeParse4(request, (value$) => V2EvalRunsRetrieveRequest$zodSchema.parse(value$), "Input validation failed");
55473
- if (!parsed$.ok) {
55474
- return [parsed$, { status: "invalid" }];
55475
- }
55476
- const payload$ = parsed$.value;
55477
- const body$ = null;
55478
- const pathParams$ = {
55479
- run_id: encodeSimple("run_id", payload$.run_id, {
55480
- explode: false,
55481
- charEncoding: "percent"
55482
- })
55483
- };
55484
- const path$ = pathToFunc("/v2/eval-runs/{run_id}")(pathParams$);
55485
- const headers$ = new Headers(compactMap({
55486
- Accept: "application/json"
55487
- }));
55488
- const securityInput = await extractSecurity(client$._options.security);
55489
- const requestSecurity = resolveGlobalSecurity(securityInput);
55490
- const context = {
55491
- options: client$._options,
55492
- baseURL: options?.serverURL ?? client$._baseURL ?? "",
55493
- operationID: "v2_eval_runs_retrieve",
55494
- oAuth2Scopes: null,
55495
- resolvedSecurity: requestSecurity,
55496
- securitySource: client$._options.security,
55497
- retryConfig: options?.retries || client$._options.retryConfig || { strategy: "none" },
55498
- retryCodes: options?.retryCodes || [
55499
- "429",
55500
- "500",
55501
- "502",
55502
- "503",
55503
- "504"
55504
- ]
55505
- };
55506
- const requestRes = client$._createRequest(context, {
55507
- security: requestSecurity,
55508
- method: "GET",
55509
- baseURL: options?.serverURL,
55510
- path: path$,
55511
- headers: headers$,
55512
- body: body$,
55513
- userAgent: client$._options.userAgent,
55514
- timeoutMs: options?.timeoutMs || client$._options.timeoutMs || -1
55515
- }, options);
55516
- if (!requestRes.ok) {
55517
- return [requestRes, { status: "invalid" }];
55518
- }
55519
- const req$ = requestRes.value;
55520
- const doResult = await client$._do(req$, {
55521
- context,
55522
- errorCodes: [],
55523
- retryConfig: context.retryConfig,
55524
- retryCodes: context.retryCodes
55525
- });
55526
- if (!doResult.ok) {
55527
- return [doResult, { status: "request-error", request: req$ }];
55528
- }
55529
- return [doResult, {
55530
- status: "complete",
55531
- request: req$,
55532
- response: doResult.value
55533
- }];
55534
- }
55535
- var init_evalRunsV2EvalRunsRetrieve = __esm(() => {
55585
+ var init_evalsGet = __esm(() => {
55536
55586
  init_encodings();
55537
55587
  init_primitives();
55538
55588
  init_schemas4();
@@ -55542,188 +55592,20 @@ var init_evalRunsV2EvalRunsRetrieve = __esm(() => {
55542
55592
  init_async();
55543
55593
  });
55544
55594
 
55545
- // src/mcp-server/tools/evalRunsV2EvalRunsRetrieve.ts
55546
- var args7, tool$evalRunsV2EvalRunsRetrieve;
55547
- var init_evalRunsV2EvalRunsRetrieve2 = __esm(() => {
55548
- init_evalRunsV2EvalRunsRetrieve();
55595
+ // src/mcp-server/tools/evalsGet.ts
55596
+ var args6, tool$evalsGet;
55597
+ var init_evalsGet2 = __esm(() => {
55598
+ init_evalsGet();
55549
55599
  init_v2evalrunsretrieveop();
55550
55600
  init_tools();
55551
- args7 = {
55601
+ args6 = {
55552
55602
  request: V2EvalRunsRetrieveRequest$zodSchema
55553
55603
  };
55554
- tool$evalRunsV2EvalRunsRetrieve = {
55555
- name: "eval-runs-v2-eval-runs-retrieve",
55556
- description: `Retrieve Run
55557
-
55558
- Returns the status and progress of an eval job. Poll until the status reaches a terminal value.`,
55559
- annotations: {
55560
- title: "",
55561
- destructiveHint: false,
55562
- idempotentHint: false,
55563
- openWorldHint: false,
55564
- readOnlyHint: true
55565
- },
55566
- args: args7,
55567
- tool: async (client, args, ctx) => {
55568
- const [result] = await evalRunsV2EvalRunsRetrieve(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
55569
- if (!result.ok) {
55570
- return {
55571
- content: [{ type: "text", text: result.error.message }],
55572
- isError: true
55573
- };
55574
- }
55575
- return formatResult(result.value);
55576
- }
55577
- };
55578
- });
55579
-
55580
- // src/models/v2evalrunanalysis.ts
55581
- var V2EvalRunAnalysis$zodSchema;
55582
- var init_v2evalrunanalysis = __esm(() => {
55583
- init_zod();
55584
- V2EvalRunAnalysis$zodSchema = object({
55585
- agent: string2().nullable().optional().describe("Runnable agent used for this concrete attempt."),
55586
- analysis: record(string2(), any()).optional().describe("Persisted eval_run_analysis payload, including verdict, outcome, findings, bottleneck, and analysis sections."),
55587
- analysis_version: string2().nullable().optional().describe("Version of the analysis pipeline that produced this payload."),
55588
- attempt_id: string2().describe("Child eval-attempt ID used to correlate the analysis with trace events."),
55589
- created_at: string2().nullable().optional(),
55590
- eval_id: string2().nullable().optional().describe("Public eval definition ID, not the backing executable eval ID."),
55591
- id: string2().describe("Persisted eval analysis ID."),
55592
- prompt: string2().nullable().optional().describe("Exact task prompt used for this concrete attempt."),
55593
- prompt_index: int().default(0).describe("Zero-based prompt index within the submitted simulation."),
55594
- repeat_index: int().default(0).describe("Zero-based repeat index within the submitted run."),
55595
- result_id: string2().describe("Concrete eval result ID analyzed for this attempt."),
55596
- updated_at: string2().nullable().optional(),
55597
- variant_key: string2().default("default").describe("Runtime variant that produced this result.")
55598
- });
55599
- });
55600
-
55601
- // src/models/v2evalrunanalysisresponse.ts
55602
- var V2EvalRunAnalysisResponse$zodSchema;
55603
- var init_v2evalrunanalysisresponse = __esm(() => {
55604
- init_zod();
55605
- init_publicevalrunprogress();
55606
- init_v2evalrunanalysis();
55607
- V2EvalRunAnalysisResponse$zodSchema = object({
55608
- data: array(V2EvalRunAnalysis$zodSchema).optional(),
55609
- error: string2().nullable().optional(),
55610
- progress: PublicEvalRunProgress$zodSchema.optional(),
55611
- run_id: string2(),
55612
- status: string2()
55613
- });
55614
- });
55615
-
55616
- // src/models/v2evalrunsanalysisop.ts
55617
- var V2EvalRunsAnalysisRequest$zodSchema, V2EvalRunsAnalysisResponseBody$zodSchema, V2EvalRunsAnalysisResponse$zodSchema;
55618
- var init_v2evalrunsanalysisop = __esm(() => {
55619
- init_zod();
55620
- init_httpvalidationerror();
55621
- init_v2evalrunanalysisresponse();
55622
- V2EvalRunsAnalysisRequest$zodSchema = object({
55623
- run_id: string2()
55624
- });
55625
- V2EvalRunsAnalysisResponseBody$zodSchema = object({
55626
- detail: string2()
55627
- }).describe("Eval run not found. Use the run ID returned when you started the eval.");
55628
- V2EvalRunsAnalysisResponse$zodSchema = union([
55629
- V2EvalRunAnalysisResponse$zodSchema,
55630
- lazy(() => V2EvalRunsAnalysisResponseBody$zodSchema),
55631
- HTTPValidationError$zodSchema
55632
- ]);
55633
- });
55634
-
55635
- // src/funcs/evalsAnalysis.ts
55636
- function evalsAnalysis(client$, request, options) {
55637
- return new APIPromise($do11(client$, request, options));
55638
- }
55639
- async function $do11(client$, request, options) {
55640
- const parsed$ = safeParse4(request, (value$) => V2EvalRunsAnalysisRequest$zodSchema.parse(value$), "Input validation failed");
55641
- if (!parsed$.ok) {
55642
- return [parsed$, { status: "invalid" }];
55643
- }
55644
- const payload$ = parsed$.value;
55645
- const body$ = null;
55646
- const pathParams$ = {
55647
- run_id: encodeSimple("run_id", payload$.run_id, {
55648
- explode: false,
55649
- charEncoding: "percent"
55650
- })
55651
- };
55652
- const path$ = pathToFunc("/v2/eval-runs/{run_id}/analysis")(pathParams$);
55653
- const headers$ = new Headers(compactMap({
55654
- Accept: "application/json"
55655
- }));
55656
- const securityInput = await extractSecurity(client$._options.security);
55657
- const requestSecurity = resolveGlobalSecurity(securityInput);
55658
- const context = {
55659
- options: client$._options,
55660
- baseURL: options?.serverURL ?? client$._baseURL ?? "",
55661
- operationID: "v2_eval_runs_analysis",
55662
- oAuth2Scopes: null,
55663
- resolvedSecurity: requestSecurity,
55664
- securitySource: client$._options.security,
55665
- retryConfig: options?.retries || client$._options.retryConfig || { strategy: "none" },
55666
- retryCodes: options?.retryCodes || [
55667
- "429",
55668
- "500",
55669
- "502",
55670
- "503",
55671
- "504"
55672
- ]
55673
- };
55674
- const requestRes = client$._createRequest(context, {
55675
- security: requestSecurity,
55676
- method: "GET",
55677
- baseURL: options?.serverURL,
55678
- path: path$,
55679
- headers: headers$,
55680
- body: body$,
55681
- userAgent: client$._options.userAgent,
55682
- timeoutMs: options?.timeoutMs || client$._options.timeoutMs || -1
55683
- }, options);
55684
- if (!requestRes.ok) {
55685
- return [requestRes, { status: "invalid" }];
55686
- }
55687
- const req$ = requestRes.value;
55688
- const doResult = await client$._do(req$, {
55689
- context,
55690
- errorCodes: [],
55691
- retryConfig: context.retryConfig,
55692
- retryCodes: context.retryCodes
55693
- });
55694
- if (!doResult.ok) {
55695
- return [doResult, { status: "request-error", request: req$ }];
55696
- }
55697
- return [doResult, {
55698
- status: "complete",
55699
- request: req$,
55700
- response: doResult.value
55701
- }];
55702
- }
55703
- var init_evalsAnalysis = __esm(() => {
55704
- init_encodings();
55705
- init_primitives();
55706
- init_schemas4();
55707
- init_security();
55708
- init_url();
55709
- init_v2evalrunsanalysisop();
55710
- init_async();
55711
- });
55712
-
55713
- // src/mcp-server/tools/evalsAnalysis.ts
55714
- var args8, tool$evalsAnalysis;
55715
- var init_evalsAnalysis2 = __esm(() => {
55716
- init_evalsAnalysis();
55717
- init_v2evalrunsanalysisop();
55718
- init_tools();
55719
- args8 = {
55720
- request: V2EvalRunsAnalysisRequest$zodSchema
55721
- };
55722
- tool$evalsAnalysis = {
55723
- name: "evals-analysis",
55724
- description: `Retrieve eval run analysis
55604
+ tool$evalsGet = {
55605
+ name: "evals-get",
55606
+ description: `Retrieve an eval run
55725
55607
 
55726
- Returns the persisted analysis produced for each completed attempt in one eval run. The analysis includes the verdict, outcome explanation, findings, bottleneck, and analysis sections. Active runs return the analysis available so far together with current run status and progress.`,
55608
+ Returns the run's current status and progress. Once its attempts finish, the response also includes their results and analysis. Poll this until the status reaches a terminal value. Use the trace operation for the raw execution event log, or the monitor operation to follow a run live.`,
55727
55609
  annotations: {
55728
55610
  title: "",
55729
55611
  destructiveHint: false,
@@ -55731,9 +55613,9 @@ Returns the persisted analysis produced for each completed attempt in one eval r
55731
55613
  openWorldHint: false,
55732
55614
  readOnlyHint: true
55733
55615
  },
55734
- args: args8,
55616
+ args: args6,
55735
55617
  tool: async (client, args, ctx) => {
55736
- const [result] = await evalsAnalysis(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
55618
+ const [result] = await evalsGet(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
55737
55619
  if (!result.ok) {
55738
55620
  return {
55739
55621
  content: [{ type: "text", text: result.error.message }],
@@ -55745,20 +55627,11 @@ Returns the persisted analysis produced for each completed attempt in one eval r
55745
55627
  };
55746
55628
  });
55747
55629
 
55748
- // src/models/v2inlineclireference.ts
55749
- var V2InlineCliReference$zodSchema;
55750
- var init_v2inlineclireference = __esm(() => {
55751
- init_zod();
55752
- V2InlineCliReference$zodSchema = object({
55753
- install_command: string2().describe("Shell command to install this CLI before the agent starts.")
55754
- });
55755
- });
55756
-
55757
55630
  // src/models/v2inlinemcpdefinition.ts
55758
- var Type$zodSchema, V2InlineMcpDefinition$zodSchema;
55631
+ var V2InlineMcpDefinitionType$zodSchema, V2InlineMcpDefinition$zodSchema;
55759
55632
  var init_v2inlinemcpdefinition = __esm(() => {
55760
55633
  init_zod();
55761
- Type$zodSchema = _enum([
55634
+ V2InlineMcpDefinitionType$zodSchema = _enum([
55762
55635
  "http",
55763
55636
  "sse",
55764
55637
  "stdio"
@@ -55769,20 +55642,11 @@ var init_v2inlinemcpdefinition = __esm(() => {
55769
55642
  env: record(string2(), string2()).optional().describe("Environment mapping for a `stdio` MCP server. Values may reference run secrets with `${VARIABLE_NAME}`."),
55770
55643
  headers: record(string2(), string2()).optional().describe("HTTP headers. Reference run secrets with `${VARIABLE_NAME}` instead of embedding credentials."),
55771
55644
  name: string2().describe("Stable server name used in the selected agent's MCP config."),
55772
- type: Type$zodSchema,
55645
+ type: V2InlineMcpDefinitionType$zodSchema,
55773
55646
  url: string2().nullable().optional().describe("HTTPS endpoint for an `http` or `sse` MCP server.")
55774
55647
  });
55775
55648
  });
55776
55649
 
55777
- // src/models/v2savedclireference.ts
55778
- var V2SavedCliReference$zodSchema;
55779
- var init_v2savedclireference = __esm(() => {
55780
- init_zod();
55781
- V2SavedCliReference$zodSchema = object({
55782
- id: string2().describe("Saved CLI ID from `GET /v2/clis`.")
55783
- });
55784
- });
55785
-
55786
55650
  // src/models/v2savedmcpreference.ts
55787
55651
  var V2SavedMcpReference$zodSchema;
55788
55652
  var init_v2savedmcpreference = __esm(() => {
@@ -55793,27 +55657,20 @@ var init_v2savedmcpreference = __esm(() => {
55793
55657
  });
55794
55658
 
55795
55659
  // src/models/createv2evalrunrequest.ts
55796
- var CreateV2EvalRunRequestCli$zodSchema, CreateV2EvalRunRequestMcp$zodSchema, CreateV2EvalRunRequest$zodSchema;
55660
+ var CreateV2EvalRunRequestMcp$zodSchema, CreateV2EvalRunRequest$zodSchema;
55797
55661
  var init_createv2evalrunrequest = __esm(() => {
55798
55662
  init_zod();
55799
- init_v2inlineclireference();
55663
+ init_searchgroundingmode();
55800
55664
  init_v2inlinemcpdefinition();
55801
- init_v2savedclireference();
55802
55665
  init_v2savedmcpreference();
55803
- CreateV2EvalRunRequestCli$zodSchema = union([
55804
- V2SavedCliReference$zodSchema,
55805
- V2InlineCliReference$zodSchema
55806
- ]);
55807
55666
  CreateV2EvalRunRequestMcp$zodSchema = union([
55808
55667
  V2InlineMcpDefinition$zodSchema,
55809
55668
  V2SavedMcpReference$zodSchema
55810
55669
  ]);
55811
55670
  CreateV2EvalRunRequest$zodSchema = object({
55812
55671
  agent: array(string2()).nullable().optional().describe("Runnable agents from `GET /v2/agents`. Required unless the saved eval already has agents. CLI clients expose repeatable `--agent`."),
55813
- clis: array(union([
55814
- V2SavedCliReference$zodSchema,
55815
- V2InlineCliReference$zodSchema
55816
- ])).nullable().optional().describe("Saved CLI references or inline installation commands. In saved mode, an explicit empty array disables saved CLIs for this launch."),
55672
+ cleanup_commands: array(string2()).nullable().optional().describe("Best-effort shell commands run sequentially after agent execution and result capture, before sandbox deletion. In saved mode, an explicit empty array disables saved cleanup commands for this launch. CLI clients expose repeatable `--cleanup-command`."),
55673
+ clis: array(string2()).nullable().optional().describe("Saved CLI IDs or shell installation commands. UUID values resolve saved CLIs from `GET /v2/clis`; every other value is run as an installation command. In saved mode, an explicit empty array disables saved CLIs for this launch. CLI clients expose repeatable `--cli`."),
55817
55674
  configuration_id: string2().optional().describe("Saved-run form only. Optional saved repository, environment, skills, and legacy tool configuration."),
55818
55675
  env: record(string2(), string2()).optional().describe("Write-only environment variables for this launch. Values are stored only in the queued job's private configuration."),
55819
55676
  eval_id: string2().optional().describe("Saved-run form only. Existing eval definition to run. Omit this field to use the direct create-and-run form."),
@@ -55827,6 +55684,8 @@ var init_createv2evalrunrequest = __esm(() => {
55827
55684
  prompts: array(string2()).nullable().optional().describe("Task prompts. Required in direct mode; in saved mode, replaces saved prompts for this launch. CLI clients expose repeatable `--prompt`."),
55828
55685
  repository: string2().optional().describe("Direct create-and-run form only. A saved repository ID, built-in alias such as `nextjs-starter`, or a public GitHub HTTPS URL."),
55829
55686
  runs: int().optional().describe("Independent attempts for every prompt-agent pair. Defaults to the saved eval value or 1 for a direct run."),
55687
+ search_grounding: SearchGroundingMode$zodSchema.nullable().optional().describe("Web-retrieval policy for this launch. Overrides the selected configuration or saved eval when provided; direct runs default to `default`."),
55688
+ setup_commands: array(string2()).nullable().optional().describe("Shell commands run sequentially in the prepared repository before the agent starts. In saved mode, an explicit empty array disables saved setup commands for this launch. CLI clients expose repeatable `--setup-command`."),
55830
55689
  skill_ids: array(string2()).optional().describe("Direct create-and-run form only. Reusable skill IDs from `GET /v2/skills` to install before the agent starts."),
55831
55690
  skills: array(string2()).optional().describe("Additional public skill packages to install before the agent starts in either request form, using GitHub `owner/repository` references such as `resend/resend-skills`. CLI clients may expose this string array as a repeatable `--skill` flag."),
55832
55691
  tags: array(string2()).optional().describe("Direct create-and-run form only. Optional labels saved on the created eval. CLI clients may expose this string array as a repeatable `--tag` flag.")
@@ -55835,9 +55694,9 @@ var init_createv2evalrunrequest = __esm(() => {
55835
55694
 
55836
55695
  // src/funcs/evalsRun.ts
55837
55696
  function evalsRun(client$, request, options) {
55838
- return new APIPromise($do12(client$, request, options));
55697
+ return new APIPromise($do10(client$, request, options));
55839
55698
  }
55840
- async function $do12(client$, request, options) {
55699
+ async function $do10(client$, request, options) {
55841
55700
  const parsed$ = safeParse4(request, (value$) => CreateV2EvalRunRequest$zodSchema.parse(value$), "Input validation failed");
55842
55701
  if (!parsed$.ok) {
55843
55702
  return [parsed$, { status: "invalid" }];
@@ -55907,26 +55766,28 @@ var init_evalsRun = __esm(() => {
55907
55766
  });
55908
55767
 
55909
55768
  // src/mcp-server/tools/evalsRun.ts
55910
- var args9, tool$evalsRun;
55769
+ var args7, tool$evalsRun;
55911
55770
  var init_evalsRun2 = __esm(() => {
55912
55771
  init_evalsRun();
55913
55772
  init_createv2evalrunrequest();
55914
55773
  init_tools();
55915
- args9 = {
55774
+ args7 = {
55916
55775
  request: CreateV2EvalRunRequest$zodSchema
55917
55776
  };
55918
55777
  tool$evalsRun = {
55919
55778
  name: "evals-run",
55920
55779
  description: `Start an eval run
55921
55780
 
55922
- Queues a durable eval job and returns immediately with a polling ID. Poll \`GET /v2/eval-runs/{run_id}\` until its status reaches a terminal value.
55781
+ Queues a durable eval job and returns immediately with a polling ID. Poll \`GET /v2/eval-runs/{run_id}\` until its status reaches a terminal value; once finished it also returns the results and analysis. Follow a run live with the monitor operation.
55923
55782
 
55924
55783
  Send one of these flat request forms:
55925
55784
 
55926
- - **Saved eval:** provide \`eval_id\`; optionally replace its prompts, persona, agents, CLIs, MCPs, or runs for this launch. A saved configuration can still provide repository, environment, skills, and legacy tool selections.
55927
- - **Direct create-and-run:** provide \`prompts\`, \`judges\`, and \`agent\`. You may also provide a persona, tags, repository, write-only environment variables, saved or inline CLIs and MCP servers, and skills.
55785
+ - **Saved eval:** provide \`eval_id\`; optionally replace its prompts, persona, agents, CLIs, MCPs, setup commands, cleanup commands, or runs for this launch. A saved configuration can still provide repository, environment, skills, and legacy tool selections.
55786
+ - **Direct create-and-run:** provide \`prompts\`, \`judges\`, and \`agent\`. You may also provide a persona, tags, repository, write-only environment variables, saved or inline CLIs and MCP servers, skills, setup commands, and cleanup commands.
55787
+
55788
+ The run fans out across prompts × agents × runs. Persona is shared metadata and does not multiply attempts. No separate \`mode\` field is required. If the organization lacks the estimated credits, the request fails with \`402\` before dispatching work.
55928
55789
 
55929
- The run fans out across prompts × agents × runs. Persona is shared metadata and does not multiply attempts. No separate \`mode\` field is required.`,
55790
+ A saved or inline environment containing an empty or placeholder credential is rejected with \`400\` before any evaluation is dispatched.`,
55930
55791
  annotations: {
55931
55792
  title: "",
55932
55793
  destructiveHint: false,
@@ -55934,7 +55795,7 @@ The run fans out across prompts × agents × runs. Persona is shared metadata an
55934
55795
  openWorldHint: false,
55935
55796
  readOnlyHint: false
55936
55797
  },
55937
- args: args9,
55798
+ args: args7,
55938
55799
  tool: async (client, args, ctx) => {
55939
55800
  const [result] = await evalsRun(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
55940
55801
  if (!result.ok) {
@@ -55966,7 +55827,7 @@ var init_v2evalruntraceevent = __esm(() => {
55966
55827
  repeat_index: int().default(0),
55967
55828
  run_id: string2(),
55968
55829
  timestamp: string2().nullable().optional(),
55969
- type: string2()
55830
+ type: string2().describe("Event type. Events are not tool invocations: one invocation usually emits a `tool_call` event and a `tool_result` event, and can emit extra `tool_call` updates as its arguments stream. Count invocations with the result's `tool_calls`, not with trace events.")
55970
55831
  });
55971
55832
  });
55972
55833
 
@@ -56010,9 +55871,9 @@ var init_v2evalrunstraceop = __esm(() => {
56010
55871
 
56011
55872
  // src/funcs/evalsTrace.ts
56012
55873
  function evalsTrace(client$, request, options) {
56013
- return new APIPromise($do13(client$, request, options));
55874
+ return new APIPromise($do11(client$, request, options));
56014
55875
  }
56015
- async function $do13(client$, request, options) {
55876
+ async function $do11(client$, request, options) {
56016
55877
  const parsed$ = safeParse4(request, (value$) => V2EvalRunsTraceRequest$zodSchema.parse(value$), "Input validation failed");
56017
55878
  if (!parsed$.ok) {
56018
55879
  return [parsed$, { status: "invalid" }];
@@ -56092,12 +55953,12 @@ var init_evalsTrace = __esm(() => {
56092
55953
  });
56093
55954
 
56094
55955
  // src/mcp-server/tools/evalsTrace.ts
56095
- var args10, tool$evalsTrace;
55956
+ var args8, tool$evalsTrace;
56096
55957
  var init_evalsTrace2 = __esm(() => {
56097
55958
  init_evalsTrace();
56098
55959
  init_v2evalrunstraceop();
56099
55960
  init_tools();
56100
- args10 = {
55961
+ args8 = {
56101
55962
  request: V2EvalRunsTraceRequest$zodSchema
56102
55963
  };
56103
55964
  tool$evalsTrace = {
@@ -56112,7 +55973,7 @@ Returns a finite page of sanitized execution events for one eval run. Use \`next
56112
55973
  openWorldHint: false,
56113
55974
  readOnlyHint: true
56114
55975
  },
56115
- args: args10,
55976
+ args: args8,
56116
55977
  tool: async (client, args, ctx) => {
56117
55978
  const [result] = await evalsTrace(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
56118
55979
  if (!result.ok) {
@@ -56130,6 +55991,7 @@ Returns a finite page of sanitized execution events for one eval run. Use \`next
56130
55991
  var CreateV2EvalConfigurationRequest$zodSchema;
56131
55992
  var init_createv2evalconfigurationrequest = __esm(() => {
56132
55993
  init_zod();
55994
+ init_searchgroundingmode();
56133
55995
  CreateV2EvalConfigurationRequest$zodSchema = object({
56134
55996
  cli_ids: array(string2()).optional(),
56135
55997
  environment_id: string2().optional(),
@@ -56137,6 +55999,7 @@ var init_createv2evalconfigurationrequest = __esm(() => {
56137
55999
  mcp_server_ids: array(string2()).optional(),
56138
56000
  name: string2(),
56139
56001
  repository_id: string2().optional(),
56002
+ search_grounding: SearchGroundingMode$zodSchema.optional().describe("Requested web-retrieval behavior for an eval run."),
56140
56003
  skill_ids: array(string2()).optional(),
56141
56004
  skills: array(string2()).optional().describe("Public skill packages saved with this configuration, using GitHub `owner/repository` references such as `resend/resend-skills`.")
56142
56005
  });
@@ -56146,6 +56009,7 @@ var init_createv2evalconfigurationrequest = __esm(() => {
56146
56009
  var V2EvalConfiguration$zodSchema;
56147
56010
  var init_v2evalconfiguration = __esm(() => {
56148
56011
  init_zod();
56012
+ init_searchgroundingmode();
56149
56013
  V2EvalConfiguration$zodSchema = object({
56150
56014
  cli_ids: array(string2()).optional(),
56151
56015
  created_at: string2().nullable().optional(),
@@ -56156,6 +56020,7 @@ var init_v2evalconfiguration = __esm(() => {
56156
56020
  mcp_server_ids: array(string2()).optional(),
56157
56021
  name: string2(),
56158
56022
  repository_id: string2().nullable().optional(),
56023
+ search_grounding: SearchGroundingMode$zodSchema.optional().describe("Requested web-retrieval behavior for an eval run."),
56159
56024
  skill_ids: array(string2()).optional(),
56160
56025
  skills: array(string2()).optional().describe("Public GitHub `owner/repository` skill packages installed for this configuration."),
56161
56026
  updated_at: string2().nullable().optional()
@@ -56181,9 +56046,9 @@ var init_v2evalconfigurationscreateop = __esm(() => {
56181
56046
 
56182
56047
  // src/funcs/evalsV2EvalConfigurationsCreate.ts
56183
56048
  function evalsV2EvalConfigurationsCreate(client$, request, options) {
56184
- return new APIPromise($do14(client$, request, options));
56049
+ return new APIPromise($do12(client$, request, options));
56185
56050
  }
56186
- async function $do14(client$, request, options) {
56051
+ async function $do12(client$, request, options) {
56187
56052
  const parsed$ = safeParse4(request, (value$) => V2EvalConfigurationsCreateRequest$zodSchema.parse(value$), "Input validation failed");
56188
56053
  if (!parsed$.ok) {
56189
56054
  return [parsed$, { status: "invalid" }];
@@ -56259,12 +56124,12 @@ var init_evalsV2EvalConfigurationsCreate = __esm(() => {
56259
56124
  });
56260
56125
 
56261
56126
  // src/mcp-server/tools/evalsV2EvalConfigurationsCreate.ts
56262
- var args11, tool$evalsV2EvalConfigurationsCreate;
56127
+ var args9, tool$evalsV2EvalConfigurationsCreate;
56263
56128
  var init_evalsV2EvalConfigurationsCreate2 = __esm(() => {
56264
56129
  init_evalsV2EvalConfigurationsCreate();
56265
56130
  init_v2evalconfigurationscreateop();
56266
56131
  init_tools();
56267
- args11 = {
56132
+ args9 = {
56268
56133
  request: V2EvalConfigurationsCreateRequest$zodSchema
56269
56134
  };
56270
56135
  tool$evalsV2EvalConfigurationsCreate = {
@@ -56279,7 +56144,7 @@ Creates a saved runtime configuration that pairs an eval with a repository, envi
56279
56144
  openWorldHint: false,
56280
56145
  readOnlyHint: false
56281
56146
  },
56282
- args: args11,
56147
+ args: args9,
56283
56148
  tool: async (client, args, ctx) => {
56284
56149
  const [result] = await evalsV2EvalConfigurationsCreate(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
56285
56150
  if (!result.ok) {
@@ -56311,9 +56176,9 @@ var init_v2evalconfigurationsdeleteop = __esm(() => {
56311
56176
 
56312
56177
  // src/funcs/evalsV2EvalConfigurationsDelete.ts
56313
56178
  function evalsV2EvalConfigurationsDelete(client$, request, options) {
56314
- return new APIPromise($do15(client$, request, options));
56179
+ return new APIPromise($do13(client$, request, options));
56315
56180
  }
56316
- async function $do15(client$, request, options) {
56181
+ async function $do13(client$, request, options) {
56317
56182
  const parsed$ = safeParse4(request, (value$) => V2EvalConfigurationsDeleteRequest$zodSchema.parse(value$), "Input validation failed");
56318
56183
  if (!parsed$.ok) {
56319
56184
  return [parsed$, { status: "invalid" }];
@@ -56389,12 +56254,12 @@ var init_evalsV2EvalConfigurationsDelete = __esm(() => {
56389
56254
  });
56390
56255
 
56391
56256
  // src/mcp-server/tools/evalsV2EvalConfigurationsDelete.ts
56392
- var args12, tool$evalsV2EvalConfigurationsDelete;
56257
+ var args10, tool$evalsV2EvalConfigurationsDelete;
56393
56258
  var init_evalsV2EvalConfigurationsDelete2 = __esm(() => {
56394
56259
  init_evalsV2EvalConfigurationsDelete();
56395
56260
  init_v2evalconfigurationsdeleteop();
56396
56261
  init_tools();
56397
- args12 = {
56262
+ args10 = {
56398
56263
  request: V2EvalConfigurationsDeleteRequest$zodSchema
56399
56264
  };
56400
56265
  tool$evalsV2EvalConfigurationsDelete = {
@@ -56409,7 +56274,7 @@ Permanently deletes an eval configuration and returns its last saved state.`,
56409
56274
  openWorldHint: false,
56410
56275
  readOnlyHint: false
56411
56276
  },
56412
- args: args12,
56277
+ args: args10,
56413
56278
  tool: async (client, args, ctx) => {
56414
56279
  const [result] = await evalsV2EvalConfigurationsDelete(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
56415
56280
  if (!result.ok) {
@@ -56451,9 +56316,9 @@ var init_v2evalconfigurationslistop = __esm(() => {
56451
56316
 
56452
56317
  // src/funcs/evalsV2EvalConfigurationsList.ts
56453
56318
  function evalsV2EvalConfigurationsList(client$, request, options) {
56454
- return new APIPromise($do16(client$, request, options));
56319
+ return new APIPromise($do14(client$, request, options));
56455
56320
  }
56456
- async function $do16(client$, request, options) {
56321
+ async function $do14(client$, request, options) {
56457
56322
  const parsed$ = safeParse4(request, (value$) => V2EvalConfigurationsListRequest$zodSchema.parse(value$), "Input validation failed");
56458
56323
  if (!parsed$.ok) {
56459
56324
  return [parsed$, { status: "invalid" }];
@@ -56528,12 +56393,12 @@ var init_evalsV2EvalConfigurationsList = __esm(() => {
56528
56393
  });
56529
56394
 
56530
56395
  // src/mcp-server/tools/evalsV2EvalConfigurationsList.ts
56531
- var args13, tool$evalsV2EvalConfigurationsList;
56396
+ var args11, tool$evalsV2EvalConfigurationsList;
56532
56397
  var init_evalsV2EvalConfigurationsList2 = __esm(() => {
56533
56398
  init_evalsV2EvalConfigurationsList();
56534
56399
  init_v2evalconfigurationslistop();
56535
56400
  init_tools();
56536
- args13 = {
56401
+ args11 = {
56537
56402
  request: V2EvalConfigurationsListRequest$zodSchema
56538
56403
  };
56539
56404
  tool$evalsV2EvalConfigurationsList = {
@@ -56548,7 +56413,7 @@ Returns the saved runtime configurations for an eval, including repository, envi
56548
56413
  openWorldHint: false,
56549
56414
  readOnlyHint: true
56550
56415
  },
56551
- args: args13,
56416
+ args: args11,
56552
56417
  tool: async (client, args, ctx) => {
56553
56418
  const [result] = await evalsV2EvalConfigurationsList(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
56554
56419
  if (!result.ok) {
@@ -56580,9 +56445,9 @@ var init_v2evalconfigurationsretrieveop = __esm(() => {
56580
56445
 
56581
56446
  // src/funcs/evalsV2EvalConfigurationsRetrieve.ts
56582
56447
  function evalsV2EvalConfigurationsRetrieve(client$, request, options) {
56583
- return new APIPromise($do17(client$, request, options));
56448
+ return new APIPromise($do15(client$, request, options));
56584
56449
  }
56585
- async function $do17(client$, request, options) {
56450
+ async function $do15(client$, request, options) {
56586
56451
  const parsed$ = safeParse4(request, (value$) => V2EvalConfigurationsRetrieveRequest$zodSchema.parse(value$), "Input validation failed");
56587
56452
  if (!parsed$.ok) {
56588
56453
  return [parsed$, { status: "invalid" }];
@@ -56647,30 +56512,865 @@ async function $do17(client$, request, options) {
56647
56512
  response: doResult.value
56648
56513
  }];
56649
56514
  }
56650
- var init_evalsV2EvalConfigurationsRetrieve = __esm(() => {
56515
+ var init_evalsV2EvalConfigurationsRetrieve = __esm(() => {
56516
+ init_encodings();
56517
+ init_primitives();
56518
+ init_schemas4();
56519
+ init_security();
56520
+ init_url();
56521
+ init_v2evalconfigurationsretrieveop();
56522
+ init_async();
56523
+ });
56524
+
56525
+ // src/mcp-server/tools/evalsV2EvalConfigurationsRetrieve.ts
56526
+ var args12, tool$evalsV2EvalConfigurationsRetrieve;
56527
+ var init_evalsV2EvalConfigurationsRetrieve2 = __esm(() => {
56528
+ init_evalsV2EvalConfigurationsRetrieve();
56529
+ init_v2evalconfigurationsretrieveop();
56530
+ init_tools();
56531
+ args12 = {
56532
+ request: V2EvalConfigurationsRetrieveRequest$zodSchema
56533
+ };
56534
+ tool$evalsV2EvalConfigurationsRetrieve = {
56535
+ name: "evals-v2-eval-configurations-retrieve",
56536
+ description: `Retrieve Configuration
56537
+
56538
+ Returns one saved configuration for an eval.`,
56539
+ annotations: {
56540
+ title: "",
56541
+ destructiveHint: false,
56542
+ idempotentHint: false,
56543
+ openWorldHint: false,
56544
+ readOnlyHint: true
56545
+ },
56546
+ args: args12,
56547
+ tool: async (client, args, ctx) => {
56548
+ const [result] = await evalsV2EvalConfigurationsRetrieve(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
56549
+ if (!result.ok) {
56550
+ return {
56551
+ content: [{ type: "text", text: result.error.message }],
56552
+ isError: true
56553
+ };
56554
+ }
56555
+ return formatResult(result.value);
56556
+ }
56557
+ };
56558
+ });
56559
+
56560
+ // src/models/updatev2evalconfigurationrequest.ts
56561
+ var UpdateV2EvalConfigurationRequest$zodSchema;
56562
+ var init_updatev2evalconfigurationrequest = __esm(() => {
56563
+ init_zod();
56564
+ init_searchgroundingmode();
56565
+ UpdateV2EvalConfigurationRequest$zodSchema = object({
56566
+ cli_ids: array(string2()).nullable().optional(),
56567
+ environment_id: string2().optional(),
56568
+ include_docs: boolean2().optional(),
56569
+ mcp_server_ids: array(string2()).nullable().optional(),
56570
+ name: string2().optional(),
56571
+ repository_id: string2().optional(),
56572
+ search_grounding: SearchGroundingMode$zodSchema.nullable().optional().describe("Replacement web-retrieval policy. Omit it to keep the saved value."),
56573
+ skill_ids: array(string2()).nullable().optional(),
56574
+ skills: array(string2()).nullable().optional().describe("Public skill packages saved with this configuration, using GitHub `owner/repository` references.")
56575
+ });
56576
+ });
56577
+
56578
+ // src/models/v2evalconfigurationsupdateop.ts
56579
+ var V2EvalConfigurationsUpdateRequest$zodSchema, V2EvalConfigurationsUpdateResponse$zodSchema;
56580
+ var init_v2evalconfigurationsupdateop = __esm(() => {
56581
+ init_zod();
56582
+ init_httpvalidationerror();
56583
+ init_updatev2evalconfigurationrequest();
56584
+ init_v2evalconfiguration();
56585
+ V2EvalConfigurationsUpdateRequest$zodSchema = object({
56586
+ body: UpdateV2EvalConfigurationRequest$zodSchema,
56587
+ configuration_id: string2(),
56588
+ eval_id: string2()
56589
+ });
56590
+ V2EvalConfigurationsUpdateResponse$zodSchema = union([
56591
+ V2EvalConfiguration$zodSchema,
56592
+ HTTPValidationError$zodSchema
56593
+ ]);
56594
+ });
56595
+
56596
+ // src/funcs/evalsV2EvalConfigurationsUpdate.ts
56597
+ function evalsV2EvalConfigurationsUpdate(client$, request, options) {
56598
+ return new APIPromise($do16(client$, request, options));
56599
+ }
56600
+ async function $do16(client$, request, options) {
56601
+ const parsed$ = safeParse4(request, (value$) => V2EvalConfigurationsUpdateRequest$zodSchema.parse(value$), "Input validation failed");
56602
+ if (!parsed$.ok) {
56603
+ return [parsed$, { status: "invalid" }];
56604
+ }
56605
+ const payload$ = parsed$.value;
56606
+ const body$ = encodeJSON("body", payload$.body, { explode: true });
56607
+ const pathParams$ = {
56608
+ configuration_id: encodeSimple("configuration_id", payload$.configuration_id, { explode: false, charEncoding: "percent" }),
56609
+ eval_id: encodeSimple("eval_id", payload$.eval_id, {
56610
+ explode: false,
56611
+ charEncoding: "percent"
56612
+ })
56613
+ };
56614
+ const path$ = pathToFunc("/v2/evals/{eval_id}/configurations/{configuration_id}")(pathParams$);
56615
+ const headers$ = new Headers(compactMap({
56616
+ "Content-Type": "application/json",
56617
+ Accept: "application/json"
56618
+ }));
56619
+ const securityInput = await extractSecurity(client$._options.security);
56620
+ const requestSecurity = resolveGlobalSecurity(securityInput);
56621
+ const context = {
56622
+ options: client$._options,
56623
+ baseURL: options?.serverURL ?? client$._baseURL ?? "",
56624
+ operationID: "v2_eval_configurations_update",
56625
+ oAuth2Scopes: null,
56626
+ resolvedSecurity: requestSecurity,
56627
+ securitySource: client$._options.security,
56628
+ retryConfig: options?.retries || client$._options.retryConfig || { strategy: "none" },
56629
+ retryCodes: options?.retryCodes || [
56630
+ "429",
56631
+ "500",
56632
+ "502",
56633
+ "503",
56634
+ "504"
56635
+ ]
56636
+ };
56637
+ const requestRes = client$._createRequest(context, {
56638
+ security: requestSecurity,
56639
+ method: "PATCH",
56640
+ baseURL: options?.serverURL,
56641
+ path: path$,
56642
+ headers: headers$,
56643
+ body: body$,
56644
+ userAgent: client$._options.userAgent,
56645
+ timeoutMs: options?.timeoutMs || client$._options.timeoutMs || -1
56646
+ }, options);
56647
+ if (!requestRes.ok) {
56648
+ return [requestRes, { status: "invalid" }];
56649
+ }
56650
+ const req$ = requestRes.value;
56651
+ const doResult = await client$._do(req$, {
56652
+ context,
56653
+ errorCodes: [],
56654
+ retryConfig: context.retryConfig,
56655
+ retryCodes: context.retryCodes
56656
+ });
56657
+ if (!doResult.ok) {
56658
+ return [doResult, { status: "request-error", request: req$ }];
56659
+ }
56660
+ return [doResult, {
56661
+ status: "complete",
56662
+ request: req$,
56663
+ response: doResult.value
56664
+ }];
56665
+ }
56666
+ var init_evalsV2EvalConfigurationsUpdate = __esm(() => {
56667
+ init_encodings();
56668
+ init_primitives();
56669
+ init_schemas4();
56670
+ init_security();
56671
+ init_url();
56672
+ init_v2evalconfigurationsupdateop();
56673
+ init_async();
56674
+ });
56675
+
56676
+ // src/mcp-server/tools/evalsV2EvalConfigurationsUpdate.ts
56677
+ var args13, tool$evalsV2EvalConfigurationsUpdate;
56678
+ var init_evalsV2EvalConfigurationsUpdate2 = __esm(() => {
56679
+ init_evalsV2EvalConfigurationsUpdate();
56680
+ init_v2evalconfigurationsupdateop();
56681
+ init_tools();
56682
+ args13 = {
56683
+ request: V2EvalConfigurationsUpdateRequest$zodSchema
56684
+ };
56685
+ tool$evalsV2EvalConfigurationsUpdate = {
56686
+ name: "evals-v2-eval-configurations-update",
56687
+ description: `Update Configuration
56688
+
56689
+ Updates an eval configuration; omitted fields remain unchanged.`,
56690
+ annotations: {
56691
+ title: "",
56692
+ destructiveHint: false,
56693
+ idempotentHint: false,
56694
+ openWorldHint: false,
56695
+ readOnlyHint: false
56696
+ },
56697
+ args: args13,
56698
+ tool: async (client, args, ctx) => {
56699
+ const [result] = await evalsV2EvalConfigurationsUpdate(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
56700
+ if (!result.ok) {
56701
+ return {
56702
+ content: [{ type: "text", text: result.error.message }],
56703
+ isError: true
56704
+ };
56705
+ }
56706
+ return formatResult(result.value);
56707
+ }
56708
+ };
56709
+ });
56710
+
56711
+ // src/models/evaltemplatebatchitem.ts
56712
+ var EvalTemplateBatchItemType$zodSchema, EvalTemplateBatchItem$zodSchema;
56713
+ var init_evaltemplatebatchitem = __esm(() => {
56714
+ init_zod();
56715
+ EvalTemplateBatchItemType$zodSchema = _enum([
56716
+ "workflow",
56717
+ "mcp",
56718
+ "cli",
56719
+ "skill"
56720
+ ]).describe("Eval subject type used for organization and reporting.");
56721
+ EvalTemplateBatchItem$zodSchema = object({
56722
+ name: string2().describe("Name of the prompt-specific eval to create."),
56723
+ prompts: array(string2()).describe("Task prompts owned by this eval. Judges, targets, scheduling, and runtime resources come from the template."),
56724
+ subject_id: string2().nullable().optional().describe("Optional subject identifier associated with this eval."),
56725
+ type: EvalTemplateBatchItemType$zodSchema.default("workflow").describe("Eval subject type used for organization and reporting.")
56726
+ }).describe("Prompt-specific fields combined with the referenced template.");
56727
+ });
56728
+
56729
+ // src/models/createevaltemplatebatchrequest.ts
56730
+ var CreateEvalTemplateBatchRequest$zodSchema;
56731
+ var init_createevaltemplatebatchrequest = __esm(() => {
56732
+ init_zod();
56733
+ init_evaltemplatebatchitem();
56734
+ CreateEvalTemplateBatchRequest$zodSchema = object({
56735
+ evals: array(EvalTemplateBatchItem$zodSchema).describe("Prompt-specific evals to create from the referenced template. Template-owned fields are intentionally rejected here to avoid ambiguous overrides."),
56736
+ expected_template_version: int().optional().describe("Template `version` returned when the template was read or created. When supplied, creation fails if the template changed before reuse."),
56737
+ template_id: string2().describe("Template `id` returned by `POST /v2/eval-templates`.")
56738
+ }).describe("Reuse one template across up to 100 prompt-specific eval definitions.");
56739
+ });
56740
+
56741
+ // src/funcs/evalsV2EvalsBatchCreate.ts
56742
+ function evalsV2EvalsBatchCreate(client$, request, options) {
56743
+ return new APIPromise($do17(client$, request, options));
56744
+ }
56745
+ async function $do17(client$, request, options) {
56746
+ const parsed$ = safeParse4(request, (value$) => CreateEvalTemplateBatchRequest$zodSchema.parse(value$), "Input validation failed");
56747
+ if (!parsed$.ok) {
56748
+ return [parsed$, { status: "invalid" }];
56749
+ }
56750
+ const payload$ = parsed$.value;
56751
+ const body$ = encodeJSON("body", payload$, { explode: true });
56752
+ const path$ = pathToFunc("/v2/evals/batch")();
56753
+ const headers$ = new Headers(compactMap({
56754
+ "Content-Type": "application/json",
56755
+ Accept: "application/json"
56756
+ }));
56757
+ const securityInput = await extractSecurity(client$._options.security);
56758
+ const requestSecurity = resolveGlobalSecurity(securityInput);
56759
+ const context = {
56760
+ options: client$._options,
56761
+ baseURL: options?.serverURL ?? client$._baseURL ?? "",
56762
+ operationID: "v2_evals_batch_create",
56763
+ oAuth2Scopes: null,
56764
+ resolvedSecurity: requestSecurity,
56765
+ securitySource: client$._options.security,
56766
+ retryConfig: options?.retries || client$._options.retryConfig || { strategy: "none" },
56767
+ retryCodes: options?.retryCodes || [
56768
+ "429",
56769
+ "500",
56770
+ "502",
56771
+ "503",
56772
+ "504"
56773
+ ]
56774
+ };
56775
+ const requestRes = client$._createRequest(context, {
56776
+ security: requestSecurity,
56777
+ method: "POST",
56778
+ baseURL: options?.serverURL,
56779
+ path: path$,
56780
+ headers: headers$,
56781
+ body: body$,
56782
+ userAgent: client$._options.userAgent,
56783
+ timeoutMs: options?.timeoutMs || client$._options.timeoutMs || -1
56784
+ }, options);
56785
+ if (!requestRes.ok) {
56786
+ return [requestRes, { status: "invalid" }];
56787
+ }
56788
+ const req$ = requestRes.value;
56789
+ const doResult = await client$._do(req$, {
56790
+ context,
56791
+ errorCodes: [],
56792
+ retryConfig: context.retryConfig,
56793
+ retryCodes: context.retryCodes
56794
+ });
56795
+ if (!doResult.ok) {
56796
+ return [doResult, { status: "request-error", request: req$ }];
56797
+ }
56798
+ return [doResult, {
56799
+ status: "complete",
56800
+ request: req$,
56801
+ response: doResult.value
56802
+ }];
56803
+ }
56804
+ var init_evalsV2EvalsBatchCreate = __esm(() => {
56805
+ init_encodings();
56806
+ init_primitives();
56807
+ init_schemas4();
56808
+ init_security();
56809
+ init_url();
56810
+ init_createevaltemplatebatchrequest();
56811
+ init_async();
56812
+ });
56813
+
56814
+ // src/mcp-server/tools/evalsV2EvalsBatchCreate.ts
56815
+ var args14, tool$evalsV2EvalsBatchCreate;
56816
+ var init_evalsV2EvalsBatchCreate2 = __esm(() => {
56817
+ init_evalsV2EvalsBatchCreate();
56818
+ init_createevaltemplatebatchrequest();
56819
+ init_tools();
56820
+ args14 = {
56821
+ request: CreateEvalTemplateBatchRequest$zodSchema
56822
+ };
56823
+ tool$evalsV2EvalsBatchCreate = {
56824
+ name: "evals-v2-evals-batch-create",
56825
+ description: `Create evals from a template
56826
+
56827
+ Step 2 of the reusable template workflow. Provide the \`id\` and \`version\` returned by \`POST /v2/eval-templates\` as \`template_id\` and \`expected_template_version\`, then provide only prompt-specific fields in \`evals\`. The template supplies only the settings named in its \`included_components\`; omitted settings keep the values supplied by the creation workflow or its defaults. Included values are copied at creation time, so later template changes do not alter the created evals. Competitor selections create one additional eval per selected competitor for every input item, and the response groups each base eval with those variants. Brand-scoped runtime resources are validated against the organization's primary brand before any eval is created.`,
56828
+ annotations: {
56829
+ title: "",
56830
+ destructiveHint: false,
56831
+ idempotentHint: false,
56832
+ openWorldHint: false,
56833
+ readOnlyHint: false
56834
+ },
56835
+ args: args14,
56836
+ tool: async (client, args, ctx) => {
56837
+ const [result] = await evalsV2EvalsBatchCreate(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
56838
+ if (!result.ok) {
56839
+ return {
56840
+ content: [{ type: "text", text: result.error.message }],
56841
+ isError: true
56842
+ };
56843
+ }
56844
+ return formatResult(result.value);
56845
+ }
56846
+ };
56847
+ });
56848
+
56849
+ // src/models/createv2evalrequest.ts
56850
+ var CreateV2EvalRequestMcp$zodSchema, CreateV2EvalRequest$zodSchema;
56851
+ var init_createv2evalrequest = __esm(() => {
56852
+ init_zod();
56853
+ init_searchgroundingmode();
56854
+ init_v2inlinemcpdefinition();
56855
+ init_v2savedmcpreference();
56856
+ CreateV2EvalRequestMcp$zodSchema = union([
56857
+ V2InlineMcpDefinition$zodSchema,
56858
+ V2SavedMcpReference$zodSchema
56859
+ ]);
56860
+ CreateV2EvalRequest$zodSchema = object({
56861
+ agent: array(string2()).optional().describe("Runnable agents from `GET /v2/agents`. An empty array saves the eval without default agents. CLI clients expose repeatable `--agent`."),
56862
+ cleanup_commands: array(string2()).optional().describe("Best-effort shell commands run sequentially after agent execution and result capture, before sandbox deletion. CLI clients expose repeatable `--cleanup-command`."),
56863
+ clis: array(string2()).optional().describe("Saved CLI IDs or shell installation commands. UUID values resolve saved CLIs from `GET /v2/clis`; every other value is run as an installation command. CLI clients expose repeatable `--cli`."),
56864
+ judges: array(string2()).describe("Plain-text success criteria. CLI clients expose this as repeatable `--judge`."),
56865
+ mcps: array(union([
56866
+ V2InlineMcpDefinition$zodSchema,
56867
+ V2SavedMcpReference$zodSchema
56868
+ ])).optional().describe("Saved MCP references or inline portable MCP definitions."),
56869
+ name: string2(),
56870
+ persona: string2().optional().describe("Optional perspective applied to the whole eval. It is stored as metadata and does not change prompts or multiply attempts."),
56871
+ prompts: array(string2()).describe("Task prompts. CLI clients expose this as repeatable `--prompt`."),
56872
+ runs: int().default(1).describe("Independent attempts for every prompt-agent pair."),
56873
+ search_grounding: SearchGroundingMode$zodSchema.optional().describe("Requested web-retrieval behavior for an eval run."),
56874
+ setup_commands: array(string2()).optional().describe("Shell commands run sequentially in the prepared repository before the agent starts. CLI clients expose repeatable `--setup-command`."),
56875
+ tags: array(string2()).optional().describe("Optional labels for organizing the eval. CLI clients may expose this string array as a repeatable `--tag` flag.")
56876
+ });
56877
+ });
56878
+
56879
+ // src/funcs/evalsV2EvalsCreate.ts
56880
+ function evalsV2EvalsCreate(client$, request, options) {
56881
+ return new APIPromise($do18(client$, request, options));
56882
+ }
56883
+ async function $do18(client$, request, options) {
56884
+ const parsed$ = safeParse4(request, (value$) => CreateV2EvalRequest$zodSchema.parse(value$), "Input validation failed");
56885
+ if (!parsed$.ok) {
56886
+ return [parsed$, { status: "invalid" }];
56887
+ }
56888
+ const payload$ = parsed$.value;
56889
+ const body$ = encodeJSON("body", payload$, { explode: true });
56890
+ const path$ = pathToFunc("/v2/evals")();
56891
+ const headers$ = new Headers(compactMap({
56892
+ "Content-Type": "application/json",
56893
+ Accept: "application/json"
56894
+ }));
56895
+ const securityInput = await extractSecurity(client$._options.security);
56896
+ const requestSecurity = resolveGlobalSecurity(securityInput);
56897
+ const context = {
56898
+ options: client$._options,
56899
+ baseURL: options?.serverURL ?? client$._baseURL ?? "",
56900
+ operationID: "v2_evals_create",
56901
+ oAuth2Scopes: null,
56902
+ resolvedSecurity: requestSecurity,
56903
+ securitySource: client$._options.security,
56904
+ retryConfig: options?.retries || client$._options.retryConfig || { strategy: "none" },
56905
+ retryCodes: options?.retryCodes || [
56906
+ "429",
56907
+ "500",
56908
+ "502",
56909
+ "503",
56910
+ "504"
56911
+ ]
56912
+ };
56913
+ const requestRes = client$._createRequest(context, {
56914
+ security: requestSecurity,
56915
+ method: "POST",
56916
+ baseURL: options?.serverURL,
56917
+ path: path$,
56918
+ headers: headers$,
56919
+ body: body$,
56920
+ userAgent: client$._options.userAgent,
56921
+ timeoutMs: options?.timeoutMs || client$._options.timeoutMs || -1
56922
+ }, options);
56923
+ if (!requestRes.ok) {
56924
+ return [requestRes, { status: "invalid" }];
56925
+ }
56926
+ const req$ = requestRes.value;
56927
+ const doResult = await client$._do(req$, {
56928
+ context,
56929
+ errorCodes: [],
56930
+ retryConfig: context.retryConfig,
56931
+ retryCodes: context.retryCodes
56932
+ });
56933
+ if (!doResult.ok) {
56934
+ return [doResult, { status: "request-error", request: req$ }];
56935
+ }
56936
+ return [doResult, {
56937
+ status: "complete",
56938
+ request: req$,
56939
+ response: doResult.value
56940
+ }];
56941
+ }
56942
+ var init_evalsV2EvalsCreate = __esm(() => {
56943
+ init_encodings();
56944
+ init_primitives();
56945
+ init_schemas4();
56946
+ init_security();
56947
+ init_url();
56948
+ init_createv2evalrequest();
56949
+ init_async();
56950
+ });
56951
+
56952
+ // src/mcp-server/tools/evalsV2EvalsCreate.ts
56953
+ var args15, tool$evalsV2EvalsCreate;
56954
+ var init_evalsV2EvalsCreate2 = __esm(() => {
56955
+ init_evalsV2EvalsCreate();
56956
+ init_createv2evalrequest();
56957
+ init_tools();
56958
+ args15 = {
56959
+ request: CreateV2EvalRequest$zodSchema
56960
+ };
56961
+ tool$evalsV2EvalsCreate = {
56962
+ name: "evals-v2-evals-create",
56963
+ description: `Create Eval
56964
+
56965
+ Creates a reusable simulation definition with one or more task prompts and judges. Optional agents, persona, CLI installations, MCP servers, setup commands, cleanup commands, grounding policy, repeat count, and tags are saved with the eval. \`search_grounding\` supports \`disabled\`, \`enabled\`, and \`default\`; \`default\` preserves normal platform behavior. Creating the eval does not start a run and does not accept secret environment values.`,
56966
+ annotations: {
56967
+ title: "",
56968
+ destructiveHint: false,
56969
+ idempotentHint: false,
56970
+ openWorldHint: false,
56971
+ readOnlyHint: false
56972
+ },
56973
+ args: args15,
56974
+ tool: async (client, args, ctx) => {
56975
+ const [result] = await evalsV2EvalsCreate(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
56976
+ if (!result.ok) {
56977
+ return {
56978
+ content: [{ type: "text", text: result.error.message }],
56979
+ isError: true
56980
+ };
56981
+ }
56982
+ return formatResult(result.value);
56983
+ }
56984
+ };
56985
+ });
56986
+
56987
+ // src/models/v2inlineclireference.ts
56988
+ var V2InlineCliReference$zodSchema;
56989
+ var init_v2inlineclireference = __esm(() => {
56990
+ init_zod();
56991
+ V2InlineCliReference$zodSchema = object({
56992
+ install_command: string2().describe("Shell command to install this CLI before the agent starts.")
56993
+ });
56994
+ });
56995
+
56996
+ // src/models/v2savedclireference.ts
56997
+ var V2SavedCliReference$zodSchema;
56998
+ var init_v2savedclireference = __esm(() => {
56999
+ init_zod();
57000
+ V2SavedCliReference$zodSchema = object({
57001
+ id: string2().describe("Saved CLI ID from `GET /v2/clis`.")
57002
+ });
57003
+ });
57004
+
57005
+ // src/models/v2eval.ts
57006
+ var Cli$zodSchema, V2EvalMcp$zodSchema, V2Eval$zodSchema;
57007
+ var init_v2eval = __esm(() => {
57008
+ init_zod();
57009
+ init_searchgroundingmode();
57010
+ init_v2inlineclireference();
57011
+ init_v2inlinemcpdefinition();
57012
+ init_v2savedclireference();
57013
+ init_v2savedmcpreference();
57014
+ Cli$zodSchema = union([
57015
+ V2SavedCliReference$zodSchema,
57016
+ V2InlineCliReference$zodSchema
57017
+ ]);
57018
+ V2EvalMcp$zodSchema = union([
57019
+ V2InlineMcpDefinition$zodSchema,
57020
+ V2SavedMcpReference$zodSchema
57021
+ ]);
57022
+ V2Eval$zodSchema = object({
57023
+ agent: array(string2()).optional(),
57024
+ cleanup_commands: array(string2()).optional(),
57025
+ clis: array(union([
57026
+ V2SavedCliReference$zodSchema,
57027
+ V2InlineCliReference$zodSchema
57028
+ ])).optional(),
57029
+ created_at: string2().nullable().optional(),
57030
+ id: string2(),
57031
+ judges: array(string2()).optional(),
57032
+ mcps: array(union([
57033
+ V2InlineMcpDefinition$zodSchema,
57034
+ V2SavedMcpReference$zodSchema
57035
+ ])).optional(),
57036
+ name: string2(),
57037
+ persona: string2().nullable().optional(),
57038
+ prompts: array(string2()).optional(),
57039
+ runs: int().default(1),
57040
+ search_grounding: SearchGroundingMode$zodSchema.optional().describe("Requested web-retrieval behavior for an eval run."),
57041
+ setup_commands: array(string2()).optional(),
57042
+ tags: array(string2()).optional(),
57043
+ updated_at: string2().nullable().optional()
57044
+ });
57045
+ });
57046
+
57047
+ // src/models/v2evalsdeleteop.ts
57048
+ var V2EvalsDeleteRequest$zodSchema, V2EvalsDeleteResponse$zodSchema;
57049
+ var init_v2evalsdeleteop = __esm(() => {
57050
+ init_zod();
57051
+ init_httpvalidationerror();
57052
+ init_v2eval();
57053
+ V2EvalsDeleteRequest$zodSchema = object({
57054
+ eval_id: string2()
57055
+ });
57056
+ V2EvalsDeleteResponse$zodSchema = union([
57057
+ V2Eval$zodSchema,
57058
+ HTTPValidationError$zodSchema
57059
+ ]);
57060
+ });
57061
+
57062
+ // src/funcs/evalsV2EvalsDelete.ts
57063
+ function evalsV2EvalsDelete(client$, request, options) {
57064
+ return new APIPromise($do19(client$, request, options));
57065
+ }
57066
+ async function $do19(client$, request, options) {
57067
+ const parsed$ = safeParse4(request, (value$) => V2EvalsDeleteRequest$zodSchema.parse(value$), "Input validation failed");
57068
+ if (!parsed$.ok) {
57069
+ return [parsed$, { status: "invalid" }];
57070
+ }
57071
+ const payload$ = parsed$.value;
57072
+ const body$ = null;
57073
+ const pathParams$ = {
57074
+ eval_id: encodeSimple("eval_id", payload$.eval_id, {
57075
+ explode: false,
57076
+ charEncoding: "percent"
57077
+ })
57078
+ };
57079
+ const path$ = pathToFunc("/v2/evals/{eval_id}")(pathParams$);
57080
+ const headers$ = new Headers(compactMap({
57081
+ Accept: "application/json"
57082
+ }));
57083
+ const securityInput = await extractSecurity(client$._options.security);
57084
+ const requestSecurity = resolveGlobalSecurity(securityInput);
57085
+ const context = {
57086
+ options: client$._options,
57087
+ baseURL: options?.serverURL ?? client$._baseURL ?? "",
57088
+ operationID: "v2_evals_delete",
57089
+ oAuth2Scopes: null,
57090
+ resolvedSecurity: requestSecurity,
57091
+ securitySource: client$._options.security,
57092
+ retryConfig: options?.retries || client$._options.retryConfig || { strategy: "none" },
57093
+ retryCodes: options?.retryCodes || [
57094
+ "429",
57095
+ "500",
57096
+ "502",
57097
+ "503",
57098
+ "504"
57099
+ ]
57100
+ };
57101
+ const requestRes = client$._createRequest(context, {
57102
+ security: requestSecurity,
57103
+ method: "DELETE",
57104
+ baseURL: options?.serverURL,
57105
+ path: path$,
57106
+ headers: headers$,
57107
+ body: body$,
57108
+ userAgent: client$._options.userAgent,
57109
+ timeoutMs: options?.timeoutMs || client$._options.timeoutMs || -1
57110
+ }, options);
57111
+ if (!requestRes.ok) {
57112
+ return [requestRes, { status: "invalid" }];
57113
+ }
57114
+ const req$ = requestRes.value;
57115
+ const doResult = await client$._do(req$, {
57116
+ context,
57117
+ errorCodes: [],
57118
+ retryConfig: context.retryConfig,
57119
+ retryCodes: context.retryCodes
57120
+ });
57121
+ if (!doResult.ok) {
57122
+ return [doResult, { status: "request-error", request: req$ }];
57123
+ }
57124
+ return [doResult, {
57125
+ status: "complete",
57126
+ request: req$,
57127
+ response: doResult.value
57128
+ }];
57129
+ }
57130
+ var init_evalsV2EvalsDelete = __esm(() => {
57131
+ init_encodings();
57132
+ init_primitives();
57133
+ init_schemas4();
57134
+ init_security();
57135
+ init_url();
57136
+ init_v2evalsdeleteop();
57137
+ init_async();
57138
+ });
57139
+
57140
+ // src/mcp-server/tools/evalsV2EvalsDelete.ts
57141
+ var args16, tool$evalsV2EvalsDelete;
57142
+ var init_evalsV2EvalsDelete2 = __esm(() => {
57143
+ init_evalsV2EvalsDelete();
57144
+ init_v2evalsdeleteop();
57145
+ init_tools();
57146
+ args16 = {
57147
+ request: V2EvalsDeleteRequest$zodSchema
57148
+ };
57149
+ tool$evalsV2EvalsDelete = {
57150
+ name: "evals-v2-evals-delete",
57151
+ description: `Delete Eval
57152
+
57153
+ Permanently deletes an eval and all of its saved configurations.`,
57154
+ annotations: {
57155
+ title: "",
57156
+ destructiveHint: true,
57157
+ idempotentHint: false,
57158
+ openWorldHint: false,
57159
+ readOnlyHint: false
57160
+ },
57161
+ args: args16,
57162
+ tool: async (client, args, ctx) => {
57163
+ const [result] = await evalsV2EvalsDelete(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
57164
+ if (!result.ok) {
57165
+ return {
57166
+ content: [{ type: "text", text: result.error.message }],
57167
+ isError: true
57168
+ };
57169
+ }
57170
+ return formatResult(result.value);
57171
+ }
57172
+ };
57173
+ });
57174
+
57175
+ // src/funcs/evalsV2EvalsList.ts
57176
+ function evalsV2EvalsList(client$, options) {
57177
+ return new APIPromise($do20(client$, options));
57178
+ }
57179
+ async function $do20(client$, options) {
57180
+ const path$ = pathToFunc("/v2/evals")();
57181
+ const headers$ = new Headers(compactMap({
57182
+ Accept: "application/json"
57183
+ }));
57184
+ const securityInput = await extractSecurity(client$._options.security);
57185
+ const requestSecurity = resolveGlobalSecurity(securityInput);
57186
+ const context = {
57187
+ options: client$._options,
57188
+ baseURL: options?.serverURL ?? client$._baseURL ?? "",
57189
+ operationID: "v2_evals_list",
57190
+ oAuth2Scopes: null,
57191
+ resolvedSecurity: requestSecurity,
57192
+ securitySource: client$._options.security,
57193
+ retryConfig: options?.retries || client$._options.retryConfig || { strategy: "none" },
57194
+ retryCodes: options?.retryCodes || [
57195
+ "429",
57196
+ "500",
57197
+ "502",
57198
+ "503",
57199
+ "504"
57200
+ ]
57201
+ };
57202
+ const requestRes = client$._createRequest(context, {
57203
+ security: requestSecurity,
57204
+ method: "GET",
57205
+ baseURL: options?.serverURL,
57206
+ path: path$,
57207
+ headers: headers$,
57208
+ userAgent: client$._options.userAgent,
57209
+ timeoutMs: options?.timeoutMs || client$._options.timeoutMs || -1
57210
+ }, options);
57211
+ if (!requestRes.ok) {
57212
+ return [requestRes, { status: "invalid" }];
57213
+ }
57214
+ const req$ = requestRes.value;
57215
+ const doResult = await client$._do(req$, {
57216
+ context,
57217
+ errorCodes: [],
57218
+ retryConfig: context.retryConfig,
57219
+ retryCodes: context.retryCodes
57220
+ });
57221
+ if (!doResult.ok) {
57222
+ return [doResult, { status: "request-error", request: req$ }];
57223
+ }
57224
+ return [doResult, {
57225
+ status: "complete",
57226
+ request: req$,
57227
+ response: doResult.value
57228
+ }];
57229
+ }
57230
+ var init_evalsV2EvalsList = __esm(() => {
57231
+ init_primitives();
57232
+ init_security();
57233
+ init_url();
57234
+ init_async();
57235
+ });
57236
+
57237
+ // src/mcp-server/tools/evalsV2EvalsList.ts
57238
+ var tool$evalsV2EvalsList;
57239
+ var init_evalsV2EvalsList2 = __esm(() => {
57240
+ init_evalsV2EvalsList();
57241
+ init_tools();
57242
+ tool$evalsV2EvalsList = {
57243
+ name: "evals-v2-evals-list",
57244
+ description: `List Evals
57245
+
57246
+ Returns all active eval definitions owned by the organization.`,
57247
+ annotations: {
57248
+ title: "",
57249
+ destructiveHint: false,
57250
+ idempotentHint: false,
57251
+ openWorldHint: false,
57252
+ readOnlyHint: true
57253
+ },
57254
+ tool: async (client, ctx) => {
57255
+ const [result] = await evalsV2EvalsList(client, { fetchOptions: { signal: ctx.signal } }).$inspect();
57256
+ if (!result.ok) {
57257
+ return {
57258
+ content: [{ type: "text", text: result.error.message }],
57259
+ isError: true
57260
+ };
57261
+ }
57262
+ return formatResult(result.value);
57263
+ }
57264
+ };
57265
+ });
57266
+
57267
+ // src/models/v2evalsretrieveop.ts
57268
+ var V2EvalsRetrieveRequest$zodSchema, V2EvalsRetrieveResponse$zodSchema;
57269
+ var init_v2evalsretrieveop = __esm(() => {
57270
+ init_zod();
57271
+ init_httpvalidationerror();
57272
+ init_v2eval();
57273
+ V2EvalsRetrieveRequest$zodSchema = object({
57274
+ eval_id: string2()
57275
+ });
57276
+ V2EvalsRetrieveResponse$zodSchema = union([
57277
+ V2Eval$zodSchema,
57278
+ HTTPValidationError$zodSchema
57279
+ ]);
57280
+ });
57281
+
57282
+ // src/funcs/evalsV2EvalsRetrieve.ts
57283
+ function evalsV2EvalsRetrieve(client$, request, options) {
57284
+ return new APIPromise($do21(client$, request, options));
57285
+ }
57286
+ async function $do21(client$, request, options) {
57287
+ const parsed$ = safeParse4(request, (value$) => V2EvalsRetrieveRequest$zodSchema.parse(value$), "Input validation failed");
57288
+ if (!parsed$.ok) {
57289
+ return [parsed$, { status: "invalid" }];
57290
+ }
57291
+ const payload$ = parsed$.value;
57292
+ const body$ = null;
57293
+ const pathParams$ = {
57294
+ eval_id: encodeSimple("eval_id", payload$.eval_id, {
57295
+ explode: false,
57296
+ charEncoding: "percent"
57297
+ })
57298
+ };
57299
+ const path$ = pathToFunc("/v2/evals/{eval_id}")(pathParams$);
57300
+ const headers$ = new Headers(compactMap({
57301
+ Accept: "application/json"
57302
+ }));
57303
+ const securityInput = await extractSecurity(client$._options.security);
57304
+ const requestSecurity = resolveGlobalSecurity(securityInput);
57305
+ const context = {
57306
+ options: client$._options,
57307
+ baseURL: options?.serverURL ?? client$._baseURL ?? "",
57308
+ operationID: "v2_evals_retrieve",
57309
+ oAuth2Scopes: null,
57310
+ resolvedSecurity: requestSecurity,
57311
+ securitySource: client$._options.security,
57312
+ retryConfig: options?.retries || client$._options.retryConfig || { strategy: "none" },
57313
+ retryCodes: options?.retryCodes || [
57314
+ "429",
57315
+ "500",
57316
+ "502",
57317
+ "503",
57318
+ "504"
57319
+ ]
57320
+ };
57321
+ const requestRes = client$._createRequest(context, {
57322
+ security: requestSecurity,
57323
+ method: "GET",
57324
+ baseURL: options?.serverURL,
57325
+ path: path$,
57326
+ headers: headers$,
57327
+ body: body$,
57328
+ userAgent: client$._options.userAgent,
57329
+ timeoutMs: options?.timeoutMs || client$._options.timeoutMs || -1
57330
+ }, options);
57331
+ if (!requestRes.ok) {
57332
+ return [requestRes, { status: "invalid" }];
57333
+ }
57334
+ const req$ = requestRes.value;
57335
+ const doResult = await client$._do(req$, {
57336
+ context,
57337
+ errorCodes: [],
57338
+ retryConfig: context.retryConfig,
57339
+ retryCodes: context.retryCodes
57340
+ });
57341
+ if (!doResult.ok) {
57342
+ return [doResult, { status: "request-error", request: req$ }];
57343
+ }
57344
+ return [doResult, {
57345
+ status: "complete",
57346
+ request: req$,
57347
+ response: doResult.value
57348
+ }];
57349
+ }
57350
+ var init_evalsV2EvalsRetrieve = __esm(() => {
56651
57351
  init_encodings();
56652
57352
  init_primitives();
56653
57353
  init_schemas4();
56654
57354
  init_security();
56655
57355
  init_url();
56656
- init_v2evalconfigurationsretrieveop();
57356
+ init_v2evalsretrieveop();
56657
57357
  init_async();
56658
57358
  });
56659
57359
 
56660
- // src/mcp-server/tools/evalsV2EvalConfigurationsRetrieve.ts
56661
- var args14, tool$evalsV2EvalConfigurationsRetrieve;
56662
- var init_evalsV2EvalConfigurationsRetrieve2 = __esm(() => {
56663
- init_evalsV2EvalConfigurationsRetrieve();
56664
- init_v2evalconfigurationsretrieveop();
57360
+ // src/mcp-server/tools/evalsV2EvalsRetrieve.ts
57361
+ var args17, tool$evalsV2EvalsRetrieve;
57362
+ var init_evalsV2EvalsRetrieve2 = __esm(() => {
57363
+ init_evalsV2EvalsRetrieve();
57364
+ init_v2evalsretrieveop();
56665
57365
  init_tools();
56666
- args14 = {
56667
- request: V2EvalConfigurationsRetrieveRequest$zodSchema
57366
+ args17 = {
57367
+ request: V2EvalsRetrieveRequest$zodSchema
56668
57368
  };
56669
- tool$evalsV2EvalConfigurationsRetrieve = {
56670
- name: "evals-v2-eval-configurations-retrieve",
56671
- description: `Retrieve Configuration
57369
+ tool$evalsV2EvalsRetrieve = {
57370
+ name: "evals-v2-evals-retrieve",
57371
+ description: `Retrieve Eval
56672
57372
 
56673
- Returns one saved configuration for an eval.`,
57373
+ Returns one reusable eval simulation definition.`,
56674
57374
  annotations: {
56675
57375
  title: "",
56676
57376
  destructiveHint: false,
@@ -56678,9 +57378,9 @@ Returns one saved configuration for an eval.`,
56678
57378
  openWorldHint: false,
56679
57379
  readOnlyHint: true
56680
57380
  },
56681
- args: args14,
57381
+ args: args17,
56682
57382
  tool: async (client, args, ctx) => {
56683
- const [result] = await evalsV2EvalConfigurationsRetrieve(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
57383
+ const [result] = await evalsV2EvalsRetrieve(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
56684
57384
  if (!result.ok) {
56685
57385
  return {
56686
57386
  content: [{ type: "text", text: result.error.message }],
@@ -56692,59 +57392,71 @@ Returns one saved configuration for an eval.`,
56692
57392
  };
56693
57393
  });
56694
57394
 
56695
- // src/models/updatev2evalconfigurationrequest.ts
56696
- var UpdateV2EvalConfigurationRequest$zodSchema;
56697
- var init_updatev2evalconfigurationrequest = __esm(() => {
57395
+ // src/models/updatev2evalrequest.ts
57396
+ var UpdateV2EvalRequestMcp$zodSchema, UpdateV2EvalRequest$zodSchema;
57397
+ var init_updatev2evalrequest = __esm(() => {
56698
57398
  init_zod();
56699
- UpdateV2EvalConfigurationRequest$zodSchema = object({
56700
- cli_ids: array(string2()).nullable().optional(),
56701
- environment_id: string2().optional(),
56702
- include_docs: boolean2().optional(),
56703
- mcp_server_ids: array(string2()).nullable().optional(),
57399
+ init_searchgroundingmode();
57400
+ init_v2inlinemcpdefinition();
57401
+ init_v2savedmcpreference();
57402
+ UpdateV2EvalRequestMcp$zodSchema = union([
57403
+ V2InlineMcpDefinition$zodSchema,
57404
+ V2SavedMcpReference$zodSchema
57405
+ ]);
57406
+ UpdateV2EvalRequest$zodSchema = object({
57407
+ agent: array(string2()).nullable().optional(),
57408
+ cleanup_commands: array(string2()).nullable().optional().describe("Replacement cleanup commands. Pass an empty array to clear all saved cleanup commands."),
57409
+ clis: array(string2()).nullable().optional().describe("Replacement saved CLI IDs or shell installation commands. UUID values resolve saved CLIs; every other value is an installation command. Pass an empty array to clear all CLIs."),
57410
+ judges: array(string2()).nullable().optional().describe("Replacement success criteria for the eval."),
57411
+ mcps: array(union([
57412
+ V2InlineMcpDefinition$zodSchema,
57413
+ V2SavedMcpReference$zodSchema
57414
+ ])).nullable().optional(),
56704
57415
  name: string2().optional(),
56705
- repository_id: string2().optional(),
56706
- skill_ids: array(string2()).nullable().optional(),
56707
- skills: array(string2()).nullable().optional().describe("Public skill packages saved with this configuration, using GitHub `owner/repository` references.")
57416
+ persona: string2().optional().describe("Replacement eval persona. Pass null to clear it; omit it to leave the saved value unchanged."),
57417
+ prompts: array(string2()).nullable().optional(),
57418
+ runs: int().optional(),
57419
+ search_grounding: SearchGroundingMode$zodSchema.nullable().optional().describe("Replacement web-retrieval policy. Omit it to keep the saved value."),
57420
+ setup_commands: array(string2()).nullable().optional().describe("Replacement setup commands. Pass an empty array to clear all saved setup commands."),
57421
+ tags: array(string2()).nullable().optional().describe("Replacement labels for the eval. Pass an empty array to clear all tags.")
56708
57422
  });
56709
57423
  });
56710
57424
 
56711
- // src/models/v2evalconfigurationsupdateop.ts
56712
- var V2EvalConfigurationsUpdateRequest$zodSchema, V2EvalConfigurationsUpdateResponse$zodSchema;
56713
- var init_v2evalconfigurationsupdateop = __esm(() => {
57425
+ // src/models/v2evalsupdateop.ts
57426
+ var V2EvalsUpdateRequest$zodSchema, V2EvalsUpdateResponse$zodSchema;
57427
+ var init_v2evalsupdateop = __esm(() => {
56714
57428
  init_zod();
56715
57429
  init_httpvalidationerror();
56716
- init_updatev2evalconfigurationrequest();
56717
- init_v2evalconfiguration();
56718
- V2EvalConfigurationsUpdateRequest$zodSchema = object({
56719
- body: UpdateV2EvalConfigurationRequest$zodSchema,
56720
- configuration_id: string2(),
57430
+ init_updatev2evalrequest();
57431
+ init_v2eval();
57432
+ V2EvalsUpdateRequest$zodSchema = object({
57433
+ body: UpdateV2EvalRequest$zodSchema,
56721
57434
  eval_id: string2()
56722
57435
  });
56723
- V2EvalConfigurationsUpdateResponse$zodSchema = union([
56724
- V2EvalConfiguration$zodSchema,
57436
+ V2EvalsUpdateResponse$zodSchema = union([
57437
+ V2Eval$zodSchema,
56725
57438
  HTTPValidationError$zodSchema
56726
57439
  ]);
56727
57440
  });
56728
57441
 
56729
- // src/funcs/evalsV2EvalConfigurationsUpdate.ts
56730
- function evalsV2EvalConfigurationsUpdate(client$, request, options) {
56731
- return new APIPromise($do18(client$, request, options));
57442
+ // src/funcs/evalsV2EvalsUpdate.ts
57443
+ function evalsV2EvalsUpdate(client$, request, options) {
57444
+ return new APIPromise($do22(client$, request, options));
56732
57445
  }
56733
- async function $do18(client$, request, options) {
56734
- const parsed$ = safeParse4(request, (value$) => V2EvalConfigurationsUpdateRequest$zodSchema.parse(value$), "Input validation failed");
57446
+ async function $do22(client$, request, options) {
57447
+ const parsed$ = safeParse4(request, (value$) => V2EvalsUpdateRequest$zodSchema.parse(value$), "Input validation failed");
56735
57448
  if (!parsed$.ok) {
56736
57449
  return [parsed$, { status: "invalid" }];
56737
57450
  }
56738
57451
  const payload$ = parsed$.value;
56739
57452
  const body$ = encodeJSON("body", payload$.body, { explode: true });
56740
57453
  const pathParams$ = {
56741
- configuration_id: encodeSimple("configuration_id", payload$.configuration_id, { explode: false, charEncoding: "percent" }),
56742
57454
  eval_id: encodeSimple("eval_id", payload$.eval_id, {
56743
57455
  explode: false,
56744
57456
  charEncoding: "percent"
56745
57457
  })
56746
57458
  };
56747
- const path$ = pathToFunc("/v2/evals/{eval_id}/configurations/{configuration_id}")(pathParams$);
57459
+ const path$ = pathToFunc("/v2/evals/{eval_id}")(pathParams$);
56748
57460
  const headers$ = new Headers(compactMap({
56749
57461
  "Content-Type": "application/json",
56750
57462
  Accept: "application/json"
@@ -56754,7 +57466,7 @@ async function $do18(client$, request, options) {
56754
57466
  const context = {
56755
57467
  options: client$._options,
56756
57468
  baseURL: options?.serverURL ?? client$._baseURL ?? "",
56757
- operationID: "v2_eval_configurations_update",
57469
+ operationID: "v2_evals_update",
56758
57470
  oAuth2Scopes: null,
56759
57471
  resolvedSecurity: requestSecurity,
56760
57472
  securitySource: client$._options.security,
@@ -56796,30 +57508,30 @@ async function $do18(client$, request, options) {
56796
57508
  response: doResult.value
56797
57509
  }];
56798
57510
  }
56799
- var init_evalsV2EvalConfigurationsUpdate = __esm(() => {
57511
+ var init_evalsV2EvalsUpdate = __esm(() => {
56800
57512
  init_encodings();
56801
57513
  init_primitives();
56802
57514
  init_schemas4();
56803
57515
  init_security();
56804
57516
  init_url();
56805
- init_v2evalconfigurationsupdateop();
57517
+ init_v2evalsupdateop();
56806
57518
  init_async();
56807
57519
  });
56808
57520
 
56809
- // src/mcp-server/tools/evalsV2EvalConfigurationsUpdate.ts
56810
- var args15, tool$evalsV2EvalConfigurationsUpdate;
56811
- var init_evalsV2EvalConfigurationsUpdate2 = __esm(() => {
56812
- init_evalsV2EvalConfigurationsUpdate();
56813
- init_v2evalconfigurationsupdateop();
57521
+ // src/mcp-server/tools/evalsV2EvalsUpdate.ts
57522
+ var args18, tool$evalsV2EvalsUpdate;
57523
+ var init_evalsV2EvalsUpdate2 = __esm(() => {
57524
+ init_evalsV2EvalsUpdate();
57525
+ init_v2evalsupdateop();
56814
57526
  init_tools();
56815
- args15 = {
56816
- request: V2EvalConfigurationsUpdateRequest$zodSchema
57527
+ args18 = {
57528
+ request: V2EvalsUpdateRequest$zodSchema
56817
57529
  };
56818
- tool$evalsV2EvalConfigurationsUpdate = {
56819
- name: "evals-v2-eval-configurations-update",
56820
- description: `Update Configuration
57530
+ tool$evalsV2EvalsUpdate = {
57531
+ name: "evals-v2-evals-update",
57532
+ description: `Update Eval
56821
57533
 
56822
- Updates an eval configuration; omitted fields remain unchanged.`,
57534
+ Updates a reusable eval simulation definition. Omitted fields remain unchanged. Empty \`agent\`, \`clis\`, \`mcps\`, \`setup_commands\`, \`cleanup_commands\`, or \`tags\` arrays clear those collections, and \`persona: null\` clears the saved persona.`,
56823
57535
  annotations: {
56824
57536
  title: "",
56825
57537
  destructiveHint: false,
@@ -56827,9 +57539,9 @@ Updates an eval configuration; omitted fields remain unchanged.`,
56827
57539
  openWorldHint: false,
56828
57540
  readOnlyHint: false
56829
57541
  },
56830
- args: args15,
57542
+ args: args18,
56831
57543
  tool: async (client, args, ctx) => {
56832
- const [result] = await evalsV2EvalConfigurationsUpdate(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
57544
+ const [result] = await evalsV2EvalsUpdate(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
56833
57545
  if (!result.ok) {
56834
57546
  return {
56835
57547
  content: [{ type: "text", text: result.error.message }],
@@ -56841,53 +57553,117 @@ Updates an eval configuration; omitted fields remain unchanged.`,
56841
57553
  };
56842
57554
  });
56843
57555
 
56844
- // src/models/createv2evalrequest.ts
56845
- var CreateV2EvalRequestCli$zodSchema, CreateV2EvalRequestMcp$zodSchema, CreateV2EvalRequest$zodSchema;
56846
- var init_createv2evalrequest = __esm(() => {
57556
+ // src/models/evaltemplateruntime.ts
57557
+ var EvalTemplateRuntime$zodSchema;
57558
+ var init_evaltemplateruntime = __esm(() => {
56847
57559
  init_zod();
56848
- init_v2inlineclireference();
56849
- init_v2inlinemcpdefinition();
56850
- init_v2savedclireference();
56851
- init_v2savedmcpreference();
56852
- CreateV2EvalRequestCli$zodSchema = union([
56853
- V2SavedCliReference$zodSchema,
56854
- V2InlineCliReference$zodSchema
56855
- ]);
56856
- CreateV2EvalRequestMcp$zodSchema = union([
56857
- V2InlineMcpDefinition$zodSchema,
56858
- V2SavedMcpReference$zodSchema
56859
- ]);
56860
- CreateV2EvalRequest$zodSchema = object({
56861
- agent: array(string2()).optional().describe("Runnable agents from `GET /v2/agents`. An empty array saves the eval without default agents. CLI clients expose repeatable `--agent`."),
56862
- clis: array(union([
56863
- V2SavedCliReference$zodSchema,
56864
- V2InlineCliReference$zodSchema
56865
- ])).optional().describe("Saved CLI references or inline installation commands."),
56866
- judges: array(string2()).describe("Plain-text success criteria. CLI clients expose this as repeatable `--judge`."),
56867
- mcps: array(union([
56868
- V2InlineMcpDefinition$zodSchema,
56869
- V2SavedMcpReference$zodSchema
56870
- ])).optional().describe("Saved MCP references or inline portable MCP definitions."),
57560
+ init_searchgroundingmode();
57561
+ EvalTemplateRuntime$zodSchema = object({
57562
+ cleanup_commands: array(string2()).optional().describe("Shell commands run after each eval attempt, in order."),
57563
+ cli_ids: array(string2()).optional().describe("Reusable CLI IDs copied into each created eval."),
57564
+ environment_profile_ids: array(string2()).optional().describe("Environment profile IDs copied into each created eval."),
57565
+ include_brand_docs: boolean2().default(false).describe("Whether organization documentation is included at runtime."),
57566
+ include_env_vars: boolean2().default(true).describe("Whether selected environment profiles are enabled."),
57567
+ mcp_server_ids: array(string2()).optional().describe("Reusable MCP server IDs copied into each created eval."),
57568
+ run_on_github_pr: boolean2().default(false).describe("Whether each created eval is configured for GitHub pull requests."),
57569
+ search_grounding: SearchGroundingMode$zodSchema.optional().describe("Requested web-retrieval behavior for an eval run."),
57570
+ setup_commands: array(string2()).optional().describe("Shell commands run before each eval attempt, in order."),
57571
+ skill_ids: array(string2()).optional().describe("Reusable skill IDs copied into each created eval."),
57572
+ starting_project_id: string2().nullable().optional().describe("Optional reusable repository workspace ID.")
57573
+ }).describe("Reusable runtime resources copied into each created eval.");
57574
+ });
57575
+
57576
+ // src/models/evaltemplateschedule.ts
57577
+ var Frequency$zodSchema, EvalTemplateSchedule$zodSchema;
57578
+ var init_evaltemplateschedule = __esm(() => {
57579
+ init_zod();
57580
+ Frequency$zodSchema = _enum([
57581
+ "daily",
57582
+ "every_other_day",
57583
+ "weekly"
57584
+ ]).describe("Schedule cadence used when `enabled` is true.");
57585
+ EvalTemplateSchedule$zodSchema = object({
57586
+ enabled: boolean2().default(false).describe("Whether each eval created from the template receives a schedule."),
57587
+ frequency: Frequency$zodSchema.default("weekly").describe("Schedule cadence used when `enabled` is true.")
57588
+ }).describe("Schedule copied into every eval created from the template.");
57589
+ });
57590
+
57591
+ // src/models/graderdefinition.ts
57592
+ var GraderDefinition$zodSchema;
57593
+ var init_graderdefinition = __esm(() => {
57594
+ init_zod();
57595
+ GraderDefinition$zodSchema = object({
57596
+ blocking: boolean2().default(false).describe("Whether a failed or missing result from this grader fails the entire eval regardless of the aggregate score."),
57597
+ case_sensitive: boolean2().default(false),
57598
+ criteria: string2().nullable().optional(),
56871
57599
  name: string2(),
56872
- persona: string2().optional().describe("Optional perspective applied to the whole eval. It is stored as metadata and does not change prompts or multiply attempts."),
56873
- prompts: array(string2()).describe("Task prompts. CLI clients expose this as repeatable `--prompt`."),
56874
- runs: int().default(1).describe("Independent attempts for every prompt-agent pair."),
56875
- tags: array(string2()).optional().describe("Optional labels for organizing the eval. CLI clients may expose this string array as a repeatable `--tag` flag.")
57600
+ needle: string2().nullable().optional(),
57601
+ needles: array(string2()).nullable().optional(),
57602
+ pattern: string2().nullable().optional(),
57603
+ type: string2(),
57604
+ weight: number2().default(1)
56876
57605
  });
56877
57606
  });
56878
57607
 
56879
- // src/funcs/evalsV2EvalsCreate.ts
56880
- function evalsV2EvalsCreate(client$, request, options) {
56881
- return new APIPromise($do19(client$, request, options));
57608
+ // src/models/createevaltemplaterequest.ts
57609
+ var CreateEvalTemplateRequestEvalType$zodSchema, CreateEvalTemplateRequestIncludedComponent$zodSchema, CreateEvalTemplateRequest$zodSchema;
57610
+ var init_createevaltemplaterequest = __esm(() => {
57611
+ init_zod();
57612
+ init_evaltemplateruntime();
57613
+ init_evaltemplateschedule();
57614
+ init_graderdefinition();
57615
+ CreateEvalTemplateRequestEvalType$zodSchema = _enum([
57616
+ "custom_eval",
57617
+ "agent_preference"
57618
+ ]).describe("Kind of eval the template creates. A template only applies to evals of the same type.");
57619
+ CreateEvalTemplateRequestIncludedComponent$zodSchema = _enum([
57620
+ "judges",
57621
+ "targets",
57622
+ "persona",
57623
+ "competitors",
57624
+ "tag",
57625
+ "category",
57626
+ "runs",
57627
+ "schedule",
57628
+ "repository",
57629
+ "environment",
57630
+ "search_grounding",
57631
+ "skills",
57632
+ "mcp_servers",
57633
+ "clis",
57634
+ "brand_docs",
57635
+ "github_pr",
57636
+ "runtime_commands"
57637
+ ]);
57638
+ CreateEvalTemplateRequest$zodSchema = object({
57639
+ category_name: string2().optional().describe("Optional category copied when `category` is included."),
57640
+ competitor_ids: array(string2()).optional().describe("Competitor IDs that create additional variants when `competitors` is included."),
57641
+ description: string2().optional().describe("Optional guidance describing when to use the template."),
57642
+ eval_type: CreateEvalTemplateRequestEvalType$zodSchema.default("custom_eval").describe("Kind of eval the template creates. A template only applies to evals of the same type."),
57643
+ included_components: array(CreateEvalTemplateRequestIncludedComponent$zodSchema).describe("Settings copied when the template is applied. Omitted components do not override the new eval; included components copy their saved value, including an empty value."),
57644
+ judges: array(GraderDefinition$zodSchema).optional().describe("Judge definitions copied when `judges` is included. May be empty to explicitly create the eval without template judges."),
57645
+ name: string2().describe("Organization-visible name for this reusable template."),
57646
+ persona: string2().optional().describe("Optional persona copied when `persona` is included."),
57647
+ runs_per_execution: int().default(1).describe("Attempt count copied when `runs` is included."),
57648
+ runtime: EvalTemplateRuntime$zodSchema.optional().describe("Reusable runtime resources copied into each created eval."),
57649
+ schedule: EvalTemplateSchedule$zodSchema.optional().describe("Schedule copied into every eval created from the template."),
57650
+ tag: string2().optional().describe("Optional tag copied when `tag` is included."),
57651
+ target_keys: array(string2()).optional().describe("Runnable target keys copied when `targets` is included, such as `agent:codex:gpt-5.6-terra`.")
57652
+ }).describe("Create reusable configuration before supplying prompt-specific evals.");
57653
+ });
57654
+
57655
+ // src/funcs/evalTemplatesV2EvalTemplatesCreate.ts
57656
+ function evalTemplatesV2EvalTemplatesCreate(client$, request, options) {
57657
+ return new APIPromise($do23(client$, request, options));
56882
57658
  }
56883
- async function $do19(client$, request, options) {
56884
- const parsed$ = safeParse4(request, (value$) => CreateV2EvalRequest$zodSchema.parse(value$), "Input validation failed");
57659
+ async function $do23(client$, request, options) {
57660
+ const parsed$ = safeParse4(request, (value$) => CreateEvalTemplateRequest$zodSchema.parse(value$), "Input validation failed");
56885
57661
  if (!parsed$.ok) {
56886
57662
  return [parsed$, { status: "invalid" }];
56887
57663
  }
56888
57664
  const payload$ = parsed$.value;
56889
57665
  const body$ = encodeJSON("body", payload$, { explode: true });
56890
- const path$ = pathToFunc("/v2/evals")();
57666
+ const path$ = pathToFunc("/v2/eval-templates")();
56891
57667
  const headers$ = new Headers(compactMap({
56892
57668
  "Content-Type": "application/json",
56893
57669
  Accept: "application/json"
@@ -56897,7 +57673,7 @@ async function $do19(client$, request, options) {
56897
57673
  const context = {
56898
57674
  options: client$._options,
56899
57675
  baseURL: options?.serverURL ?? client$._baseURL ?? "",
56900
- operationID: "v2_evals_create",
57676
+ operationID: "v2_eval_templates_create",
56901
57677
  oAuth2Scopes: null,
56902
57678
  resolvedSecurity: requestSecurity,
56903
57679
  securitySource: client$._options.security,
@@ -56939,30 +57715,30 @@ async function $do19(client$, request, options) {
56939
57715
  response: doResult.value
56940
57716
  }];
56941
57717
  }
56942
- var init_evalsV2EvalsCreate = __esm(() => {
57718
+ var init_evalTemplatesV2EvalTemplatesCreate = __esm(() => {
56943
57719
  init_encodings();
56944
57720
  init_primitives();
56945
57721
  init_schemas4();
56946
57722
  init_security();
56947
57723
  init_url();
56948
- init_createv2evalrequest();
57724
+ init_createevaltemplaterequest();
56949
57725
  init_async();
56950
57726
  });
56951
57727
 
56952
- // src/mcp-server/tools/evalsV2EvalsCreate.ts
56953
- var args16, tool$evalsV2EvalsCreate;
56954
- var init_evalsV2EvalsCreate2 = __esm(() => {
56955
- init_evalsV2EvalsCreate();
56956
- init_createv2evalrequest();
57728
+ // src/mcp-server/tools/evalTemplatesV2EvalTemplatesCreate.ts
57729
+ var args19, tool$evalTemplatesV2EvalTemplatesCreate;
57730
+ var init_evalTemplatesV2EvalTemplatesCreate2 = __esm(() => {
57731
+ init_evalTemplatesV2EvalTemplatesCreate();
57732
+ init_createevaltemplaterequest();
56957
57733
  init_tools();
56958
- args16 = {
56959
- request: CreateV2EvalRequest$zodSchema
57734
+ args19 = {
57735
+ request: CreateEvalTemplateRequest$zodSchema
56960
57736
  };
56961
- tool$evalsV2EvalsCreate = {
56962
- name: "evals-v2-evals-create",
56963
- description: `Create Eval
57737
+ tool$evalTemplatesV2EvalTemplatesCreate = {
57738
+ name: "eval-templates-v2-eval-templates-create",
57739
+ description: `Create an Eval template
56964
57740
 
56965
- Creates a reusable simulation definition with one or more task prompts and judges. Optional agents, persona, CLI installations, MCP servers, repeat count, and tags are saved with the eval. Creating the eval does not start a run and does not accept secret environment values.`,
57741
+ Step 1 of the reusable template workflow. Creates an organization-owned Eval template containing only the configuration components selected in \`included_components\`. Do not include task prompts in the template. The response returns the template \`id\` and \`version\`; pass them as \`template_id\` and \`expected_template_version\` to \`POST /v2/evals/batch\` with each eval's name and prompts. Applying a template copies its values into the new evals without creating a live link.`,
56966
57742
  annotations: {
56967
57743
  title: "",
56968
57744
  destructiveHint: false,
@@ -56970,9 +57746,9 @@ Creates a reusable simulation definition with one or more task prompts and judge
56970
57746
  openWorldHint: false,
56971
57747
  readOnlyHint: false
56972
57748
  },
56973
- args: args16,
57749
+ args: args19,
56974
57750
  tool: async (client, args, ctx) => {
56975
- const [result] = await evalsV2EvalsCreate(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
57751
+ const [result] = await evalTemplatesV2EvalTemplatesCreate(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
56976
57752
  if (!result.ok) {
56977
57753
  return {
56978
57754
  content: [{ type: "text", text: result.error.message }],
@@ -56984,77 +57760,101 @@ Creates a reusable simulation definition with one or more task prompts and judge
56984
57760
  };
56985
57761
  });
56986
57762
 
56987
- // src/models/v2eval.ts
56988
- var V2EvalCli$zodSchema, V2EvalMcp$zodSchema, V2Eval$zodSchema;
56989
- var init_v2eval = __esm(() => {
57763
+ // src/models/evaltemplate.ts
57764
+ var EvalTemplateEvalType$zodSchema, EvalTemplateIncludedComponent$zodSchema, EvalTemplate$zodSchema;
57765
+ var init_evaltemplate = __esm(() => {
56990
57766
  init_zod();
56991
- init_v2inlineclireference();
56992
- init_v2inlinemcpdefinition();
56993
- init_v2savedclireference();
56994
- init_v2savedmcpreference();
56995
- V2EvalCli$zodSchema = union([
56996
- V2SavedCliReference$zodSchema,
56997
- V2InlineCliReference$zodSchema
56998
- ]);
56999
- V2EvalMcp$zodSchema = union([
57000
- V2InlineMcpDefinition$zodSchema,
57001
- V2SavedMcpReference$zodSchema
57767
+ init_evaltemplateruntime();
57768
+ init_evaltemplateschedule();
57769
+ init_graderdefinition();
57770
+ EvalTemplateEvalType$zodSchema = _enum([
57771
+ "custom_eval",
57772
+ "agent_preference"
57773
+ ]).describe("Kind of eval the template creates. A template only applies to evals of the same type.");
57774
+ EvalTemplateIncludedComponent$zodSchema = _enum([
57775
+ "judges",
57776
+ "targets",
57777
+ "persona",
57778
+ "competitors",
57779
+ "tag",
57780
+ "category",
57781
+ "runs",
57782
+ "schedule",
57783
+ "repository",
57784
+ "environment",
57785
+ "search_grounding",
57786
+ "skills",
57787
+ "mcp_servers",
57788
+ "clis",
57789
+ "brand_docs",
57790
+ "github_pr",
57791
+ "runtime_commands"
57002
57792
  ]);
57003
- V2Eval$zodSchema = object({
57004
- agent: array(string2()).optional(),
57005
- clis: array(union([
57006
- V2SavedCliReference$zodSchema,
57007
- V2InlineCliReference$zodSchema
57008
- ])).optional(),
57009
- created_at: string2().nullable().optional(),
57793
+ EvalTemplate$zodSchema = object({
57794
+ category_name: string2().nullable().optional().describe("Optional category copied when `category` is included."),
57795
+ competitor_ids: array(string2()).optional().describe("Competitor IDs that create additional variants when `competitors` is included."),
57796
+ created_at: datetime2({ offset: true }),
57797
+ created_by: string2().nullable().optional(),
57798
+ deleted_at: datetime2({ offset: true }).nullable().optional(),
57799
+ description: string2().nullable().optional().describe("Optional guidance describing when to use the template."),
57800
+ eval_type: EvalTemplateEvalType$zodSchema.default("custom_eval").describe("Kind of eval the template creates. A template only applies to evals of the same type."),
57010
57801
  id: string2(),
57011
- judges: array(string2()).optional(),
57012
- mcps: array(union([
57013
- V2InlineMcpDefinition$zodSchema,
57014
- V2SavedMcpReference$zodSchema
57015
- ])).optional(),
57016
- name: string2(),
57017
- persona: string2().nullable().optional(),
57018
- prompts: array(string2()).optional(),
57019
- runs: int().default(1),
57020
- tags: array(string2()).optional(),
57021
- updated_at: string2().nullable().optional()
57022
- });
57023
- });
57024
-
57025
- // src/models/v2evalsdeleteop.ts
57026
- var V2EvalsDeleteRequest$zodSchema, V2EvalsDeleteResponse$zodSchema;
57027
- var init_v2evalsdeleteop = __esm(() => {
57802
+ included_components: array(EvalTemplateIncludedComponent$zodSchema).describe("Settings copied when the template is applied. Omitted components do not override the new eval; included components copy their saved value, including an empty value."),
57803
+ judges: array(GraderDefinition$zodSchema).optional().describe("Judge definitions copied when `judges` is included. May be empty to explicitly create the eval without template judges."),
57804
+ name: string2().describe("Organization-visible name for this reusable template."),
57805
+ org_id: string2(),
57806
+ persona: string2().nullable().optional().describe("Optional persona copied when `persona` is included."),
57807
+ runs_per_execution: int().default(1).describe("Attempt count copied when `runs` is included."),
57808
+ runtime: EvalTemplateRuntime$zodSchema.optional().describe("Reusable runtime resources copied into each created eval."),
57809
+ schedule: EvalTemplateSchedule$zodSchema.optional().describe("Schedule copied into every eval created from the template."),
57810
+ tag: string2().nullable().optional().describe("Optional tag copied when `tag` is included."),
57811
+ target_keys: array(string2()).optional().describe("Runnable target keys copied when `targets` is included, such as `agent:codex:gpt-5.6-terra`."),
57812
+ updated_at: datetime2({ offset: true }),
57813
+ version: int()
57814
+ });
57815
+ });
57816
+
57817
+ // src/models/v2evaltemplatesdeleteop.ts
57818
+ var V2EvalTemplatesDeleteRequest$zodSchema, V2EvalTemplatesDeleteConflictResponseBody$zodSchema, V2EvalTemplatesDeleteNotFoundResponseBody$zodSchema, V2EvalTemplatesDeleteResponse$zodSchema;
57819
+ var init_v2evaltemplatesdeleteop = __esm(() => {
57028
57820
  init_zod();
57821
+ init_evaltemplate();
57029
57822
  init_httpvalidationerror();
57030
- init_v2eval();
57031
- V2EvalsDeleteRequest$zodSchema = object({
57032
- eval_id: string2()
57823
+ V2EvalTemplatesDeleteRequest$zodSchema = object({
57824
+ template_id: string2()
57033
57825
  });
57034
- V2EvalsDeleteResponse$zodSchema = union([
57035
- V2Eval$zodSchema,
57826
+ V2EvalTemplatesDeleteConflictResponseBody$zodSchema = object({
57827
+ detail: string2()
57828
+ }).describe("The supplied template version is stale.");
57829
+ V2EvalTemplatesDeleteNotFoundResponseBody$zodSchema = object({
57830
+ detail: string2()
57831
+ }).describe("The Eval template does not exist in the organization.");
57832
+ V2EvalTemplatesDeleteResponse$zodSchema = union([
57833
+ EvalTemplate$zodSchema,
57834
+ lazy(() => V2EvalTemplatesDeleteNotFoundResponseBody$zodSchema),
57835
+ lazy(() => V2EvalTemplatesDeleteConflictResponseBody$zodSchema),
57036
57836
  HTTPValidationError$zodSchema
57037
57837
  ]);
57038
57838
  });
57039
57839
 
57040
- // src/funcs/evalsV2EvalsDelete.ts
57041
- function evalsV2EvalsDelete(client$, request, options) {
57042
- return new APIPromise($do20(client$, request, options));
57840
+ // src/funcs/evalTemplatesV2EvalTemplatesDelete.ts
57841
+ function evalTemplatesV2EvalTemplatesDelete(client$, request, options) {
57842
+ return new APIPromise($do24(client$, request, options));
57043
57843
  }
57044
- async function $do20(client$, request, options) {
57045
- const parsed$ = safeParse4(request, (value$) => V2EvalsDeleteRequest$zodSchema.parse(value$), "Input validation failed");
57844
+ async function $do24(client$, request, options) {
57845
+ const parsed$ = safeParse4(request, (value$) => V2EvalTemplatesDeleteRequest$zodSchema.parse(value$), "Input validation failed");
57046
57846
  if (!parsed$.ok) {
57047
57847
  return [parsed$, { status: "invalid" }];
57048
57848
  }
57049
57849
  const payload$ = parsed$.value;
57050
57850
  const body$ = null;
57051
57851
  const pathParams$ = {
57052
- eval_id: encodeSimple("eval_id", payload$.eval_id, {
57852
+ template_id: encodeSimple("template_id", payload$.template_id, {
57053
57853
  explode: false,
57054
57854
  charEncoding: "percent"
57055
57855
  })
57056
57856
  };
57057
- const path$ = pathToFunc("/v2/evals/{eval_id}")(pathParams$);
57857
+ const path$ = pathToFunc("/v2/eval-templates/{template_id}")(pathParams$);
57058
57858
  const headers$ = new Headers(compactMap({
57059
57859
  Accept: "application/json"
57060
57860
  }));
@@ -57063,7 +57863,7 @@ async function $do20(client$, request, options) {
57063
57863
  const context = {
57064
57864
  options: client$._options,
57065
57865
  baseURL: options?.serverURL ?? client$._baseURL ?? "",
57066
- operationID: "v2_evals_delete",
57866
+ operationID: "v2_eval_templates_delete",
57067
57867
  oAuth2Scopes: null,
57068
57868
  resolvedSecurity: requestSecurity,
57069
57869
  securitySource: client$._options.security,
@@ -57105,30 +57905,30 @@ async function $do20(client$, request, options) {
57105
57905
  response: doResult.value
57106
57906
  }];
57107
57907
  }
57108
- var init_evalsV2EvalsDelete = __esm(() => {
57908
+ var init_evalTemplatesV2EvalTemplatesDelete = __esm(() => {
57109
57909
  init_encodings();
57110
57910
  init_primitives();
57111
57911
  init_schemas4();
57112
57912
  init_security();
57113
57913
  init_url();
57114
- init_v2evalsdeleteop();
57914
+ init_v2evaltemplatesdeleteop();
57115
57915
  init_async();
57116
57916
  });
57117
57917
 
57118
- // src/mcp-server/tools/evalsV2EvalsDelete.ts
57119
- var args17, tool$evalsV2EvalsDelete;
57120
- var init_evalsV2EvalsDelete2 = __esm(() => {
57121
- init_evalsV2EvalsDelete();
57122
- init_v2evalsdeleteop();
57918
+ // src/mcp-server/tools/evalTemplatesV2EvalTemplatesDelete.ts
57919
+ var args20, tool$evalTemplatesV2EvalTemplatesDelete;
57920
+ var init_evalTemplatesV2EvalTemplatesDelete2 = __esm(() => {
57921
+ init_evalTemplatesV2EvalTemplatesDelete();
57922
+ init_v2evaltemplatesdeleteop();
57123
57923
  init_tools();
57124
- args17 = {
57125
- request: V2EvalsDeleteRequest$zodSchema
57924
+ args20 = {
57925
+ request: V2EvalTemplatesDeleteRequest$zodSchema
57126
57926
  };
57127
- tool$evalsV2EvalsDelete = {
57128
- name: "evals-v2-evals-delete",
57129
- description: `Delete Eval
57927
+ tool$evalTemplatesV2EvalTemplatesDelete = {
57928
+ name: "eval-templates-v2-eval-templates-delete",
57929
+ description: `Delete an Eval template
57130
57930
 
57131
- Permanently deletes an eval and all of its saved configurations.`,
57931
+ Soft-deletes an Eval template. Evals previously created from it remain independent and continue to run with their copied configuration.`,
57132
57932
  annotations: {
57133
57933
  title: "",
57134
57934
  destructiveHint: true,
@@ -57136,9 +57936,9 @@ Permanently deletes an eval and all of its saved configurations.`,
57136
57936
  openWorldHint: false,
57137
57937
  readOnlyHint: false
57138
57938
  },
57139
- args: args17,
57939
+ args: args20,
57140
57940
  tool: async (client, args, ctx) => {
57141
- const [result] = await evalsV2EvalsDelete(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
57941
+ const [result] = await evalTemplatesV2EvalTemplatesDelete(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
57142
57942
  if (!result.ok) {
57143
57943
  return {
57144
57944
  content: [{ type: "text", text: result.error.message }],
@@ -57150,12 +57950,12 @@ Permanently deletes an eval and all of its saved configurations.`,
57150
57950
  };
57151
57951
  });
57152
57952
 
57153
- // src/funcs/evalsV2EvalsList.ts
57154
- function evalsV2EvalsList(client$, options) {
57155
- return new APIPromise($do21(client$, options));
57953
+ // src/funcs/evalTemplatesV2EvalTemplatesList.ts
57954
+ function evalTemplatesV2EvalTemplatesList(client$, options) {
57955
+ return new APIPromise($do25(client$, options));
57156
57956
  }
57157
- async function $do21(client$, options) {
57158
- const path$ = pathToFunc("/v2/evals")();
57957
+ async function $do25(client$, options) {
57958
+ const path$ = pathToFunc("/v2/eval-templates")();
57159
57959
  const headers$ = new Headers(compactMap({
57160
57960
  Accept: "application/json"
57161
57961
  }));
@@ -57164,7 +57964,7 @@ async function $do21(client$, options) {
57164
57964
  const context = {
57165
57965
  options: client$._options,
57166
57966
  baseURL: options?.serverURL ?? client$._baseURL ?? "",
57167
- operationID: "v2_evals_list",
57967
+ operationID: "v2_eval_templates_list",
57168
57968
  oAuth2Scopes: null,
57169
57969
  resolvedSecurity: requestSecurity,
57170
57970
  securitySource: client$._options.security,
@@ -57205,23 +58005,23 @@ async function $do21(client$, options) {
57205
58005
  response: doResult.value
57206
58006
  }];
57207
58007
  }
57208
- var init_evalsV2EvalsList = __esm(() => {
58008
+ var init_evalTemplatesV2EvalTemplatesList = __esm(() => {
57209
58009
  init_primitives();
57210
58010
  init_security();
57211
58011
  init_url();
57212
58012
  init_async();
57213
58013
  });
57214
58014
 
57215
- // src/mcp-server/tools/evalsV2EvalsList.ts
57216
- var tool$evalsV2EvalsList;
57217
- var init_evalsV2EvalsList2 = __esm(() => {
57218
- init_evalsV2EvalsList();
58015
+ // src/mcp-server/tools/evalTemplatesV2EvalTemplatesList.ts
58016
+ var tool$evalTemplatesV2EvalTemplatesList;
58017
+ var init_evalTemplatesV2EvalTemplatesList2 = __esm(() => {
58018
+ init_evalTemplatesV2EvalTemplatesList();
57219
58019
  init_tools();
57220
- tool$evalsV2EvalsList = {
57221
- name: "evals-v2-evals-list",
57222
- description: `List Evals
58020
+ tool$evalTemplatesV2EvalTemplatesList = {
58021
+ name: "eval-templates-v2-eval-templates-list",
58022
+ description: `List Eval templates
57223
58023
 
57224
- Returns all active eval definitions owned by the organization.`,
58024
+ Returns active reusable Eval templates owned by the authenticated organization. Templates contain reusable configuration and never task prompts.`,
57225
58025
  annotations: {
57226
58026
  title: "",
57227
58027
  destructiveHint: false,
@@ -57230,7 +58030,7 @@ Returns all active eval definitions owned by the organization.`,
57230
58030
  readOnlyHint: true
57231
58031
  },
57232
58032
  tool: async (client, ctx) => {
57233
- const [result] = await evalsV2EvalsList(client, { fetchOptions: { signal: ctx.signal } }).$inspect();
58033
+ const [result] = await evalTemplatesV2EvalTemplatesList(client, { fetchOptions: { signal: ctx.signal } }).$inspect();
57234
58034
  if (!result.ok) {
57235
58035
  return {
57236
58036
  content: [{ type: "text", text: result.error.message }],
@@ -57242,39 +58042,43 @@ Returns all active eval definitions owned by the organization.`,
57242
58042
  };
57243
58043
  });
57244
58044
 
57245
- // src/models/v2evalsretrieveop.ts
57246
- var V2EvalsRetrieveRequest$zodSchema, V2EvalsRetrieveResponse$zodSchema;
57247
- var init_v2evalsretrieveop = __esm(() => {
58045
+ // src/models/v2evaltemplatesretrieveop.ts
58046
+ var V2EvalTemplatesRetrieveRequest$zodSchema, V2EvalTemplatesRetrieveResponseBody$zodSchema, V2EvalTemplatesRetrieveResponse$zodSchema;
58047
+ var init_v2evaltemplatesretrieveop = __esm(() => {
57248
58048
  init_zod();
58049
+ init_evaltemplate();
57249
58050
  init_httpvalidationerror();
57250
- init_v2eval();
57251
- V2EvalsRetrieveRequest$zodSchema = object({
57252
- eval_id: string2()
58051
+ V2EvalTemplatesRetrieveRequest$zodSchema = object({
58052
+ template_id: string2()
57253
58053
  });
57254
- V2EvalsRetrieveResponse$zodSchema = union([
57255
- V2Eval$zodSchema,
58054
+ V2EvalTemplatesRetrieveResponseBody$zodSchema = object({
58055
+ detail: string2()
58056
+ }).describe("The Eval template does not exist in the organization.");
58057
+ V2EvalTemplatesRetrieveResponse$zodSchema = union([
58058
+ EvalTemplate$zodSchema,
58059
+ lazy(() => V2EvalTemplatesRetrieveResponseBody$zodSchema),
57256
58060
  HTTPValidationError$zodSchema
57257
58061
  ]);
57258
58062
  });
57259
58063
 
57260
- // src/funcs/evalsV2EvalsRetrieve.ts
57261
- function evalsV2EvalsRetrieve(client$, request, options) {
57262
- return new APIPromise($do22(client$, request, options));
58064
+ // src/funcs/evalTemplatesV2EvalTemplatesRetrieve.ts
58065
+ function evalTemplatesV2EvalTemplatesRetrieve(client$, request, options) {
58066
+ return new APIPromise($do26(client$, request, options));
57263
58067
  }
57264
- async function $do22(client$, request, options) {
57265
- const parsed$ = safeParse4(request, (value$) => V2EvalsRetrieveRequest$zodSchema.parse(value$), "Input validation failed");
58068
+ async function $do26(client$, request, options) {
58069
+ const parsed$ = safeParse4(request, (value$) => V2EvalTemplatesRetrieveRequest$zodSchema.parse(value$), "Input validation failed");
57266
58070
  if (!parsed$.ok) {
57267
58071
  return [parsed$, { status: "invalid" }];
57268
58072
  }
57269
58073
  const payload$ = parsed$.value;
57270
58074
  const body$ = null;
57271
58075
  const pathParams$ = {
57272
- eval_id: encodeSimple("eval_id", payload$.eval_id, {
58076
+ template_id: encodeSimple("template_id", payload$.template_id, {
57273
58077
  explode: false,
57274
58078
  charEncoding: "percent"
57275
58079
  })
57276
58080
  };
57277
- const path$ = pathToFunc("/v2/evals/{eval_id}")(pathParams$);
58081
+ const path$ = pathToFunc("/v2/eval-templates/{template_id}")(pathParams$);
57278
58082
  const headers$ = new Headers(compactMap({
57279
58083
  Accept: "application/json"
57280
58084
  }));
@@ -57283,7 +58087,7 @@ async function $do22(client$, request, options) {
57283
58087
  const context = {
57284
58088
  options: client$._options,
57285
58089
  baseURL: options?.serverURL ?? client$._baseURL ?? "",
57286
- operationID: "v2_evals_retrieve",
58090
+ operationID: "v2_eval_templates_retrieve",
57287
58091
  oAuth2Scopes: null,
57288
58092
  resolvedSecurity: requestSecurity,
57289
58093
  securitySource: client$._options.security,
@@ -57325,30 +58129,30 @@ async function $do22(client$, request, options) {
57325
58129
  response: doResult.value
57326
58130
  }];
57327
58131
  }
57328
- var init_evalsV2EvalsRetrieve = __esm(() => {
58132
+ var init_evalTemplatesV2EvalTemplatesRetrieve = __esm(() => {
57329
58133
  init_encodings();
57330
58134
  init_primitives();
57331
58135
  init_schemas4();
57332
58136
  init_security();
57333
58137
  init_url();
57334
- init_v2evalsretrieveop();
58138
+ init_v2evaltemplatesretrieveop();
57335
58139
  init_async();
57336
58140
  });
57337
58141
 
57338
- // src/mcp-server/tools/evalsV2EvalsRetrieve.ts
57339
- var args18, tool$evalsV2EvalsRetrieve;
57340
- var init_evalsV2EvalsRetrieve2 = __esm(() => {
57341
- init_evalsV2EvalsRetrieve();
57342
- init_v2evalsretrieveop();
58142
+ // src/mcp-server/tools/evalTemplatesV2EvalTemplatesRetrieve.ts
58143
+ var args21, tool$evalTemplatesV2EvalTemplatesRetrieve;
58144
+ var init_evalTemplatesV2EvalTemplatesRetrieve2 = __esm(() => {
58145
+ init_evalTemplatesV2EvalTemplatesRetrieve();
58146
+ init_v2evaltemplatesretrieveop();
57343
58147
  init_tools();
57344
- args18 = {
57345
- request: V2EvalsRetrieveRequest$zodSchema
58148
+ args21 = {
58149
+ request: V2EvalTemplatesRetrieveRequest$zodSchema
57346
58150
  };
57347
- tool$evalsV2EvalsRetrieve = {
57348
- name: "evals-v2-evals-retrieve",
57349
- description: `Retrieve Eval
58151
+ tool$evalTemplatesV2EvalTemplatesRetrieve = {
58152
+ name: "eval-templates-v2-eval-templates-retrieve",
58153
+ description: `Retrieve an Eval template
57350
58154
 
57351
- Returns one reusable eval simulation definition.`,
58155
+ Returns one active reusable Eval template.`,
57352
58156
  annotations: {
57353
58157
  title: "",
57354
58158
  destructiveHint: false,
@@ -57356,9 +58160,9 @@ Returns one reusable eval simulation definition.`,
57356
58160
  openWorldHint: false,
57357
58161
  readOnlyHint: true
57358
58162
  },
57359
- args: args18,
58163
+ args: args21,
57360
58164
  tool: async (client, args, ctx) => {
57361
- const [result] = await evalsV2EvalsRetrieve(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
58165
+ const [result] = await evalTemplatesV2EvalTemplatesRetrieve(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
57362
58166
  if (!result.ok) {
57363
58167
  return {
57364
58168
  content: [{ type: "text", text: result.error.message }],
@@ -57370,76 +58174,101 @@ Returns one reusable eval simulation definition.`,
57370
58174
  };
57371
58175
  });
57372
58176
 
57373
- // src/models/updatev2evalrequest.ts
57374
- var UpdateV2EvalRequestCli$zodSchema, UpdateV2EvalRequestMcp$zodSchema, UpdateV2EvalRequest$zodSchema;
57375
- var init_updatev2evalrequest = __esm(() => {
58177
+ // src/models/updateevaltemplaterequest.ts
58178
+ var UpdateEvalTemplateRequestEvalType$zodSchema, UpdateEvalTemplateRequestIncludedComponent$zodSchema, UpdateEvalTemplateRequest$zodSchema;
58179
+ var init_updateevaltemplaterequest = __esm(() => {
57376
58180
  init_zod();
57377
- init_v2inlineclireference();
57378
- init_v2inlinemcpdefinition();
57379
- init_v2savedclireference();
57380
- init_v2savedmcpreference();
57381
- UpdateV2EvalRequestCli$zodSchema = union([
57382
- V2SavedCliReference$zodSchema,
57383
- V2InlineCliReference$zodSchema
58181
+ init_evaltemplateruntime();
58182
+ init_evaltemplateschedule();
58183
+ init_graderdefinition();
58184
+ UpdateEvalTemplateRequestEvalType$zodSchema = _enum([
58185
+ "custom_eval",
58186
+ "agent_preference"
57384
58187
  ]);
57385
- UpdateV2EvalRequestMcp$zodSchema = union([
57386
- V2InlineMcpDefinition$zodSchema,
57387
- V2SavedMcpReference$zodSchema
58188
+ UpdateEvalTemplateRequestIncludedComponent$zodSchema = _enum([
58189
+ "judges",
58190
+ "targets",
58191
+ "persona",
58192
+ "competitors",
58193
+ "tag",
58194
+ "category",
58195
+ "runs",
58196
+ "schedule",
58197
+ "repository",
58198
+ "environment",
58199
+ "search_grounding",
58200
+ "skills",
58201
+ "mcp_servers",
58202
+ "clis",
58203
+ "brand_docs",
58204
+ "github_pr",
58205
+ "runtime_commands"
57388
58206
  ]);
57389
- UpdateV2EvalRequest$zodSchema = object({
57390
- agent: array(string2()).nullable().optional(),
57391
- clis: array(union([
57392
- V2SavedCliReference$zodSchema,
57393
- V2InlineCliReference$zodSchema
57394
- ])).nullable().optional(),
57395
- judges: array(string2()).nullable().optional().describe("Replacement success criteria for the eval."),
57396
- mcps: array(union([
57397
- V2InlineMcpDefinition$zodSchema,
57398
- V2SavedMcpReference$zodSchema
57399
- ])).nullable().optional(),
58207
+ UpdateEvalTemplateRequest$zodSchema = object({
58208
+ category_name: string2().optional(),
58209
+ competitor_ids: array(string2()).nullable().optional(),
58210
+ description: string2().optional(),
58211
+ eval_type: UpdateEvalTemplateRequestEvalType$zodSchema.optional(),
58212
+ expected_template_version: int().optional(),
58213
+ included_components: array(UpdateEvalTemplateRequestIncludedComponent$zodSchema).nullable().optional(),
58214
+ judges: array(GraderDefinition$zodSchema).nullable().optional(),
57400
58215
  name: string2().optional(),
57401
- persona: string2().optional().describe("Replacement eval persona. Pass null to clear it; omit it to leave the saved value unchanged."),
57402
- prompts: array(string2()).nullable().optional(),
57403
- runs: int().optional(),
57404
- tags: array(string2()).nullable().optional().describe("Replacement labels for the eval. Pass an empty array to clear all tags.")
58216
+ persona: string2().optional(),
58217
+ runs_per_execution: int().optional(),
58218
+ runtime: EvalTemplateRuntime$zodSchema.nullable().optional(),
58219
+ schedule: EvalTemplateSchedule$zodSchema.nullable().optional(),
58220
+ tag: string2().optional(),
58221
+ target_keys: array(string2()).nullable().optional()
57405
58222
  });
57406
58223
  });
57407
58224
 
57408
- // src/models/v2evalsupdateop.ts
57409
- var V2EvalsUpdateRequest$zodSchema, V2EvalsUpdateResponse$zodSchema;
57410
- var init_v2evalsupdateop = __esm(() => {
58225
+ // src/models/v2evaltemplatesupdateop.ts
58226
+ var V2EvalTemplatesUpdateRequest$zodSchema, V2EvalTemplatesUpdateConflictResponseBody$zodSchema, V2EvalTemplatesUpdateNotFoundResponseBody$zodSchema, V2EvalTemplatesUpdateBadRequestResponseBody$zodSchema, V2EvalTemplatesUpdateResponse$zodSchema;
58227
+ var init_v2evaltemplatesupdateop = __esm(() => {
57411
58228
  init_zod();
58229
+ init_evaltemplate();
57412
58230
  init_httpvalidationerror();
57413
- init_updatev2evalrequest();
57414
- init_v2eval();
57415
- V2EvalsUpdateRequest$zodSchema = object({
57416
- body: UpdateV2EvalRequest$zodSchema,
57417
- eval_id: string2()
58231
+ init_updateevaltemplaterequest();
58232
+ V2EvalTemplatesUpdateRequest$zodSchema = object({
58233
+ body: UpdateEvalTemplateRequest$zodSchema,
58234
+ template_id: string2()
57418
58235
  });
57419
- V2EvalsUpdateResponse$zodSchema = union([
57420
- V2Eval$zodSchema,
58236
+ V2EvalTemplatesUpdateConflictResponseBody$zodSchema = object({
58237
+ detail: string2()
58238
+ }).describe("The supplied template version is stale.");
58239
+ V2EvalTemplatesUpdateNotFoundResponseBody$zodSchema = object({
58240
+ detail: string2()
58241
+ }).describe("The Eval template does not exist in the organization.");
58242
+ V2EvalTemplatesUpdateBadRequestResponseBody$zodSchema = object({
58243
+ detail: string2()
58244
+ }).describe("The template contains invalid judges, targets, scheduling, metadata, or organization resources.");
58245
+ V2EvalTemplatesUpdateResponse$zodSchema = union([
58246
+ EvalTemplate$zodSchema,
58247
+ lazy(() => V2EvalTemplatesUpdateBadRequestResponseBody$zodSchema),
58248
+ lazy(() => V2EvalTemplatesUpdateNotFoundResponseBody$zodSchema),
58249
+ lazy(() => V2EvalTemplatesUpdateConflictResponseBody$zodSchema),
57421
58250
  HTTPValidationError$zodSchema
57422
58251
  ]);
57423
58252
  });
57424
58253
 
57425
- // src/funcs/evalsV2EvalsUpdate.ts
57426
- function evalsV2EvalsUpdate(client$, request, options) {
57427
- return new APIPromise($do23(client$, request, options));
58254
+ // src/funcs/evalTemplatesV2EvalTemplatesUpdate.ts
58255
+ function evalTemplatesV2EvalTemplatesUpdate(client$, request, options) {
58256
+ return new APIPromise($do27(client$, request, options));
57428
58257
  }
57429
- async function $do23(client$, request, options) {
57430
- const parsed$ = safeParse4(request, (value$) => V2EvalsUpdateRequest$zodSchema.parse(value$), "Input validation failed");
58258
+ async function $do27(client$, request, options) {
58259
+ const parsed$ = safeParse4(request, (value$) => V2EvalTemplatesUpdateRequest$zodSchema.parse(value$), "Input validation failed");
57431
58260
  if (!parsed$.ok) {
57432
58261
  return [parsed$, { status: "invalid" }];
57433
58262
  }
57434
58263
  const payload$ = parsed$.value;
57435
58264
  const body$ = encodeJSON("body", payload$.body, { explode: true });
57436
58265
  const pathParams$ = {
57437
- eval_id: encodeSimple("eval_id", payload$.eval_id, {
58266
+ template_id: encodeSimple("template_id", payload$.template_id, {
57438
58267
  explode: false,
57439
58268
  charEncoding: "percent"
57440
58269
  })
57441
58270
  };
57442
- const path$ = pathToFunc("/v2/evals/{eval_id}")(pathParams$);
58271
+ const path$ = pathToFunc("/v2/eval-templates/{template_id}")(pathParams$);
57443
58272
  const headers$ = new Headers(compactMap({
57444
58273
  "Content-Type": "application/json",
57445
58274
  Accept: "application/json"
@@ -57449,7 +58278,7 @@ async function $do23(client$, request, options) {
57449
58278
  const context = {
57450
58279
  options: client$._options,
57451
58280
  baseURL: options?.serverURL ?? client$._baseURL ?? "",
57452
- operationID: "v2_evals_update",
58281
+ operationID: "v2_eval_templates_update",
57453
58282
  oAuth2Scopes: null,
57454
58283
  resolvedSecurity: requestSecurity,
57455
58284
  securitySource: client$._options.security,
@@ -57491,30 +58320,30 @@ async function $do23(client$, request, options) {
57491
58320
  response: doResult.value
57492
58321
  }];
57493
58322
  }
57494
- var init_evalsV2EvalsUpdate = __esm(() => {
58323
+ var init_evalTemplatesV2EvalTemplatesUpdate = __esm(() => {
57495
58324
  init_encodings();
57496
58325
  init_primitives();
57497
58326
  init_schemas4();
57498
58327
  init_security();
57499
58328
  init_url();
57500
- init_v2evalsupdateop();
58329
+ init_v2evaltemplatesupdateop();
57501
58330
  init_async();
57502
58331
  });
57503
58332
 
57504
- // src/mcp-server/tools/evalsV2EvalsUpdate.ts
57505
- var args19, tool$evalsV2EvalsUpdate;
57506
- var init_evalsV2EvalsUpdate2 = __esm(() => {
57507
- init_evalsV2EvalsUpdate();
57508
- init_v2evalsupdateop();
58333
+ // src/mcp-server/tools/evalTemplatesV2EvalTemplatesUpdate.ts
58334
+ var args22, tool$evalTemplatesV2EvalTemplatesUpdate;
58335
+ var init_evalTemplatesV2EvalTemplatesUpdate2 = __esm(() => {
58336
+ init_evalTemplatesV2EvalTemplatesUpdate();
58337
+ init_v2evaltemplatesupdateop();
57509
58338
  init_tools();
57510
- args19 = {
57511
- request: V2EvalsUpdateRequest$zodSchema
58339
+ args22 = {
58340
+ request: V2EvalTemplatesUpdateRequest$zodSchema
57512
58341
  };
57513
- tool$evalsV2EvalsUpdate = {
57514
- name: "evals-v2-evals-update",
57515
- description: `Update Eval
58342
+ tool$evalTemplatesV2EvalTemplatesUpdate = {
58343
+ name: "eval-templates-v2-eval-templates-update",
58344
+ description: `Update an Eval template
57516
58345
 
57517
- Updates a reusable eval simulation definition. Omitted fields remain unchanged. Empty \`agent\`, \`clis\`, \`mcps\`, or \`tags\` arrays clear those collections, and \`persona: null\` clears the saved persona.`,
58346
+ Updates an Eval template. Supply \`expected_template_version\` to reject an edit based on stale data. Existing evals are never changed.`,
57518
58347
  annotations: {
57519
58348
  title: "",
57520
58349
  destructiveHint: false,
@@ -57522,9 +58351,9 @@ Updates a reusable eval simulation definition. Omitted fields remain unchanged.
57522
58351
  openWorldHint: false,
57523
58352
  readOnlyHint: false
57524
58353
  },
57525
- args: args19,
58354
+ args: args22,
57526
58355
  tool: async (client, args, ctx) => {
57527
- const [result] = await evalsV2EvalsUpdate(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
58356
+ const [result] = await evalTemplatesV2EvalTemplatesUpdate(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
57528
58357
  if (!result.ok) {
57529
58358
  return {
57530
58359
  content: [{ type: "text", text: result.error.message }],
@@ -57538,9 +58367,9 @@ Updates a reusable eval simulation definition. Omitted fields remain unchanged.
57538
58367
 
57539
58368
  // src/funcs/mcpServersV2MCPServersList.ts
57540
58369
  function mcpServersV2MCPServersList(client$, options) {
57541
- return new APIPromise($do24(client$, options));
58370
+ return new APIPromise($do28(client$, options));
57542
58371
  }
57543
- async function $do24(client$, options) {
58372
+ async function $do28(client$, options) {
57544
58373
  const path$ = pathToFunc("/v2/mcp-servers")();
57545
58374
  const headers$ = new Headers(compactMap({
57546
58375
  Accept: "application/json"
@@ -57722,9 +58551,9 @@ var init_v2opportunitieslistop = __esm(() => {
57722
58551
 
57723
58552
  // src/funcs/opportunitiesV2OpportunitiesList.ts
57724
58553
  function opportunitiesV2OpportunitiesList(client$, request, options) {
57725
- return new APIPromise($do25(client$, request, options));
58554
+ return new APIPromise($do29(client$, request, options));
57726
58555
  }
57727
- async function $do25(client$, request, options) {
58556
+ async function $do29(client$, request, options) {
57728
58557
  const parsed$ = safeParse4(request, (value$) => V2OpportunitiesListRequest$zodSchema.optional().parse(value$), "Input validation failed");
57729
58558
  if (!parsed$.ok) {
57730
58559
  return [parsed$, { status: "invalid" }];
@@ -57802,12 +58631,12 @@ var init_opportunitiesV2OpportunitiesList = __esm(() => {
57802
58631
  });
57803
58632
 
57804
58633
  // src/mcp-server/tools/opportunitiesV2OpportunitiesList.ts
57805
- var args20, tool$opportunitiesV2OpportunitiesList;
58634
+ var args23, tool$opportunitiesV2OpportunitiesList;
57806
58635
  var init_opportunitiesV2OpportunitiesList2 = __esm(() => {
57807
58636
  init_opportunitiesV2OpportunitiesList();
57808
58637
  init_v2opportunitieslistop();
57809
58638
  init_tools();
57810
- args20 = {
58639
+ args23 = {
57811
58640
  request: V2OpportunitiesListRequest$zodSchema.optional()
57812
58641
  };
57813
58642
  tool$opportunitiesV2OpportunitiesList = {
@@ -57822,7 +58651,7 @@ Returns ranked opportunities computed from stored prompt visibility results. Eac
57822
58651
  openWorldHint: false,
57823
58652
  readOnlyHint: true
57824
58653
  },
57825
- args: args20,
58654
+ args: args23,
57826
58655
  tool: async (client, args, ctx) => {
57827
58656
  const [result] = await opportunitiesV2OpportunitiesList(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
57828
58657
  if (!result.ok) {
@@ -57894,9 +58723,9 @@ var init_v2opportunitiesretrieveop = __esm(() => {
57894
58723
 
57895
58724
  // src/funcs/opportunitiesV2OpportunitiesRetrieve.ts
57896
58725
  function opportunitiesV2OpportunitiesRetrieve(client$, request, options) {
57897
- return new APIPromise($do26(client$, request, options));
58726
+ return new APIPromise($do30(client$, request, options));
57898
58727
  }
57899
- async function $do26(client$, request, options) {
58728
+ async function $do30(client$, request, options) {
57900
58729
  const parsed$ = safeParse4(request, (value$) => V2OpportunitiesRetrieveRequest$zodSchema.parse(value$), "Input validation failed");
57901
58730
  if (!parsed$.ok) {
57902
58731
  return [parsed$, { status: "invalid" }];
@@ -57977,12 +58806,12 @@ var init_opportunitiesV2OpportunitiesRetrieve = __esm(() => {
57977
58806
  });
57978
58807
 
57979
58808
  // src/mcp-server/tools/opportunitiesV2OpportunitiesRetrieve.ts
57980
- var args21, tool$opportunitiesV2OpportunitiesRetrieve;
58809
+ var args24, tool$opportunitiesV2OpportunitiesRetrieve;
57981
58810
  var init_opportunitiesV2OpportunitiesRetrieve2 = __esm(() => {
57982
58811
  init_opportunitiesV2OpportunitiesRetrieve();
57983
58812
  init_v2opportunitiesretrieveop();
57984
58813
  init_tools();
57985
- args21 = {
58814
+ args24 = {
57986
58815
  request: V2OpportunitiesRetrieveRequest$zodSchema
57987
58816
  };
57988
58817
  tool$opportunitiesV2OpportunitiesRetrieve = {
@@ -57997,7 +58826,7 @@ Returns one computed opportunity and its execution context for the selected date
57997
58826
  openWorldHint: false,
57998
58827
  readOnlyHint: true
57999
58828
  },
58000
- args: args21,
58829
+ args: args24,
58001
58830
  tool: async (client, args, ctx) => {
58002
58831
  const [result] = await opportunitiesV2OpportunitiesRetrieve(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
58003
58832
  if (!result.ok) {
@@ -58066,7 +58895,7 @@ var init_v2promptrunsummary = __esm(() => {
58066
58895
  response_preview: string2().nullable().optional(),
58067
58896
  run_date: string2(),
58068
58897
  target_kind: V2PromptRunSummaryTargetKind$zodSchema.nullable().optional(),
58069
- tool_call_count: int().nullable().optional()
58898
+ tool_call_count: int().nullable().optional().describe("Coding-agent tool invocations, counting a call and its result once. Matches the length of `tool_calls` on the detail response; null for answer-engine runs.")
58070
58899
  });
58071
58900
  });
58072
58901
 
@@ -58105,9 +58934,9 @@ var init_v2promptrunslistop = __esm(() => {
58105
58934
 
58106
58935
  // src/funcs/promptsV2PromptRunsList.ts
58107
58936
  function promptsV2PromptRunsList(client$, request, options) {
58108
- return new APIPromise($do27(client$, request, options));
58937
+ return new APIPromise($do31(client$, request, options));
58109
58938
  }
58110
- async function $do27(client$, request, options) {
58939
+ async function $do31(client$, request, options) {
58111
58940
  const parsed$ = safeParse4(request, (value$) => V2PromptRunsListRequest$zodSchema.optional().parse(value$), "Input validation failed");
58112
58941
  if (!parsed$.ok) {
58113
58942
  return [parsed$, { status: "invalid" }];
@@ -58185,19 +59014,19 @@ var init_promptsV2PromptRunsList = __esm(() => {
58185
59014
  });
58186
59015
 
58187
59016
  // src/mcp-server/tools/promptsV2PromptRunsList.ts
58188
- var args22, tool$promptsV2PromptRunsList;
59017
+ var args25, tool$promptsV2PromptRunsList;
58189
59018
  var init_promptsV2PromptRunsList2 = __esm(() => {
58190
59019
  init_promptsV2PromptRunsList();
58191
59020
  init_v2promptrunslistop();
58192
59021
  init_tools();
58193
- args22 = {
59022
+ args25 = {
58194
59023
  request: V2PromptRunsListRequest$zodSchema.optional()
58195
59024
  };
58196
59025
  tool$promptsV2PromptRunsList = {
58197
59026
  name: "prompts-v2-prompt-runs-list",
58198
59027
  description: `List Prompt Runs
58199
59028
 
58200
- Returns stored prompt runs without exposing raw runtime or provider configuration. Filterable by prompt, platform, and date range.`,
59029
+ Returns stored prompt runs with limit-and-offset pagination, without exposing raw runtime or provider configuration. Filter by prompt, platform, or date range.`,
58201
59030
  annotations: {
58202
59031
  title: "",
58203
59032
  destructiveHint: false,
@@ -58205,7 +59034,7 @@ Returns stored prompt runs without exposing raw runtime or provider configuratio
58205
59034
  openWorldHint: false,
58206
59035
  readOnlyHint: true
58207
59036
  },
58208
- args: args22,
59037
+ args: args25,
58209
59038
  tool: async (client, args, ctx) => {
58210
59039
  const [result] = await promptsV2PromptRunsList(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
58211
59040
  if (!result.ok) {
@@ -58276,6 +59105,7 @@ var init_v2promptrundetail = __esm(() => {
58276
59105
  V2PromptRunDetail$zodSchema = object({
58277
59106
  citation_count: int().default(0),
58278
59107
  citations: array(V2PromptRunCitation$zodSchema).optional(),
59108
+ conversation_turns: array(record(string2(), any())).optional().describe("Sanitized interleaved text and tool-call trace captured for this run. It interleaves the agent's `text` turns with `tool_call` turns, so its length is not a tool-call count; use `tool_calls` or `tool_call_count`."),
58279
59109
  created_at: string2().nullable().optional(),
58280
59110
  harness: string2().nullable().optional().describe("Coding-agent harness for agent runs; otherwise the answer platform identifier."),
58281
59111
  id: string2(),
@@ -58289,7 +59119,8 @@ var init_v2promptrundetail = __esm(() => {
58289
59119
  response_preview: string2().nullable().optional(),
58290
59120
  run_date: string2(),
58291
59121
  target_kind: V2PromptRunDetailTargetKind$zodSchema.nullable().optional(),
58292
- tool_call_count: int().nullable().optional()
59122
+ tool_call_count: int().nullable().optional().describe("Coding-agent tool invocations, counting a call and its result once. Matches the length of `tool_calls` on the detail response; null for answer-engine runs."),
59123
+ tool_calls: array(record(string2(), any())).optional().describe("Sanitized coding-agent tool invocations captured for this run, one entry per invocation with its result merged in.")
58293
59124
  });
58294
59125
  });
58295
59126
 
@@ -58310,9 +59141,9 @@ var init_v2promptrunsretrieveop = __esm(() => {
58310
59141
 
58311
59142
  // src/funcs/promptsV2PromptRunsRetrieve.ts
58312
59143
  function promptsV2PromptRunsRetrieve(client$, request, options) {
58313
- return new APIPromise($do28(client$, request, options));
59144
+ return new APIPromise($do32(client$, request, options));
58314
59145
  }
58315
- async function $do28(client$, request, options) {
59146
+ async function $do32(client$, request, options) {
58316
59147
  const parsed$ = safeParse4(request, (value$) => V2PromptRunsRetrieveRequest$zodSchema.parse(value$), "Input validation failed");
58317
59148
  if (!parsed$.ok) {
58318
59149
  return [parsed$, { status: "invalid" }];
@@ -58387,19 +59218,19 @@ var init_promptsV2PromptRunsRetrieve = __esm(() => {
58387
59218
  });
58388
59219
 
58389
59220
  // src/mcp-server/tools/promptsV2PromptRunsRetrieve.ts
58390
- var args23, tool$promptsV2PromptRunsRetrieve;
59221
+ var args26, tool$promptsV2PromptRunsRetrieve;
58391
59222
  var init_promptsV2PromptRunsRetrieve2 = __esm(() => {
58392
59223
  init_promptsV2PromptRunsRetrieve();
58393
59224
  init_v2promptrunsretrieveop();
58394
59225
  init_tools();
58395
- args23 = {
59226
+ args26 = {
58396
59227
  request: V2PromptRunsRetrieveRequest$zodSchema
58397
59228
  };
58398
59229
  tool$promptsV2PromptRunsRetrieve = {
58399
59230
  name: "prompts-v2-prompt-runs-retrieve",
58400
59231
  description: `Retrieve Prompt Run
58401
59232
 
58402
- Returns one stored prompt run, including its mentions and citations.`,
59233
+ Returns one stored prompt run, including mentions, citations, and any sanitized coding-agent trace captured for the run.`,
58403
59234
  annotations: {
58404
59235
  title: "",
58405
59236
  destructiveHint: false,
@@ -58407,7 +59238,7 @@ Returns one stored prompt run, including its mentions and citations.`,
58407
59238
  openWorldHint: false,
58408
59239
  readOnlyHint: true
58409
59240
  },
58410
- args: args23,
59241
+ args: args26,
58411
59242
  tool: async (client, args, ctx) => {
58412
59243
  const [result] = await promptsV2PromptRunsRetrieve(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
58413
59244
  if (!result.ok) {
@@ -58444,9 +59275,9 @@ var init_createv2promptsrequest = __esm(() => {
58444
59275
 
58445
59276
  // src/funcs/promptsV2PromptsCreate.ts
58446
59277
  function promptsV2PromptsCreate(client$, request, options) {
58447
- return new APIPromise($do29(client$, request, options));
59278
+ return new APIPromise($do33(client$, request, options));
58448
59279
  }
58449
- async function $do29(client$, request, options) {
59280
+ async function $do33(client$, request, options) {
58450
59281
  const parsed$ = safeParse4(request, (value$) => CreateV2PromptsRequest$zodSchema.parse(value$), "Input validation failed");
58451
59282
  if (!parsed$.ok) {
58452
59283
  return [parsed$, { status: "invalid" }];
@@ -58516,12 +59347,12 @@ var init_promptsV2PromptsCreate = __esm(() => {
58516
59347
  });
58517
59348
 
58518
59349
  // src/mcp-server/tools/promptsV2PromptsCreate.ts
58519
- var args24, tool$promptsV2PromptsCreate;
59350
+ var args27, tool$promptsV2PromptsCreate;
58520
59351
  var init_promptsV2PromptsCreate2 = __esm(() => {
58521
59352
  init_promptsV2PromptsCreate();
58522
59353
  init_createv2promptsrequest();
58523
59354
  init_tools();
58524
- args24 = {
59355
+ args27 = {
58525
59356
  request: CreateV2PromptsRequest$zodSchema
58526
59357
  };
58527
59358
  tool$promptsV2PromptsCreate = {
@@ -58536,7 +59367,7 @@ Creates prompts in a batch. Each item is validated independently, so the respons
58536
59367
  openWorldHint: false,
58537
59368
  readOnlyHint: false
58538
59369
  },
58539
- args: args24,
59370
+ args: args27,
58540
59371
  tool: async (client, args, ctx) => {
58541
59372
  const [result] = await promptsV2PromptsCreate(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
58542
59373
  if (!result.ok) {
@@ -58576,9 +59407,9 @@ var init_v2promptsdeleteop = __esm(() => {
58576
59407
 
58577
59408
  // src/funcs/promptsV2PromptsDelete.ts
58578
59409
  function promptsV2PromptsDelete(client$, request, options) {
58579
- return new APIPromise($do30(client$, request, options));
59410
+ return new APIPromise($do34(client$, request, options));
58580
59411
  }
58581
- async function $do30(client$, request, options) {
59412
+ async function $do34(client$, request, options) {
58582
59413
  const parsed$ = safeParse4(request, (value$) => V2PromptsDeleteRequest$zodSchema.parse(value$), "Input validation failed");
58583
59414
  if (!parsed$.ok) {
58584
59415
  return [parsed$, { status: "invalid" }];
@@ -58653,12 +59484,12 @@ var init_promptsV2PromptsDelete = __esm(() => {
58653
59484
  });
58654
59485
 
58655
59486
  // src/mcp-server/tools/promptsV2PromptsDelete.ts
58656
- var args25, tool$promptsV2PromptsDelete;
59487
+ var args28, tool$promptsV2PromptsDelete;
58657
59488
  var init_promptsV2PromptsDelete2 = __esm(() => {
58658
59489
  init_promptsV2PromptsDelete();
58659
59490
  init_v2promptsdeleteop();
58660
59491
  init_tools();
58661
- args25 = {
59492
+ args28 = {
58662
59493
  request: V2PromptsDeleteRequest$zodSchema
58663
59494
  };
58664
59495
  tool$promptsV2PromptsDelete = {
@@ -58673,7 +59504,7 @@ Archives the prompt and excludes it from normal listings while preserving its hi
58673
59504
  openWorldHint: false,
58674
59505
  readOnlyHint: false
58675
59506
  },
58676
- args: args25,
59507
+ args: args28,
58677
59508
  tool: async (client, args, ctx) => {
58678
59509
  const [result] = await promptsV2PromptsDelete(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
58679
59510
  if (!result.ok) {
@@ -58783,9 +59614,9 @@ var init_v2promptslistop = __esm(() => {
58783
59614
 
58784
59615
  // src/funcs/promptsV2PromptsList.ts
58785
59616
  function promptsV2PromptsList(client$, request, options) {
58786
- return new APIPromise($do31(client$, request, options));
59617
+ return new APIPromise($do35(client$, request, options));
58787
59618
  }
58788
- async function $do31(client$, request, options) {
59619
+ async function $do35(client$, request, options) {
58789
59620
  const parsed$ = safeParse4(request, (value$) => V2PromptsListRequest$zodSchema.optional().parse(value$), "Input validation failed");
58790
59621
  if (!parsed$.ok) {
58791
59622
  return [parsed$, { status: "invalid" }];
@@ -58860,12 +59691,12 @@ var init_promptsV2PromptsList = __esm(() => {
58860
59691
  });
58861
59692
 
58862
59693
  // src/mcp-server/tools/promptsV2PromptsList.ts
58863
- var args26, tool$promptsV2PromptsList;
59694
+ var args29, tool$promptsV2PromptsList;
58864
59695
  var init_promptsV2PromptsList2 = __esm(() => {
58865
59696
  init_promptsV2PromptsList();
58866
59697
  init_v2promptslistop();
58867
59698
  init_tools();
58868
- args26 = {
59699
+ args29 = {
58869
59700
  request: V2PromptsListRequest$zodSchema.optional()
58870
59701
  };
58871
59702
  tool$promptsV2PromptsList = {
@@ -58880,7 +59711,7 @@ Returns active and inactive prompts by default. Set \`include_archived\` to incl
58880
59711
  openWorldHint: false,
58881
59712
  readOnlyHint: true
58882
59713
  },
58883
- args: args26,
59714
+ args: args29,
58884
59715
  tool: async (client, args, ctx) => {
58885
59716
  const [result] = await promptsV2PromptsList(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
58886
59717
  if (!result.ok) {
@@ -58911,9 +59742,9 @@ var init_v2promptsretrieveop = __esm(() => {
58911
59742
 
58912
59743
  // src/funcs/promptsV2PromptsRetrieve.ts
58913
59744
  function promptsV2PromptsRetrieve(client$, request, options) {
58914
- return new APIPromise($do32(client$, request, options));
59745
+ return new APIPromise($do36(client$, request, options));
58915
59746
  }
58916
- async function $do32(client$, request, options) {
59747
+ async function $do36(client$, request, options) {
58917
59748
  const parsed$ = safeParse4(request, (value$) => V2PromptsRetrieveRequest$zodSchema.parse(value$), "Input validation failed");
58918
59749
  if (!parsed$.ok) {
58919
59750
  return [parsed$, { status: "invalid" }];
@@ -58988,12 +59819,12 @@ var init_promptsV2PromptsRetrieve = __esm(() => {
58988
59819
  });
58989
59820
 
58990
59821
  // src/mcp-server/tools/promptsV2PromptsRetrieve.ts
58991
- var args27, tool$promptsV2PromptsRetrieve;
59822
+ var args30, tool$promptsV2PromptsRetrieve;
58992
59823
  var init_promptsV2PromptsRetrieve2 = __esm(() => {
58993
59824
  init_promptsV2PromptsRetrieve();
58994
59825
  init_v2promptsretrieveop();
58995
59826
  init_tools();
58996
- args27 = {
59827
+ args30 = {
58997
59828
  request: V2PromptsRetrieveRequest$zodSchema
58998
59829
  };
58999
59830
  tool$promptsV2PromptsRetrieve = {
@@ -59008,7 +59839,7 @@ Returns a prompt with its topic, platforms, tags, and lifecycle status.`,
59008
59839
  openWorldHint: false,
59009
59840
  readOnlyHint: true
59010
59841
  },
59011
- args: args27,
59842
+ args: args30,
59012
59843
  tool: async (client, args, ctx) => {
59013
59844
  const [result] = await promptsV2PromptsRetrieve(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
59014
59845
  if (!result.ok) {
@@ -59057,9 +59888,9 @@ var init_v2promptsupdateop = __esm(() => {
59057
59888
 
59058
59889
  // src/funcs/promptsV2PromptsUpdate.ts
59059
59890
  function promptsV2PromptsUpdate(client$, request, options) {
59060
- return new APIPromise($do33(client$, request, options));
59891
+ return new APIPromise($do37(client$, request, options));
59061
59892
  }
59062
- async function $do33(client$, request, options) {
59893
+ async function $do37(client$, request, options) {
59063
59894
  const parsed$ = safeParse4(request, (value$) => V2PromptsUpdateRequest$zodSchema.parse(value$), "Input validation failed");
59064
59895
  if (!parsed$.ok) {
59065
59896
  return [parsed$, { status: "invalid" }];
@@ -59135,12 +59966,12 @@ var init_promptsV2PromptsUpdate = __esm(() => {
59135
59966
  });
59136
59967
 
59137
59968
  // src/mcp-server/tools/promptsV2PromptsUpdate.ts
59138
- var args28, tool$promptsV2PromptsUpdate;
59969
+ var args31, tool$promptsV2PromptsUpdate;
59139
59970
  var init_promptsV2PromptsUpdate2 = __esm(() => {
59140
59971
  init_promptsV2PromptsUpdate();
59141
59972
  init_v2promptsupdateop();
59142
59973
  init_tools();
59143
- args28 = {
59974
+ args31 = {
59144
59975
  request: V2PromptsUpdateRequest$zodSchema
59145
59976
  };
59146
59977
  tool$promptsV2PromptsUpdate = {
@@ -59155,7 +59986,7 @@ Updates a prompt's text, topic, platforms, or lifecycle status. Omitted fields r
59155
59986
  openWorldHint: false,
59156
59987
  readOnlyHint: false
59157
59988
  },
59158
- args: args28,
59989
+ args: args31,
59159
59990
  tool: async (client, args, ctx) => {
59160
59991
  const [result] = await promptsV2PromptsUpdate(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
59161
59992
  if (!result.ok) {
@@ -59180,9 +60011,9 @@ var init_createprompttopicrequest = __esm(() => {
59180
60011
 
59181
60012
  // src/funcs/promptsV2PromptTopicsCreate.ts
59182
60013
  function promptsV2PromptTopicsCreate(client$, request, options) {
59183
- return new APIPromise($do34(client$, request, options));
60014
+ return new APIPromise($do38(client$, request, options));
59184
60015
  }
59185
- async function $do34(client$, request, options) {
60016
+ async function $do38(client$, request, options) {
59186
60017
  const parsed$ = safeParse4(request, (value$) => CreatePromptTopicRequest$zodSchema.parse(value$), "Input validation failed");
59187
60018
  if (!parsed$.ok) {
59188
60019
  return [parsed$, { status: "invalid" }];
@@ -59252,12 +60083,12 @@ var init_promptsV2PromptTopicsCreate = __esm(() => {
59252
60083
  });
59253
60084
 
59254
60085
  // src/mcp-server/tools/promptsV2PromptTopicsCreate.ts
59255
- var args29, tool$promptsV2PromptTopicsCreate;
60086
+ var args32, tool$promptsV2PromptTopicsCreate;
59256
60087
  var init_promptsV2PromptTopicsCreate2 = __esm(() => {
59257
60088
  init_promptsV2PromptTopicsCreate();
59258
60089
  init_createprompttopicrequest();
59259
60090
  init_tools();
59260
- args29 = {
60091
+ args32 = {
59261
60092
  request: CreatePromptTopicRequest$zodSchema
59262
60093
  };
59263
60094
  tool$promptsV2PromptTopicsCreate = {
@@ -59272,7 +60103,7 @@ Creates a prompt topic and returns the saved topic.`,
59272
60103
  openWorldHint: false,
59273
60104
  readOnlyHint: false
59274
60105
  },
59275
- args: args29,
60106
+ args: args32,
59276
60107
  tool: async (client, args, ctx) => {
59277
60108
  const [result] = await promptsV2PromptTopicsCreate(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
59278
60109
  if (!result.ok) {
@@ -59304,9 +60135,9 @@ var init_v2prompttopicsdeleteop = __esm(() => {
59304
60135
 
59305
60136
  // src/funcs/promptsV2PromptTopicsDelete.ts
59306
60137
  function promptsV2PromptTopicsDelete(client$, request, options) {
59307
- return new APIPromise($do35(client$, request, options));
60138
+ return new APIPromise($do39(client$, request, options));
59308
60139
  }
59309
- async function $do35(client$, request, options) {
60140
+ async function $do39(client$, request, options) {
59310
60141
  const parsed$ = safeParse4(request, (value$) => V2PromptTopicsDeleteRequest$zodSchema.parse(value$), "Input validation failed");
59311
60142
  if (!parsed$.ok) {
59312
60143
  return [parsed$, { status: "invalid" }];
@@ -59385,12 +60216,12 @@ var init_promptsV2PromptTopicsDelete = __esm(() => {
59385
60216
  });
59386
60217
 
59387
60218
  // src/mcp-server/tools/promptsV2PromptTopicsDelete.ts
59388
- var args30, tool$promptsV2PromptTopicsDelete;
60219
+ var args33, tool$promptsV2PromptTopicsDelete;
59389
60220
  var init_promptsV2PromptTopicsDelete2 = __esm(() => {
59390
60221
  init_promptsV2PromptTopicsDelete();
59391
60222
  init_v2prompttopicsdeleteop();
59392
60223
  init_tools();
59393
- args30 = {
60224
+ args33 = {
59394
60225
  request: V2PromptTopicsDeleteRequest$zodSchema
59395
60226
  };
59396
60227
  tool$promptsV2PromptTopicsDelete = {
@@ -59405,7 +60236,7 @@ Archives a prompt topic. Set \`archive_prompts\` to archive remaining prompts; o
59405
60236
  openWorldHint: false,
59406
60237
  readOnlyHint: false
59407
60238
  },
59408
- args: args30,
60239
+ args: args33,
59409
60240
  tool: async (client, args, ctx) => {
59410
60241
  const [result] = await promptsV2PromptTopicsDelete(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
59411
60242
  if (!result.ok) {
@@ -59421,9 +60252,9 @@ Archives a prompt topic. Set \`archive_prompts\` to archive remaining prompts; o
59421
60252
 
59422
60253
  // src/funcs/promptsV2PromptTopicsList.ts
59423
60254
  function promptsV2PromptTopicsList(client$, options) {
59424
- return new APIPromise($do36(client$, options));
60255
+ return new APIPromise($do40(client$, options));
59425
60256
  }
59426
- async function $do36(client$, options) {
60257
+ async function $do40(client$, options) {
59427
60258
  const path$ = pathToFunc("/v2/prompts/topics")();
59428
60259
  const headers$ = new Headers(compactMap({
59429
60260
  Accept: "application/json"
@@ -59539,9 +60370,9 @@ var init_v2prompttopicsupdateop = __esm(() => {
59539
60370
 
59540
60371
  // src/funcs/promptsV2PromptTopicsUpdate.ts
59541
60372
  function promptsV2PromptTopicsUpdate(client$, request, options) {
59542
- return new APIPromise($do37(client$, request, options));
60373
+ return new APIPromise($do41(client$, request, options));
59543
60374
  }
59544
- async function $do37(client$, request, options) {
60375
+ async function $do41(client$, request, options) {
59545
60376
  const parsed$ = safeParse4(request, (value$) => V2PromptTopicsUpdateRequest$zodSchema.parse(value$), "Input validation failed");
59546
60377
  if (!parsed$.ok) {
59547
60378
  return [parsed$, { status: "invalid" }];
@@ -59617,12 +60448,12 @@ var init_promptsV2PromptTopicsUpdate = __esm(() => {
59617
60448
  });
59618
60449
 
59619
60450
  // src/mcp-server/tools/promptsV2PromptTopicsUpdate.ts
59620
- var args31, tool$promptsV2PromptTopicsUpdate;
60451
+ var args34, tool$promptsV2PromptTopicsUpdate;
59621
60452
  var init_promptsV2PromptTopicsUpdate2 = __esm(() => {
59622
60453
  init_promptsV2PromptTopicsUpdate();
59623
60454
  init_v2prompttopicsupdateop();
59624
60455
  init_tools();
59625
- args31 = {
60456
+ args34 = {
59626
60457
  request: V2PromptTopicsUpdateRequest$zodSchema
59627
60458
  };
59628
60459
  tool$promptsV2PromptTopicsUpdate = {
@@ -59637,7 +60468,7 @@ Renames a prompt topic and returns the updated topic.`,
59637
60468
  openWorldHint: false,
59638
60469
  readOnlyHint: false
59639
60470
  },
59640
- args: args31,
60471
+ args: args34,
59641
60472
  tool: async (client, args, ctx) => {
59642
60473
  const [result] = await promptsV2PromptTopicsUpdate(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
59643
60474
  if (!result.ok) {
@@ -59680,9 +60511,9 @@ var init_createv2repositoryrequest = __esm(() => {
59680
60511
 
59681
60512
  // src/funcs/repositoriesV2RepositoriesCreate.ts
59682
60513
  function repositoriesV2RepositoriesCreate(client$, request, options) {
59683
- return new APIPromise($do38(client$, request, options));
60514
+ return new APIPromise($do42(client$, request, options));
59684
60515
  }
59685
- async function $do38(client$, request, options) {
60516
+ async function $do42(client$, request, options) {
59686
60517
  const parsed$ = safeParse4(request, (value$) => CreateV2RepositoryRequest$zodSchema.parse(value$), "Input validation failed");
59687
60518
  if (!parsed$.ok) {
59688
60519
  return [parsed$, { status: "invalid" }];
@@ -59752,12 +60583,12 @@ var init_repositoriesV2RepositoriesCreate = __esm(() => {
59752
60583
  });
59753
60584
 
59754
60585
  // src/mcp-server/tools/repositoriesV2RepositoriesCreate.ts
59755
- var args32, tool$repositoriesV2RepositoriesCreate;
60586
+ var args35, tool$repositoriesV2RepositoriesCreate;
59756
60587
  var init_repositoriesV2RepositoriesCreate2 = __esm(() => {
59757
60588
  init_repositoriesV2RepositoriesCreate();
59758
60589
  init_createv2repositoryrequest();
59759
60590
  init_tools();
59760
- args32 = {
60591
+ args35 = {
59761
60592
  request: CreateV2RepositoryRequest$zodSchema
59762
60593
  };
59763
60594
  tool$repositoriesV2RepositoriesCreate = {
@@ -59772,7 +60603,7 @@ Adds a reusable Git repository workspace for eval configurations.`,
59772
60603
  openWorldHint: false,
59773
60604
  readOnlyHint: false
59774
60605
  },
59775
- args: args32,
60606
+ args: args35,
59776
60607
  tool: async (client, args, ctx) => {
59777
60608
  const [result] = await repositoriesV2RepositoriesCreate(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
59778
60609
  if (!result.ok) {
@@ -59838,9 +60669,9 @@ var init_v2repositoriesdeleteop = __esm(() => {
59838
60669
 
59839
60670
  // src/funcs/repositoriesV2RepositoriesDelete.ts
59840
60671
  function repositoriesV2RepositoriesDelete(client$, request, options) {
59841
- return new APIPromise($do39(client$, request, options));
60672
+ return new APIPromise($do43(client$, request, options));
59842
60673
  }
59843
- async function $do39(client$, request, options) {
60674
+ async function $do43(client$, request, options) {
59844
60675
  const parsed$ = safeParse4(request, (value$) => V2RepositoriesDeleteRequest$zodSchema.parse(value$), "Input validation failed");
59845
60676
  if (!parsed$.ok) {
59846
60677
  return [parsed$, { status: "invalid" }];
@@ -59915,12 +60746,12 @@ var init_repositoriesV2RepositoriesDelete = __esm(() => {
59915
60746
  });
59916
60747
 
59917
60748
  // src/mcp-server/tools/repositoriesV2RepositoriesDelete.ts
59918
- var args33, tool$repositoriesV2RepositoriesDelete;
60749
+ var args36, tool$repositoriesV2RepositoriesDelete;
59919
60750
  var init_repositoriesV2RepositoriesDelete2 = __esm(() => {
59920
60751
  init_repositoriesV2RepositoriesDelete();
59921
60752
  init_v2repositoriesdeleteop();
59922
60753
  init_tools();
59923
- args33 = {
60754
+ args36 = {
59924
60755
  request: V2RepositoriesDeleteRequest$zodSchema
59925
60756
  };
59926
60757
  tool$repositoriesV2RepositoriesDelete = {
@@ -59935,7 +60766,7 @@ Permanently deletes a repository workspace and returns its last saved state.`,
59935
60766
  openWorldHint: false,
59936
60767
  readOnlyHint: false
59937
60768
  },
59938
- args: args33,
60769
+ args: args36,
59939
60770
  tool: async (client, args, ctx) => {
59940
60771
  const [result] = await repositoriesV2RepositoriesDelete(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
59941
60772
  if (!result.ok) {
@@ -59951,9 +60782,9 @@ Permanently deletes a repository workspace and returns its last saved state.`,
59951
60782
 
59952
60783
  // src/funcs/repositoriesV2RepositoriesList.ts
59953
60784
  function repositoriesV2RepositoriesList(client$, options) {
59954
- return new APIPromise($do40(client$, options));
60785
+ return new APIPromise($do44(client$, options));
59955
60786
  }
59956
- async function $do40(client$, options) {
60787
+ async function $do44(client$, options) {
59957
60788
  const path$ = pathToFunc("/v2/repositories")();
59958
60789
  const headers$ = new Headers(compactMap({
59959
60790
  Accept: "application/json"
@@ -60058,9 +60889,9 @@ var init_v2repositoriesretrieveop = __esm(() => {
60058
60889
 
60059
60890
  // src/funcs/repositoriesV2RepositoriesRetrieve.ts
60060
60891
  function repositoriesV2RepositoriesRetrieve(client$, request, options) {
60061
- return new APIPromise($do41(client$, request, options));
60892
+ return new APIPromise($do45(client$, request, options));
60062
60893
  }
60063
- async function $do41(client$, request, options) {
60894
+ async function $do45(client$, request, options) {
60064
60895
  const parsed$ = safeParse4(request, (value$) => V2RepositoriesRetrieveRequest$zodSchema.parse(value$), "Input validation failed");
60065
60896
  if (!parsed$.ok) {
60066
60897
  return [parsed$, { status: "invalid" }];
@@ -60135,12 +60966,12 @@ var init_repositoriesV2RepositoriesRetrieve = __esm(() => {
60135
60966
  });
60136
60967
 
60137
60968
  // src/mcp-server/tools/repositoriesV2RepositoriesRetrieve.ts
60138
- var args34, tool$repositoriesV2RepositoriesRetrieve;
60969
+ var args37, tool$repositoriesV2RepositoriesRetrieve;
60139
60970
  var init_repositoriesV2RepositoriesRetrieve2 = __esm(() => {
60140
60971
  init_repositoriesV2RepositoriesRetrieve();
60141
60972
  init_v2repositoriesretrieveop();
60142
60973
  init_tools();
60143
- args34 = {
60974
+ args37 = {
60144
60975
  request: V2RepositoriesRetrieveRequest$zodSchema
60145
60976
  };
60146
60977
  tool$repositoriesV2RepositoriesRetrieve = {
@@ -60155,7 +60986,7 @@ Returns one reusable repository workspace.`,
60155
60986
  openWorldHint: false,
60156
60987
  readOnlyHint: true
60157
60988
  },
60158
- args: args34,
60989
+ args: args37,
60159
60990
  tool: async (client, args, ctx) => {
60160
60991
  const [result] = await repositoriesV2RepositoriesRetrieve(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
60161
60992
  if (!result.ok) {
@@ -60215,9 +61046,9 @@ var init_v2repositoriesupdateop = __esm(() => {
60215
61046
 
60216
61047
  // src/funcs/repositoriesV2RepositoriesUpdate.ts
60217
61048
  function repositoriesV2RepositoriesUpdate(client$, request, options) {
60218
- return new APIPromise($do42(client$, request, options));
61049
+ return new APIPromise($do46(client$, request, options));
60219
61050
  }
60220
- async function $do42(client$, request, options) {
61051
+ async function $do46(client$, request, options) {
60221
61052
  const parsed$ = safeParse4(request, (value$) => V2RepositoriesUpdateRequest$zodSchema.parse(value$), "Input validation failed");
60222
61053
  if (!parsed$.ok) {
60223
61054
  return [parsed$, { status: "invalid" }];
@@ -60293,12 +61124,12 @@ var init_repositoriesV2RepositoriesUpdate = __esm(() => {
60293
61124
  });
60294
61125
 
60295
61126
  // src/mcp-server/tools/repositoriesV2RepositoriesUpdate.ts
60296
- var args35, tool$repositoriesV2RepositoriesUpdate;
61127
+ var args38, tool$repositoriesV2RepositoriesUpdate;
60297
61128
  var init_repositoriesV2RepositoriesUpdate2 = __esm(() => {
60298
61129
  init_repositoriesV2RepositoriesUpdate();
60299
61130
  init_v2repositoriesupdateop();
60300
61131
  init_tools();
60301
- args35 = {
61132
+ args38 = {
60302
61133
  request: V2RepositoriesUpdateRequest$zodSchema
60303
61134
  };
60304
61135
  tool$repositoriesV2RepositoriesUpdate = {
@@ -60313,7 +61144,7 @@ Updates a repository workspace; omitted fields remain unchanged.`,
60313
61144
  openWorldHint: false,
60314
61145
  readOnlyHint: false
60315
61146
  },
60316
- args: args35,
61147
+ args: args38,
60317
61148
  tool: async (client, args, ctx) => {
60318
61149
  const [result] = await repositoriesV2RepositoriesUpdate(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
60319
61150
  if (!result.ok) {
@@ -60329,9 +61160,9 @@ Updates a repository workspace; omitted fields remain unchanged.`,
60329
61160
 
60330
61161
  // src/funcs/siteAuditsV2SiteAuditsList.ts
60331
61162
  function siteAuditsV2SiteAuditsList(client$, options) {
60332
- return new APIPromise($do43(client$, options));
61163
+ return new APIPromise($do47(client$, options));
60333
61164
  }
60334
- async function $do43(client$, options) {
61165
+ async function $do47(client$, options) {
60335
61166
  const path$ = pathToFunc("/v2/site-audits")();
60336
61167
  const headers$ = new Headers(compactMap({
60337
61168
  Accept: "application/json"
@@ -60472,9 +61303,9 @@ var init_v2siteauditsretrieveop = __esm(() => {
60472
61303
 
60473
61304
  // src/funcs/siteAuditsV2SiteAuditsRetrieve.ts
60474
61305
  function siteAuditsV2SiteAuditsRetrieve(client$, request, options) {
60475
- return new APIPromise($do44(client$, request, options));
61306
+ return new APIPromise($do48(client$, request, options));
60476
61307
  }
60477
- async function $do44(client$, request, options) {
61308
+ async function $do48(client$, request, options) {
60478
61309
  const parsed$ = safeParse4(request, (value$) => V2SiteAuditsRetrieveRequest$zodSchema.parse(value$), "Input validation failed");
60479
61310
  if (!parsed$.ok) {
60480
61311
  return [parsed$, { status: "invalid" }];
@@ -60549,12 +61380,12 @@ var init_siteAuditsV2SiteAuditsRetrieve = __esm(() => {
60549
61380
  });
60550
61381
 
60551
61382
  // src/mcp-server/tools/siteAuditsV2SiteAuditsRetrieve.ts
60552
- var args36, tool$siteAuditsV2SiteAuditsRetrieve;
61383
+ var args39, tool$siteAuditsV2SiteAuditsRetrieve;
60553
61384
  var init_siteAuditsV2SiteAuditsRetrieve2 = __esm(() => {
60554
61385
  init_siteAuditsV2SiteAuditsRetrieve();
60555
61386
  init_v2siteauditsretrieveop();
60556
61387
  init_tools();
60557
- args36 = {
61388
+ args39 = {
60558
61389
  request: V2SiteAuditsRetrieveRequest$zodSchema
60559
61390
  };
60560
61391
  tool$siteAuditsV2SiteAuditsRetrieve = {
@@ -60569,7 +61400,7 @@ Returns one audited page with its score, priority, supporting evidence, and reco
60569
61400
  openWorldHint: false,
60570
61401
  readOnlyHint: true
60571
61402
  },
60572
- args: args36,
61403
+ args: args39,
60573
61404
  tool: async (client, args, ctx) => {
60574
61405
  const [result] = await siteAuditsV2SiteAuditsRetrieve(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
60575
61406
  if (!result.ok) {
@@ -60585,9 +61416,9 @@ Returns one audited page with its score, priority, supporting evidence, and reco
60585
61416
 
60586
61417
  // src/funcs/skillsV2SkillsList.ts
60587
61418
  function skillsV2SkillsList(client$, options) {
60588
- return new APIPromise($do45(client$, options));
61419
+ return new APIPromise($do49(client$, options));
60589
61420
  }
60590
- async function $do45(client$, options) {
61421
+ async function $do49(client$, options) {
60591
61422
  const path$ = pathToFunc("/v2/skills")();
60592
61423
  const headers$ = new Headers(compactMap({
60593
61424
  Accept: "application/json"
@@ -60677,9 +61508,9 @@ Returns reusable skills available to workflows and eval configurations.`,
60677
61508
 
60678
61509
  // src/funcs/statusV2AuthStatusGet.ts
60679
61510
  function statusV2AuthStatusGet(client$, options) {
60680
- return new APIPromise($do46(client$, options));
61511
+ return new APIPromise($do50(client$, options));
60681
61512
  }
60682
- async function $do46(client$, options) {
61513
+ async function $do50(client$, options) {
60683
61514
  const path$ = pathToFunc("/v2/auth/status")();
60684
61515
  const headers$ = new Headers(compactMap({
60685
61516
  Accept: "application/json"
@@ -60769,9 +61600,9 @@ Validates the API key and returns its organization ID, API key ID, scopes, and a
60769
61600
 
60770
61601
  // src/funcs/statusV2StatusGet.ts
60771
61602
  function statusV2StatusGet(client$, options) {
60772
- return new APIPromise($do47(client$, options));
61603
+ return new APIPromise($do51(client$, options));
60773
61604
  }
60774
- async function $do47(client$, options) {
61605
+ async function $do51(client$, options) {
60775
61606
  const path$ = pathToFunc("/v2/status")();
60776
61607
  const headers$ = new Headers(compactMap({
60777
61608
  Accept: "application/json"
@@ -60881,11 +61712,11 @@ var init_v2publictask = __esm(() => {
60881
61712
  priority: string2().nullable().optional(),
60882
61713
  rank: int().describe("Sort order within the batch, ascending; values need not start at 1."),
60883
61714
  reasoning: string2().nullable().optional(),
60884
- score: number2(),
61715
+ score: number2().nullable().optional().describe("Planner impact score; manually created tasks may be unscored."),
60885
61716
  source: string2().nullable().optional(),
60886
61717
  source_key: string2().describe("A finding key namespaced by source, such as `api_perf:https://example.com/docs/cli`."),
60887
61718
  source_refs: record(string2(), any()).optional().describe("Source references needed to understand or execute the task."),
60888
- source_type: string2().describe("A free-form source type such as `api_performance`; this field is not an enum."),
61719
+ source_type: string2().nullable().optional().describe("A free-form source type such as `api_performance`; manually created tasks may not have a source."),
60889
61720
  status: string2().describe("The workflow status. New tasks start as `new`; `added_to_sheet` and `snoozed` tasks are listed only when `status=all`."),
60890
61721
  task_type: string2().describe("A free-form label such as `Redirect` or `Docs`; this field is not an enum."),
60891
61722
  title: string2(),
@@ -60910,11 +61741,11 @@ var init_v2publictask = __esm(() => {
60910
61741
  priority: nullable(string2()).optional(),
60911
61742
  rank: int(),
60912
61743
  reasoning: nullable(string2()).optional(),
60913
- score: number2(),
61744
+ score: nullable(number2()).optional(),
60914
61745
  source: nullable(string2()).optional(),
60915
61746
  source_key: string2(),
60916
61747
  source_refs: record(string2(), any()).optional(),
60917
- source_type: string2(),
61748
+ source_type: nullable(string2()).optional(),
60918
61749
  status: string2(),
60919
61750
  task_type: string2(),
60920
61751
  title: string2(),
@@ -60991,9 +61822,9 @@ var init_v2taskslistop = __esm(() => {
60991
61822
 
60992
61823
  // src/funcs/tasksV2TasksList.ts
60993
61824
  function tasksV2TasksList(client$, request, options) {
60994
- return new APIPromise($do48(client$, request, options));
61825
+ return new APIPromise($do52(client$, request, options));
60995
61826
  }
60996
- async function $do48(client$, request, options) {
61827
+ async function $do52(client$, request, options) {
60997
61828
  const parsed$ = safeParse4(request, (value$) => V2TasksListRequest$zodSchema.optional().parse(value$), "Input validation failed");
60998
61829
  if (!parsed$.ok) {
60999
61830
  return [parsed$, { status: "invalid" }];
@@ -61067,12 +61898,12 @@ var init_tasksV2TasksList = __esm(() => {
61067
61898
  });
61068
61899
 
61069
61900
  // src/mcp-server/tools/tasksV2TasksList.ts
61070
- var args37, tool$tasksV2TasksList;
61901
+ var args40, tool$tasksV2TasksList;
61071
61902
  var init_tasksV2TasksList2 = __esm(() => {
61072
61903
  init_tasksV2TasksList();
61073
61904
  init_v2taskslistop();
61074
61905
  init_tools();
61075
- args37 = {
61906
+ args40 = {
61076
61907
  request: V2TasksListRequest$zodSchema.optional()
61077
61908
  };
61078
61909
  tool$tasksV2TasksList = {
@@ -61087,7 +61918,7 @@ Returns persisted daily tasks with execution instructions. Includes per-status c
61087
61918
  openWorldHint: false,
61088
61919
  readOnlyHint: true
61089
61920
  },
61090
- args: args37,
61921
+ args: args40,
61091
61922
  tool: async (client, args, ctx) => {
61092
61923
  const [result] = await tasksV2TasksList(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
61093
61924
  if (!result.ok) {
@@ -61118,9 +61949,9 @@ var init_v2tasksretrieveop = __esm(() => {
61118
61949
 
61119
61950
  // src/funcs/tasksV2TasksRetrieve.ts
61120
61951
  function tasksV2TasksRetrieve(client$, request, options) {
61121
- return new APIPromise($do49(client$, request, options));
61952
+ return new APIPromise($do53(client$, request, options));
61122
61953
  }
61123
- async function $do49(client$, request, options) {
61954
+ async function $do53(client$, request, options) {
61124
61955
  const parsed$ = safeParse4(request, (value$) => V2TasksRetrieveRequest$zodSchema.parse(value$), "Input validation failed");
61125
61956
  if (!parsed$.ok) {
61126
61957
  return [parsed$, { status: "invalid" }];
@@ -61195,12 +62026,12 @@ var init_tasksV2TasksRetrieve = __esm(() => {
61195
62026
  });
61196
62027
 
61197
62028
  // src/mcp-server/tools/tasksV2TasksRetrieve.ts
61198
- var args38, tool$tasksV2TasksRetrieve;
62029
+ var args41, tool$tasksV2TasksRetrieve;
61199
62030
  var init_tasksV2TasksRetrieve2 = __esm(() => {
61200
62031
  init_tasksV2TasksRetrieve();
61201
62032
  init_v2tasksretrieveop();
61202
62033
  init_tools();
61203
- args38 = {
62034
+ args41 = {
61204
62035
  request: V2TasksRetrieveRequest$zodSchema
61205
62036
  };
61206
62037
  tool$tasksV2TasksRetrieve = {
@@ -61215,7 +62046,7 @@ Returns one persisted task and its execution prompt.`,
61215
62046
  openWorldHint: false,
61216
62047
  readOnlyHint: true
61217
62048
  },
61218
- args: args38,
62049
+ args: args41,
61219
62050
  tool: async (client, args, ctx) => {
61220
62051
  const [result] = await tasksV2TasksRetrieve(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
61221
62052
  if (!result.ok) {
@@ -61278,9 +62109,9 @@ var init_v2tasksupdateop = __esm(() => {
61278
62109
 
61279
62110
  // src/funcs/tasksV2TasksUpdate.ts
61280
62111
  function tasksV2TasksUpdate(client$, request, options) {
61281
- return new APIPromise($do50(client$, request, options));
62112
+ return new APIPromise($do54(client$, request, options));
61282
62113
  }
61283
- async function $do50(client$, request, options) {
62114
+ async function $do54(client$, request, options) {
61284
62115
  const parsed$ = safeParse4(request, (value$) => V2TasksUpdateRequest$zodSchema.parse(value$), "Input validation failed");
61285
62116
  if (!parsed$.ok) {
61286
62117
  return [parsed$, { status: "invalid" }];
@@ -61356,12 +62187,12 @@ var init_tasksV2TasksUpdate = __esm(() => {
61356
62187
  });
61357
62188
 
61358
62189
  // src/mcp-server/tools/tasksV2TasksUpdate.ts
61359
- var args39, tool$tasksV2TasksUpdate;
62190
+ var args42, tool$tasksV2TasksUpdate;
61360
62191
  var init_tasksV2TasksUpdate2 = __esm(() => {
61361
62192
  init_tasksV2TasksUpdate();
61362
62193
  init_v2tasksupdateop();
61363
62194
  init_tools();
61364
- args39 = {
62195
+ args42 = {
61365
62196
  request: V2TasksUpdateRequest$zodSchema
61366
62197
  };
61367
62198
  tool$tasksV2TasksUpdate = {
@@ -61376,7 +62207,7 @@ Updates a task's status, priority, assignee, or action type. Only sent fields ar
61376
62207
  openWorldHint: false,
61377
62208
  readOnlyHint: false
61378
62209
  },
61379
- args: args39,
62210
+ args: args42,
61380
62211
  tool: async (client, args, ctx) => {
61381
62212
  const [result] = await tasksV2TasksUpdate(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
61382
62213
  if (!result.ok) {
@@ -61464,9 +62295,9 @@ var init_v2visibilitycitationslistop = __esm(() => {
61464
62295
 
61465
62296
  // src/funcs/visibilityV2VisibilityCitationsList.ts
61466
62297
  function visibilityV2VisibilityCitationsList(client$, request, options) {
61467
- return new APIPromise($do51(client$, request, options));
62298
+ return new APIPromise($do55(client$, request, options));
61468
62299
  }
61469
- async function $do51(client$, request, options) {
62300
+ async function $do55(client$, request, options) {
61470
62301
  const parsed$ = safeParse4(request, (value$) => V2VisibilityCitationsListRequest$zodSchema.optional().parse(value$), "Input validation failed");
61471
62302
  if (!parsed$.ok) {
61472
62303
  return [parsed$, { status: "invalid" }];
@@ -61543,12 +62374,12 @@ var init_visibilityV2VisibilityCitationsList = __esm(() => {
61543
62374
  });
61544
62375
 
61545
62376
  // src/mcp-server/tools/visibilityV2VisibilityCitationsList.ts
61546
- var args40, tool$visibilityV2VisibilityCitationsList;
62377
+ var args43, tool$visibilityV2VisibilityCitationsList;
61547
62378
  var init_visibilityV2VisibilityCitationsList2 = __esm(() => {
61548
62379
  init_visibilityV2VisibilityCitationsList();
61549
62380
  init_v2visibilitycitationslistop();
61550
62381
  init_tools();
61551
- args40 = {
62382
+ args43 = {
61552
62383
  request: V2VisibilityCitationsListRequest$zodSchema.optional()
61553
62384
  };
61554
62385
  tool$visibilityV2VisibilityCitationsList = {
@@ -61563,7 +62394,7 @@ Returns URL-level citation counts ranked from most to least cited. Results can b
61563
62394
  openWorldHint: false,
61564
62395
  readOnlyHint: true
61565
62396
  },
61566
- args: args40,
62397
+ args: args43,
61567
62398
  tool: async (client, args, ctx) => {
61568
62399
  const [result] = await visibilityV2VisibilityCitationsList(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
61569
62400
  if (!result.ok) {
@@ -61602,10 +62433,10 @@ var init_publiccurrentvisibilityresponse = __esm(() => {
61602
62433
  date: string2().nullable().optional().describe("Latest prompt-run date in the range."),
61603
62434
  end_date: string2(),
61604
62435
  has_data: boolean2(),
61605
- share_of_voice: number2().nullable().optional().describe("Share of all brand mentions across tracked companies, 0 to 1."),
62436
+ share_of_voice: number2().nullable().optional().describe("Fraction of successful prompt runs that mention the company, from 0 to 1; for example, 0.25 means one in four runs. Failed provider runs are excluded."),
61606
62437
  share_of_voice_rank: int().nullable().optional(),
61607
62438
  start_date: string2(),
61608
- visibility_score: number2().nullable().optional().describe("The visibility score, from 0 to 1."),
62439
+ visibility_score: number2().nullable().optional().describe("Mean recommendation strength across successful prompt runs, from 0 to 1. Each run contributes the company's recommendation strength (0 to 1) when it is mentioned and 0 when it is not. Failed provider runs are excluded."),
61609
62440
  visibility_score_rank: int().nullable().optional()
61610
62441
  });
61611
62442
  });
@@ -61628,9 +62459,9 @@ var init_v2visibilitycurrentretrieveop = __esm(() => {
61628
62459
 
61629
62460
  // src/funcs/visibilityV2VisibilityCurrentRetrieve.ts
61630
62461
  function visibilityV2VisibilityCurrentRetrieve(client$, request, options) {
61631
- return new APIPromise($do52(client$, request, options));
62462
+ return new APIPromise($do56(client$, request, options));
61632
62463
  }
61633
- async function $do52(client$, request, options) {
62464
+ async function $do56(client$, request, options) {
61634
62465
  const parsed$ = safeParse4(request, (value$) => V2VisibilityCurrentRetrieveRequest$zodSchema.optional().parse(value$), "Input validation failed");
61635
62466
  if (!parsed$.ok) {
61636
62467
  return [parsed$, { status: "invalid" }];
@@ -61704,12 +62535,12 @@ var init_visibilityV2VisibilityCurrentRetrieve = __esm(() => {
61704
62535
  });
61705
62536
 
61706
62537
  // src/mcp-server/tools/visibilityV2VisibilityCurrentRetrieve.ts
61707
- var args41, tool$visibilityV2VisibilityCurrentRetrieve;
62538
+ var args44, tool$visibilityV2VisibilityCurrentRetrieve;
61708
62539
  var init_visibilityV2VisibilityCurrentRetrieve2 = __esm(() => {
61709
62540
  init_visibilityV2VisibilityCurrentRetrieve();
61710
62541
  init_v2visibilitycurrentretrieveop();
61711
62542
  init_tools();
61712
- args41 = {
62543
+ args44 = {
61713
62544
  request: V2VisibilityCurrentRetrieveRequest$zodSchema.optional()
61714
62545
  };
61715
62546
  tool$visibilityV2VisibilityCurrentRetrieve = {
@@ -61724,7 +62555,7 @@ Returns visibility score, share of voice, ranks, and date-range metadata for the
61724
62555
  openWorldHint: false,
61725
62556
  readOnlyHint: true
61726
62557
  },
61727
- args: args41,
62558
+ args: args44,
61728
62559
  tool: async (client, args, ctx) => {
61729
62560
  const [result] = await visibilityV2VisibilityCurrentRetrieve(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
61730
62561
  if (!result.ok) {
@@ -61747,9 +62578,9 @@ var init_publiccompetitorrow = __esm(() => {
61747
62578
  average_position: number2().nullable().optional().describe("Mean mention position in AI responses; 1 is first and lower is better. 0 when never mentioned."),
61748
62579
  average_position_rank: int().nullable().optional().describe("Companies with an `average_position` of 0 rank last."),
61749
62580
  company: PublicCompany$zodSchema,
61750
- share_of_voice: number2().nullable().optional().describe("Share of all brand mentions across tracked companies, 0 to 1; rows sum to about 1."),
62581
+ share_of_voice: number2().nullable().optional().describe("Fraction of successful prompt runs that mention the company, from 0 to 1; for example, 0.25 means one in four runs. Failed provider runs are excluded. A run can mention several companies, so rows do not sum to 1."),
61751
62582
  share_of_voice_rank: int().nullable().optional(),
61752
- visibility_score: number2().nullable().optional().describe("The company's visibility score, from 0 to 1. Rows do not sum to 1."),
62583
+ visibility_score: number2().nullable().optional().describe("Mean recommendation strength across successful prompt runs, from 0 to 1. Each run contributes the company's recommendation strength (0 to 1) when it is mentioned and 0 when it is not. Failed provider runs are excluded. Rows do not sum to 1."),
61753
62584
  visibility_score_rank: int().nullable().optional()
61754
62585
  });
61755
62586
  });
@@ -61810,10 +62641,10 @@ var init_publicvisibilitysnapshot = __esm(() => {
61810
62641
  PublicVisibilitySnapshot$zodSchema = object({
61811
62642
  average_position: number2().nullable().optional().describe("Mean mention position in AI responses; 1 is first and lower is better."),
61812
62643
  average_position_rank: int().nullable().optional(),
61813
- share_of_voice: number2().nullable().optional().describe("Share of all brand mentions across tracked companies, 0 to 1."),
62644
+ share_of_voice: number2().nullable().optional().describe("Fraction of successful prompt runs that mention the company, from 0 to 1; for example, 0.25 means one in four runs. Failed provider runs are excluded."),
61814
62645
  share_of_voice_rank: int().nullable().optional(),
61815
62646
  total_citations: int().default(0).describe("Owned-domain citations across all platforms; 0, not `null`, with no prompt runs."),
61816
- visibility_score: number2().nullable().optional().describe("The visibility score, from 0 to 1."),
62647
+ visibility_score: number2().nullable().optional().describe("Mean recommendation strength across successful prompt runs, from 0 to 1. Each run contributes the company's recommendation strength (0 to 1) when it is mentioned and 0 when it is not. Failed provider runs are excluded."),
61817
62648
  visibility_score_rank: int().nullable().optional()
61818
62649
  });
61819
62650
  });
@@ -61867,9 +62698,9 @@ var init_v2visibilityoverviewretrieveop = __esm(() => {
61867
62698
 
61868
62699
  // src/funcs/visibilityV2VisibilityOverviewRetrieve.ts
61869
62700
  function visibilityV2VisibilityOverviewRetrieve(client$, request, options) {
61870
- return new APIPromise($do53(client$, request, options));
62701
+ return new APIPromise($do57(client$, request, options));
61871
62702
  }
61872
- async function $do53(client$, request, options) {
62703
+ async function $do57(client$, request, options) {
61873
62704
  const parsed$ = safeParse4(request, (value$) => V2VisibilityOverviewRetrieveRequest$zodSchema.optional().parse(value$), "Input validation failed");
61874
62705
  if (!parsed$.ok) {
61875
62706
  return [parsed$, { status: "invalid" }];
@@ -61943,12 +62774,12 @@ var init_visibilityV2VisibilityOverviewRetrieve = __esm(() => {
61943
62774
  });
61944
62775
 
61945
62776
  // src/mcp-server/tools/visibilityV2VisibilityOverviewRetrieve.ts
61946
- var args42, tool$visibilityV2VisibilityOverviewRetrieve;
62777
+ var args45, tool$visibilityV2VisibilityOverviewRetrieve;
61947
62778
  var init_visibilityV2VisibilityOverviewRetrieve2 = __esm(() => {
61948
62779
  init_visibilityV2VisibilityOverviewRetrieve();
61949
62780
  init_v2visibilityoverviewretrieveop();
61950
62781
  init_tools();
61951
- args42 = {
62782
+ args45 = {
61952
62783
  request: V2VisibilityOverviewRetrieveRequest$zodSchema.optional()
61953
62784
  };
61954
62785
  tool$visibilityV2VisibilityOverviewRetrieve = {
@@ -61963,7 +62794,7 @@ Returns aggregate snapshot, daily trends, per-platform breakdowns, and competito
61963
62794
  openWorldHint: false,
61964
62795
  readOnlyHint: true
61965
62796
  },
61966
- args: args42,
62797
+ args: args45,
61967
62798
  tool: async (client, args, ctx) => {
61968
62799
  const [result] = await visibilityV2VisibilityOverviewRetrieve(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
61969
62800
  if (!result.ok) {
@@ -61997,8 +62828,8 @@ var init_publicpromptvisibilitymetrics = __esm(() => {
61997
62828
  average_position: number2().nullable().optional().describe("Mean mention position in AI responses; 1 is first and lower is better. 0 when runs produced no mentions."),
61998
62829
  citation_count: int().default(0).describe("Owned-domain citations only."),
61999
62830
  run_count: int().default(0).describe("One per prompt, platform, and run date."),
62000
- share_of_voice: number2().nullable().optional().describe("Share of all brand mentions across tracked companies, 0 to 1."),
62001
- visibility_score: number2().nullable().optional().describe("The visibility score, from 0 to 1."),
62831
+ share_of_voice: number2().nullable().optional().describe("Fraction of successful prompt runs that mention the company, from 0 to 1; for example, 0.25 means one in four runs. Failed provider runs are excluded."),
62832
+ visibility_score: number2().nullable().optional().describe("Mean recommendation strength across successful prompt runs, from 0 to 1. Each run contributes the company's recommendation strength (0 to 1) when it is mentioned and 0 when it is not. Failed provider runs are excluded."),
62002
62833
  visibility_score_rank: int().nullable().optional()
62003
62834
  });
62004
62835
  });
@@ -62077,9 +62908,9 @@ var init_v2visibilitypromptsretrieveop = __esm(() => {
62077
62908
 
62078
62909
  // src/funcs/visibilityV2VisibilityPromptsRetrieve.ts
62079
62910
  function visibilityV2VisibilityPromptsRetrieve(client$, request, options) {
62080
- return new APIPromise($do54(client$, request, options));
62911
+ return new APIPromise($do58(client$, request, options));
62081
62912
  }
62082
- async function $do54(client$, request, options) {
62913
+ async function $do58(client$, request, options) {
62083
62914
  const parsed$ = safeParse4(request, (value$) => V2VisibilityPromptsRetrieveRequest$zodSchema.optional().parse(value$), "Input validation failed");
62084
62915
  if (!parsed$.ok) {
62085
62916
  return [parsed$, { status: "invalid" }];
@@ -62153,12 +62984,12 @@ var init_visibilityV2VisibilityPromptsRetrieve = __esm(() => {
62153
62984
  });
62154
62985
 
62155
62986
  // src/mcp-server/tools/visibilityV2VisibilityPromptsRetrieve.ts
62156
- var args43, tool$visibilityV2VisibilityPromptsRetrieve;
62987
+ var args46, tool$visibilityV2VisibilityPromptsRetrieve;
62157
62988
  var init_visibilityV2VisibilityPromptsRetrieve2 = __esm(() => {
62158
62989
  init_visibilityV2VisibilityPromptsRetrieve();
62159
62990
  init_v2visibilitypromptsretrieveop();
62160
62991
  init_tools();
62161
- args43 = {
62992
+ args46 = {
62162
62993
  request: V2VisibilityPromptsRetrieveRequest$zodSchema.optional()
62163
62994
  };
62164
62995
  tool$visibilityV2VisibilityPromptsRetrieve = {
@@ -62173,7 +63004,7 @@ Returns per-topic and per-prompt visibility metrics including score, rank, share
62173
63004
  openWorldHint: false,
62174
63005
  readOnlyHint: true
62175
63006
  },
62176
- args: args43,
63007
+ args: args46,
62177
63008
  tool: async (client, args, ctx) => {
62178
63009
  const [result] = await visibilityV2VisibilityPromptsRetrieve(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
62179
63010
  if (!result.ok) {
@@ -62197,7 +63028,7 @@ function createMCPServer(deps) {
62197
63028
  sizes: ["any"]
62198
63029
  }],
62199
63030
  name: "LightsageMcp",
62200
- version: "0.3.1"
63031
+ version: "0.4.1"
62201
63032
  });
62202
63033
  const getClient = deps.getSDK || (() => new LightsageMcpCore({
62203
63034
  security: deps.security,
@@ -62245,6 +63076,7 @@ function createMCPServer(deps) {
62245
63076
  tool(tool$siteAuditsV2SiteAuditsRetrieve);
62246
63077
  tool(tool$evalsV2EvalsList);
62247
63078
  tool(tool$evalsV2EvalsCreate);
63079
+ tool(tool$evalsV2EvalsBatchCreate);
62248
63080
  tool(tool$evalsV2EvalsRetrieve);
62249
63081
  tool(tool$evalsV2EvalsUpdate);
62250
63082
  tool(tool$evalsV2EvalsDelete);
@@ -62254,12 +63086,15 @@ function createMCPServer(deps) {
62254
63086
  tool(tool$evalsV2EvalConfigurationsUpdate);
62255
63087
  tool(tool$evalsV2EvalConfigurationsDelete);
62256
63088
  tool(tool$evalsRun);
62257
- tool(tool$evalsAnalysis);
63089
+ tool(tool$evalsCancel);
62258
63090
  tool(tool$evalsTrace);
63091
+ tool(tool$evalsGet);
63092
+ tool(tool$evalTemplatesV2EvalTemplatesList);
63093
+ tool(tool$evalTemplatesV2EvalTemplatesCreate);
63094
+ tool(tool$evalTemplatesV2EvalTemplatesRetrieve);
63095
+ tool(tool$evalTemplatesV2EvalTemplatesUpdate);
63096
+ tool(tool$evalTemplatesV2EvalTemplatesDelete);
62259
63097
  tool(tool$evalRunsV2EvalRunsList);
62260
- tool(tool$evalRunsV2EvalRunsRetrieve);
62261
- tool(tool$evalResultsV2EvalResultsList);
62262
- tool(tool$evalResultsV2EvalResultsRetrieve);
62263
63098
  tool(tool$agentsV2AgentsList);
62264
63099
  tool(tool$repositoriesV2RepositoriesList);
62265
63100
  tool(tool$repositoriesV2RepositoriesCreate);
@@ -62286,11 +63121,9 @@ var init_server2 = __esm(() => {
62286
63121
  init_contentV2ContentFinalize2();
62287
63122
  init_contentV2ContentRetrieve2();
62288
63123
  init_environmentsV2EnvironmentsList2();
62289
- init_evalResultsV2EvalResultsList2();
62290
- init_evalResultsV2EvalResultsRetrieve2();
62291
63124
  init_evalRunsV2EvalRunsList2();
62292
- init_evalRunsV2EvalRunsRetrieve2();
62293
- init_evalsAnalysis2();
63125
+ init_evalsCancel2();
63126
+ init_evalsGet2();
62294
63127
  init_evalsRun2();
62295
63128
  init_evalsTrace2();
62296
63129
  init_evalsV2EvalConfigurationsCreate2();
@@ -62298,11 +63131,17 @@ var init_server2 = __esm(() => {
62298
63131
  init_evalsV2EvalConfigurationsList2();
62299
63132
  init_evalsV2EvalConfigurationsRetrieve2();
62300
63133
  init_evalsV2EvalConfigurationsUpdate2();
63134
+ init_evalsV2EvalsBatchCreate2();
62301
63135
  init_evalsV2EvalsCreate2();
62302
63136
  init_evalsV2EvalsDelete2();
62303
63137
  init_evalsV2EvalsList2();
62304
63138
  init_evalsV2EvalsRetrieve2();
62305
63139
  init_evalsV2EvalsUpdate2();
63140
+ init_evalTemplatesV2EvalTemplatesCreate2();
63141
+ init_evalTemplatesV2EvalTemplatesDelete2();
63142
+ init_evalTemplatesV2EvalTemplatesList2();
63143
+ init_evalTemplatesV2EvalTemplatesRetrieve2();
63144
+ init_evalTemplatesV2EvalTemplatesUpdate2();
62306
63145
  init_mcpServersV2MCPServersList2();
62307
63146
  init_opportunitiesV2OpportunitiesList2();
62308
63147
  init_opportunitiesV2OpportunitiesRetrieve2();
@@ -62408,13 +63247,13 @@ Renames a prompt topic and returns the updated topic.`
62408
63247
  name: "prompts-v2-prompt-runs-list",
62409
63248
  description: `List Prompt Runs
62410
63249
 
62411
- Returns stored prompt runs without exposing raw runtime or provider configuration. Filterable by prompt, platform, and date range.`
63250
+ Returns stored prompt runs with limit-and-offset pagination, without exposing raw runtime or provider configuration. Filter by prompt, platform, or date range.`
62412
63251
  },
62413
63252
  {
62414
63253
  name: "prompts-v2-prompt-runs-retrieve",
62415
63254
  description: `Retrieve Prompt Run
62416
63255
 
62417
- Returns one stored prompt run, including its mentions and citations.`
63256
+ Returns one stored prompt run, including mentions, citations, and any sanitized coding-agent trace captured for the run.`
62418
63257
  },
62419
63258
  {
62420
63259
  name: "prompts-v2-prompts-retrieve",
@@ -62498,9 +63337,11 @@ Returns all active eval definitions owned by the organization.`
62498
63337
  },
62499
63338
  {
62500
63339
  name: "evals-v2-evals-create",
62501
- description: `Create Eval
62502
-
62503
- Creates a reusable simulation definition with one or more task prompts and judges. Optional agents, persona, CLI installations, MCP servers, repeat count, and tags are saved with the eval. Creating the eval does not start a run and does not accept secret environment values.`
63340
+ description: "Create Eval\n\nCreates a reusable simulation definition with one or more task prompts and judges. Optional agents, persona, CLI installations, MCP servers, setup commands, cleanup commands, grounding policy, repeat count, and tags are saved with the eval. `search_grounding` supports `disabled`, `enabled`, and `default`; `default` preserves normal platform behavior. Creating the eval does not start a run and does not accept secret environment values."
63341
+ },
63342
+ {
63343
+ name: "evals-v2-evals-batch-create",
63344
+ description: "Create evals from a template\n\nStep 2 of the reusable template workflow. Provide the `id` and `version` returned by `POST /v2/eval-templates` as `template_id` and `expected_template_version`, then provide only prompt-specific fields in `evals`. The template supplies only the settings named in its `included_components`; omitted settings keep the values supplied by the creation workflow or its defaults. Included values are copied at creation time, so later template changes do not alter the created evals. Competitor selections create one additional eval per selected competitor for every input item, and the response groups each base eval with those variants. Brand-scoped runtime resources are validated against the organization's primary brand before any eval is created."
62504
63345
  },
62505
63346
  {
62506
63347
  name: "evals-v2-evals-retrieve",
@@ -62510,7 +63351,7 @@ Returns one reusable eval simulation definition.`
62510
63351
  },
62511
63352
  {
62512
63353
  name: "evals-v2-evals-update",
62513
- description: "Update Eval\n\nUpdates a reusable eval simulation definition. Omitted fields remain unchanged. Empty `agent`, `clis`, `mcps`, or `tags` arrays clear those collections, and `persona: null` clears the saved persona."
63354
+ description: "Update Eval\n\nUpdates a reusable eval simulation definition. Omitted fields remain unchanged. Empty `agent`, `clis`, `mcps`, `setup_commands`, `cleanup_commands`, or `tags` arrays clear those collections, and `persona: null` clears the saved persona."
62514
63355
  },
62515
63356
  {
62516
63357
  name: "evals-v2-evals-delete",
@@ -62550,39 +63391,53 @@ Permanently deletes an eval configuration and returns its last saved state.`
62550
63391
  },
62551
63392
  {
62552
63393
  name: "evals-run",
62553
- description: "Start an eval run\n\nQueues a durable eval job and returns immediately with a polling ID. Poll `GET /v2/eval-runs/{run_id}` until its status reaches a terminal value.\n\nSend one of these flat request forms:\n\n- **Saved eval:** provide `eval_id`; optionally replace its prompts, persona, agents, CLIs, MCPs, or runs for this launch. A saved configuration can still provide repository, environment, skills, and legacy tool selections.\n- **Direct create-and-run:** provide `prompts`, `judges`, and `agent`. You may also provide a persona, tags, repository, write-only environment variables, saved or inline CLIs and MCP servers, and skills.\n\nThe run fans out across prompts × agents × runs. Persona is shared metadata and does not multiply attempts. No separate `mode` field is required."
63394
+ description: "Start an eval run\n\nQueues a durable eval job and returns immediately with a polling ID. Poll `GET /v2/eval-runs/{run_id}` until its status reaches a terminal value; once finished it also returns the results and analysis. Follow a run live with the monitor operation.\n\nSend one of these flat request forms:\n\n- **Saved eval:** provide `eval_id`; optionally replace its prompts, persona, agents, CLIs, MCPs, setup commands, cleanup commands, or runs for this launch. A saved configuration can still provide repository, environment, skills, and legacy tool selections.\n- **Direct create-and-run:** provide `prompts`, `judges`, and `agent`. You may also provide a persona, tags, repository, write-only environment variables, saved or inline CLIs and MCP servers, skills, setup commands, and cleanup commands.\n\nThe run fans out across prompts × agents × runs. Persona is shared metadata and does not multiply attempts. No separate `mode` field is required. If the organization lacks the estimated credits, the request fails with `402` before dispatching work.\n\nA saved or inline environment containing an empty or placeholder credential is rejected with `400` before any evaluation is dispatched."
62554
63395
  },
62555
63396
  {
62556
- name: "evals-analysis",
62557
- description: `Retrieve eval run analysis
63397
+ name: "evals-cancel",
63398
+ description: `Cancel an eval run
62558
63399
 
62559
- Returns the persisted analysis produced for each completed attempt in one eval run. The analysis includes the verdict, outcome explanation, findings, bottleneck, and analysis sections. Active runs return the analysis available so far together with current run status and progress.`
63400
+ Marks an active eval run as cancelled and prevents queued attempts from starting. Running sandbox attempts are stopped asynchronously. Requests already sent to an external provider cannot be recalled, but any later response does not change the run from cancelled. Repeating this request for an already-cancelled run is safe.`
62560
63401
  },
62561
63402
  {
62562
63403
  name: "evals-trace",
62563
63404
  description: "Retrieve an eval run trace\n\nReturns a finite page of sanitized execution events for one eval run. Use `next_after_id` as `after_id` to retrieve the next page. Use the monitor operation instead when following a run live."
62564
63405
  },
62565
63406
  {
62566
- name: "eval-runs-v2-eval-runs-list",
62567
- description: "List Runs\n\nReturns jobs created through `POST /v2/eval-runs`, ordered newest first, with lifecycle state and progress."
63407
+ name: "evals-get",
63408
+ description: `Retrieve an eval run
63409
+
63410
+ Returns the run's current status and progress. Once its attempts finish, the response also includes their results and analysis. Poll this until the status reaches a terminal value. Use the trace operation for the raw execution event log, or the monitor operation to follow a run live.`
62568
63411
  },
62569
63412
  {
62570
- name: "eval-runs-v2-eval-runs-retrieve",
62571
- description: `Retrieve Run
63413
+ name: "eval-templates-v2-eval-templates-list",
63414
+ description: `List Eval templates
62572
63415
 
62573
- Returns the status and progress of an eval job. Poll until the status reaches a terminal value.`
63416
+ Returns active reusable Eval templates owned by the authenticated organization. Templates contain reusable configuration and never task prompts.`
62574
63417
  },
62575
63418
  {
62576
- name: "eval-results-v2-eval-results-list",
62577
- description: `List Results
63419
+ name: "eval-templates-v2-eval-templates-create",
63420
+ description: "Create an Eval template\n\nStep 1 of the reusable template workflow. Creates an organization-owned Eval template containing only the configuration components selected in `included_components`. Do not include task prompts in the template. The response returns the template `id` and `version`; pass them as `template_id` and `expected_template_version` to `POST /v2/evals/batch` with each eval's name and prompts. Applying a template copies its values into the new evals without creating a live link."
63421
+ },
63422
+ {
63423
+ name: "eval-templates-v2-eval-templates-retrieve",
63424
+ description: `Retrieve an Eval template
62578
63425
 
62579
- Returns completed custom-eval results from all supported run flows, ordered newest first. Result IDs are different from eval run IDs.`
63426
+ Returns one active reusable Eval template.`
63427
+ },
63428
+ {
63429
+ name: "eval-templates-v2-eval-templates-update",
63430
+ description: "Update an Eval template\n\nUpdates an Eval template. Supply `expected_template_version` to reject an edit based on stale data. Existing evals are never changed."
62580
63431
  },
62581
63432
  {
62582
- name: "eval-results-v2-eval-results-retrieve",
62583
- description: `Retrieve Result
63433
+ name: "eval-templates-v2-eval-templates-delete",
63434
+ description: `Delete an Eval template
62584
63435
 
62585
- Returns one completed custom-eval result with its output and diagnostic details.`
63436
+ Soft-deletes an Eval template. Evals previously created from it remain independent and continue to run with their copied configuration.`
63437
+ },
63438
+ {
63439
+ name: "eval-runs-v2-eval-runs-list",
63440
+ description: "List Runs\n\nReturns jobs created through `POST /v2/eval-runs`, ordered newest first, with lifecycle state and progress."
62586
63441
  },
62587
63442
  {
62588
63443
  name: "agents-v2-agents-list",
@@ -63548,7 +64403,7 @@ http_headers = { "lightsage-api-key-auth" = "YOUR_LIGHTSAGE_API_KEY_AUTH" }`;
63548
64403
  <h1>Instructions</h1>
63549
64404
  <p>One-click installation for Claude Desktop users</p>
63550
64405
  <div class="instruction-item">
63551
- <a href="https://github.com/lightsagehq/mcp/releases/download/v0.3.1/mcp-server.mcpb" download="mcp-server.mcpb" class="action-button header-action" style="display: inline-flex; margin-bottom: 16px;">
64406
+ <a href="https://github.com/lightsagehq/mcp/releases/download/v0.4.1/mcp-server.mcpb" download="mcp-server.mcpb" class="action-button header-action" style="display: inline-flex; margin-bottom: 16px;">
63552
64407
  \uD83D\uDCE5 Download MCP Bundle
63553
64408
  </a>
63554
64409
  </div>
@@ -66426,7 +67281,7 @@ var routes = buildRouteMap({
66426
67281
  var app = buildApplication(routes, {
66427
67282
  name: "mcp",
66428
67283
  versionInfo: {
66429
- currentVersion: "0.3.1"
67284
+ currentVersion: "0.4.1"
66430
67285
  }
66431
67286
  });
66432
67287
  run(app, process3.argv.slice(2), buildContext(process3));
@@ -66434,5 +67289,5 @@ export {
66434
67289
  app
66435
67290
  };
66436
67291
 
66437
- //# debugId=D144B9026D7DE38264756E2164756E21
67292
+ //# debugId=9D9BF9F9417C659E64756E2164756E21
66438
67293
  //# sourceMappingURL=mcp-server.js.map