@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.
- package/README.md +2 -2
- package/api/well-known-openai-apps-challenge.js +25 -0
- package/bin/mcp-server.js +1878 -1023
- package/bin/mcp-server.js.map +77 -58
- package/esm/funcs/evalTemplatesV2EvalTemplatesCreate.d.ts +16 -0
- package/esm/funcs/evalTemplatesV2EvalTemplatesCreate.d.ts.map +1 -0
- package/esm/funcs/evalTemplatesV2EvalTemplatesCreate.js +83 -0
- package/esm/funcs/evalTemplatesV2EvalTemplatesCreate.js.map +1 -0
- package/esm/funcs/evalTemplatesV2EvalTemplatesDelete.d.ts +16 -0
- package/esm/funcs/evalTemplatesV2EvalTemplatesDelete.d.ts.map +1 -0
- package/esm/funcs/{evalResultsV2EvalResultsList.js → evalTemplatesV2EvalTemplatesDelete.js} +17 -16
- package/esm/funcs/evalTemplatesV2EvalTemplatesDelete.js.map +1 -0
- package/esm/funcs/evalTemplatesV2EvalTemplatesList.d.ts +15 -0
- package/esm/funcs/evalTemplatesV2EvalTemplatesList.d.ts.map +1 -0
- package/esm/funcs/evalTemplatesV2EvalTemplatesList.js +72 -0
- package/esm/funcs/evalTemplatesV2EvalTemplatesList.js.map +1 -0
- package/esm/funcs/evalTemplatesV2EvalTemplatesRetrieve.d.ts +16 -0
- package/esm/funcs/evalTemplatesV2EvalTemplatesRetrieve.d.ts.map +1 -0
- package/esm/funcs/{evalResultsV2EvalResultsRetrieve.js → evalTemplatesV2EvalTemplatesRetrieve.js} +10 -10
- package/esm/funcs/evalTemplatesV2EvalTemplatesRetrieve.js.map +1 -0
- package/esm/funcs/evalTemplatesV2EvalTemplatesUpdate.d.ts +16 -0
- package/esm/funcs/evalTemplatesV2EvalTemplatesUpdate.d.ts.map +1 -0
- package/esm/funcs/evalTemplatesV2EvalTemplatesUpdate.js +89 -0
- package/esm/funcs/evalTemplatesV2EvalTemplatesUpdate.js.map +1 -0
- package/esm/funcs/evalsCancel.d.ts +16 -0
- package/esm/funcs/evalsCancel.d.ts.map +1 -0
- package/esm/funcs/{evalsAnalysis.js → evalsCancel.js} +10 -10
- package/esm/funcs/evalsCancel.js.map +1 -0
- package/esm/funcs/evalsGet.d.ts +16 -0
- package/esm/funcs/evalsGet.d.ts.map +1 -0
- package/esm/funcs/{evalRunsV2EvalRunsRetrieve.js → evalsGet.js} +5 -5
- package/esm/funcs/evalsGet.js.map +1 -0
- package/esm/funcs/evalsRun.d.ts +6 -4
- package/esm/funcs/evalsRun.d.ts.map +1 -1
- package/esm/funcs/evalsRun.js +6 -4
- package/esm/funcs/evalsRun.js.map +1 -1
- package/esm/funcs/evalsV2EvalsBatchCreate.d.ts +16 -0
- package/esm/funcs/evalsV2EvalsBatchCreate.d.ts.map +1 -0
- package/esm/funcs/evalsV2EvalsBatchCreate.js +83 -0
- package/esm/funcs/evalsV2EvalsBatchCreate.js.map +1 -0
- package/esm/funcs/evalsV2EvalsCreate.d.ts +1 -1
- package/esm/funcs/evalsV2EvalsCreate.js +1 -1
- package/esm/funcs/evalsV2EvalsUpdate.d.ts +1 -1
- package/esm/funcs/evalsV2EvalsUpdate.js +1 -1
- package/esm/funcs/promptsV2PromptRunsList.d.ts +1 -1
- package/esm/funcs/promptsV2PromptRunsList.js +1 -1
- package/esm/funcs/promptsV2PromptRunsRetrieve.d.ts +1 -1
- package/esm/funcs/promptsV2PromptRunsRetrieve.js +1 -1
- package/esm/landing-page.js +1 -1
- package/esm/lib/config.d.ts +3 -3
- package/esm/lib/config.js +3 -3
- package/esm/lib/config.js.map +1 -1
- package/esm/mcp-server/mcp-server.js +1 -1
- package/esm/mcp-server/server.d.ts.map +1 -1
- package/esm/mcp-server/server.js +17 -9
- package/esm/mcp-server/server.js.map +1 -1
- package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesCreate.d.ts +7 -0
- package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesCreate.d.ts.map +1 -0
- package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesCreate.js +35 -0
- package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesCreate.js.map +1 -0
- package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesDelete.d.ts +7 -0
- package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesDelete.d.ts.map +1 -0
- package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesDelete.js +35 -0
- package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesDelete.js.map +1 -0
- package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesList.d.ts +3 -0
- package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesList.d.ts.map +1 -0
- package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesList.js +30 -0
- package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesList.js.map +1 -0
- package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesRetrieve.d.ts +7 -0
- package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesRetrieve.d.ts.map +1 -0
- package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesRetrieve.js +35 -0
- package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesRetrieve.js.map +1 -0
- package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesUpdate.d.ts +7 -0
- package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesUpdate.d.ts.map +1 -0
- package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesUpdate.js +35 -0
- package/esm/mcp-server/tools/evalTemplatesV2EvalTemplatesUpdate.js.map +1 -0
- package/esm/mcp-server/tools/evalsCancel.d.ts +7 -0
- package/esm/mcp-server/tools/evalsCancel.d.ts.map +1 -0
- package/esm/mcp-server/tools/evalsCancel.js +35 -0
- package/esm/mcp-server/tools/evalsCancel.js.map +1 -0
- package/esm/mcp-server/tools/{evalRunsV2EvalRunsRetrieve.d.ts → evalsGet.d.ts} +2 -2
- package/esm/mcp-server/tools/evalsGet.d.ts.map +1 -0
- package/esm/mcp-server/tools/{evalRunsV2EvalRunsRetrieve.js → evalsGet.js} +8 -8
- package/esm/mcp-server/tools/evalsGet.js.map +1 -0
- package/esm/mcp-server/tools/evalsRun.d.ts.map +1 -1
- package/esm/mcp-server/tools/evalsRun.js +6 -4
- package/esm/mcp-server/tools/evalsRun.js.map +1 -1
- package/esm/mcp-server/tools/evalsV2EvalsBatchCreate.d.ts +7 -0
- package/esm/mcp-server/tools/evalsV2EvalsBatchCreate.d.ts.map +1 -0
- package/esm/mcp-server/tools/evalsV2EvalsBatchCreate.js +35 -0
- package/esm/mcp-server/tools/evalsV2EvalsBatchCreate.js.map +1 -0
- package/esm/mcp-server/tools/evalsV2EvalsCreate.js +1 -1
- package/esm/mcp-server/tools/evalsV2EvalsCreate.js.map +1 -1
- package/esm/mcp-server/tools/evalsV2EvalsUpdate.js +1 -1
- package/esm/mcp-server/tools/evalsV2EvalsUpdate.js.map +1 -1
- package/esm/mcp-server/tools/promptsV2PromptRunsList.js +1 -1
- package/esm/mcp-server/tools/promptsV2PromptRunsList.js.map +1 -1
- package/esm/mcp-server/tools/promptsV2PromptRunsRetrieve.js +1 -1
- package/esm/mcp-server/tools/promptsV2PromptRunsRetrieve.js.map +1 -1
- package/esm/models/createevaltemplatebatchrequest.d.ts +12 -0
- package/esm/models/createevaltemplatebatchrequest.d.ts.map +1 -0
- package/esm/models/createevaltemplatebatchrequest.js +12 -0
- package/esm/models/createevaltemplatebatchrequest.js.map +1 -0
- package/esm/models/createevaltemplaterequest.d.ts +79 -0
- package/esm/models/createevaltemplaterequest.d.ts.map +1 -0
- package/esm/models/createevaltemplaterequest.js +74 -0
- package/esm/models/createevaltemplaterequest.js.map +1 -0
- package/esm/models/createv2evalconfigurationrequest.d.ts +2 -0
- package/esm/models/createv2evalconfigurationrequest.d.ts.map +1 -1
- package/esm/models/createv2evalconfigurationrequest.js +2 -0
- package/esm/models/createv2evalconfigurationrequest.js.map +1 -1
- package/esm/models/createv2evalrequest.d.ts +5 -5
- package/esm/models/createv2evalrequest.d.ts.map +1 -1
- package/esm/models/createv2evalrequest.js +5 -10
- package/esm/models/createv2evalrequest.js.map +1 -1
- package/esm/models/createv2evalrunrequest.d.ts +5 -5
- package/esm/models/createv2evalrunrequest.d.ts.map +1 -1
- package/esm/models/createv2evalrunrequest.js +6 -10
- package/esm/models/createv2evalrunrequest.js.map +1 -1
- package/esm/models/evaltemplate.d.ts +83 -0
- package/esm/models/evaltemplate.d.ts.map +1 -0
- package/esm/models/evaltemplate.js +81 -0
- package/esm/models/evaltemplate.js.map +1 -0
- package/esm/models/evaltemplatebatchcreatedeval.d.ts +11 -0
- package/esm/models/evaltemplatebatchcreatedeval.d.ts.map +1 -0
- package/esm/models/evaltemplatebatchcreatedeval.js +14 -0
- package/esm/models/evaltemplatebatchcreatedeval.js.map +1 -0
- package/esm/models/evaltemplatebatchitem.d.ts +32 -0
- package/esm/models/evaltemplatebatchitem.d.ts.map +1 -0
- package/esm/models/evaltemplatebatchitem.js +27 -0
- package/esm/models/evaltemplatebatchitem.js.map +1 -0
- package/esm/models/evaltemplatebatchresponse.d.ts +12 -0
- package/esm/models/evaltemplatebatchresponse.d.ts.map +1 -0
- package/esm/models/evaltemplatebatchresponse.js +15 -0
- package/esm/models/evaltemplatebatchresponse.js.map +1 -0
- package/esm/models/evaltemplatebatchresult.d.ts +10 -0
- package/esm/models/evaltemplatebatchresult.d.ts.map +1 -0
- package/esm/models/evaltemplatebatchresult.js +13 -0
- package/esm/models/evaltemplatebatchresult.js.map +1 -0
- package/esm/models/evaltemplateruntime.d.ts +20 -0
- package/esm/models/evaltemplateruntime.d.ts.map +1 -0
- package/esm/models/evaltemplateruntime.js +21 -0
- package/esm/models/evaltemplateruntime.js.map +1 -0
- package/esm/models/evaltemplateschedule.d.ts +28 -0
- package/esm/models/evaltemplateschedule.d.ts.map +1 -0
- package/esm/models/evaltemplateschedule.js +24 -0
- package/esm/models/evaltemplateschedule.js.map +1 -0
- package/esm/models/evaltemplatesresponse.d.ts +8 -0
- package/esm/models/evaltemplatesresponse.d.ts.map +1 -0
- package/esm/models/evaltemplatesresponse.js +11 -0
- package/esm/models/evaltemplatesresponse.js.map +1 -0
- package/esm/models/graderdefinition.d.ts +14 -0
- package/esm/models/graderdefinition.d.ts.map +1 -0
- package/esm/models/graderdefinition.js +17 -0
- package/esm/models/graderdefinition.js.map +1 -0
- package/esm/models/publiccompetitorrow.js +2 -2
- package/esm/models/publiccompetitorrow.js.map +1 -1
- package/esm/models/publiccriterionverdict.d.ts +26 -0
- package/esm/models/publiccriterionverdict.d.ts.map +1 -0
- package/esm/models/publiccriterionverdict.js +24 -0
- package/esm/models/publiccriterionverdict.js.map +1 -0
- package/esm/models/publiccurrentvisibilityresponse.js +2 -2
- package/esm/models/publiccurrentvisibilityresponse.js.map +1 -1
- package/esm/models/publicevalrunenvironment.d.ts +15 -0
- package/esm/models/publicevalrunenvironment.d.ts.map +1 -0
- package/esm/models/publicevalrunenvironment.js +15 -0
- package/esm/models/publicevalrunenvironment.js.map +1 -0
- package/esm/models/publicevalrunprogress.d.ts +1 -0
- package/esm/models/publicevalrunprogress.d.ts.map +1 -1
- package/esm/models/publicevalrunprogress.js +3 -2
- package/esm/models/publicevalrunprogress.js.map +1 -1
- package/esm/models/publicgraderresult.d.ts +13 -0
- package/esm/models/publicgraderresult.d.ts.map +1 -0
- package/esm/models/publicgraderresult.js +18 -0
- package/esm/models/publicgraderresult.js.map +1 -0
- package/esm/models/publicpromptvisibilitymetrics.js +2 -2
- package/esm/models/publicpromptvisibilitymetrics.js.map +1 -1
- package/esm/models/publicvisibilitysnapshot.js +2 -2
- package/esm/models/publicvisibilitysnapshot.js.map +1 -1
- package/esm/models/searchgroundingmode.d.ts +20 -0
- package/esm/models/searchgroundingmode.d.ts.map +1 -0
- package/esm/models/searchgroundingmode.js +19 -0
- package/esm/models/searchgroundingmode.js.map +1 -0
- package/esm/models/updateevaltemplaterequest.d.ts +71 -0
- package/esm/models/updateevaltemplaterequest.d.ts.map +1 -0
- package/esm/models/updateevaltemplaterequest.js +71 -0
- package/esm/models/updateevaltemplaterequest.js.map +1 -0
- package/esm/models/updatev2evalconfigurationrequest.d.ts +2 -0
- package/esm/models/updatev2evalconfigurationrequest.d.ts.map +1 -1
- package/esm/models/updatev2evalconfigurationrequest.js +3 -0
- package/esm/models/updatev2evalconfigurationrequest.js.map +1 -1
- package/esm/models/updatev2evalrequest.d.ts +5 -5
- package/esm/models/updatev2evalrequest.d.ts.map +1 -1
- package/esm/models/updatev2evalrequest.js +6 -10
- package/esm/models/updatev2evalrequest.js.map +1 -1
- package/esm/models/v2eval.d.ts +6 -2
- package/esm/models/v2eval.d.ts.map +1 -1
- package/esm/models/v2eval.js +5 -1
- package/esm/models/v2eval.js.map +1 -1
- package/esm/models/v2evalconfiguration.d.ts +2 -0
- package/esm/models/v2evalconfiguration.d.ts.map +1 -1
- package/esm/models/v2evalconfiguration.js +2 -0
- package/esm/models/v2evalconfiguration.js.map +1 -1
- package/esm/models/v2evalresultsummary.d.ts +10 -0
- package/esm/models/v2evalresultsummary.d.ts.map +1 -1
- package/esm/models/v2evalresultsummary.js +12 -1
- package/esm/models/v2evalresultsummary.js.map +1 -1
- package/esm/models/v2evalrun.d.ts.map +1 -1
- package/esm/models/v2evalrun.js +1 -1
- package/esm/models/v2evalrun.js.map +1 -1
- package/esm/models/v2evalrundetail.d.ts +30 -0
- package/esm/models/v2evalrundetail.d.ts.map +1 -0
- package/esm/models/v2evalrundetail.js +33 -0
- package/esm/models/v2evalrundetail.js.map +1 -0
- package/esm/models/v2evalrunscancelop.d.ts +31 -0
- package/esm/models/v2evalrunscancelop.d.ts.map +1 -0
- package/esm/models/v2evalrunscancelop.js +27 -0
- package/esm/models/v2evalrunscancelop.js.map +1 -0
- package/esm/models/v2evalrunscreateop.d.ts +23 -3
- package/esm/models/v2evalrunscreateop.d.ts.map +1 -1
- package/esm/models/v2evalrunscreateop.js +15 -2
- package/esm/models/v2evalrunscreateop.js.map +1 -1
- package/esm/models/v2evalrunsretrieveop.d.ts +9 -2
- package/esm/models/v2evalrunsretrieveop.d.ts.map +1 -1
- package/esm/models/v2evalrunsretrieveop.js +6 -2
- package/esm/models/v2evalrunsretrieveop.js.map +1 -1
- package/esm/models/v2evalruntraceevent.d.ts.map +1 -1
- package/esm/models/v2evalruntraceevent.js +1 -1
- package/esm/models/v2evalruntraceevent.js.map +1 -1
- package/esm/models/v2evalsbatchcreateop.d.ts +27 -0
- package/esm/models/v2evalsbatchcreateop.d.ts.map +1 -0
- package/esm/models/v2evalsbatchcreateop.js +24 -0
- package/esm/models/v2evalsbatchcreateop.js.map +1 -0
- package/esm/models/v2evaltemplatescreateop.d.ts +13 -0
- package/esm/models/v2evaltemplatescreateop.d.ts.map +1 -0
- package/esm/models/v2evaltemplatescreateop.js +16 -0
- package/esm/models/v2evaltemplatescreateop.js.map +1 -0
- package/esm/models/v2evaltemplatesdeleteop.d.ts +24 -0
- package/esm/models/v2evaltemplatesdeleteop.d.ts.map +1 -0
- package/esm/models/v2evaltemplatesdeleteop.js +23 -0
- package/esm/models/v2evaltemplatesdeleteop.js.map +1 -0
- package/esm/models/v2evaltemplatesretrieveop.d.ts +17 -0
- package/esm/models/v2evaltemplatesretrieveop.d.ts.map +1 -0
- package/esm/models/v2evaltemplatesretrieveop.js +19 -0
- package/esm/models/v2evaltemplatesretrieveop.js.map +1 -0
- package/esm/models/v2evaltemplatesupdateop.d.ts +33 -0
- package/esm/models/v2evaltemplatesupdateop.d.ts.map +1 -0
- package/esm/models/v2evaltemplatesupdateop.js +29 -0
- package/esm/models/v2evaltemplatesupdateop.js.map +1 -0
- package/esm/models/v2inlinemcpdefinition.d.ts +4 -4
- package/esm/models/v2inlinemcpdefinition.d.ts.map +1 -1
- package/esm/models/v2inlinemcpdefinition.js +3 -3
- package/esm/models/v2inlinemcpdefinition.js.map +1 -1
- package/esm/models/v2promptrundetail.d.ts +6 -0
- package/esm/models/v2promptrundetail.d.ts.map +1 -1
- package/esm/models/v2promptrundetail.js +4 -1
- package/esm/models/v2promptrundetail.js.map +1 -1
- package/esm/models/v2promptrunsummary.d.ts.map +1 -1
- package/esm/models/v2promptrunsummary.js +1 -1
- package/esm/models/v2promptrunsummary.js.map +1 -1
- package/esm/models/v2publictask.d.ts +2 -2
- package/esm/models/v2publictask.d.ts.map +1 -1
- package/esm/models/v2publictask.js +4 -4
- package/esm/models/v2publictask.js.map +1 -1
- package/esm/tool-names.d.ts.map +1 -1
- package/esm/tool-names.js +31 -15
- package/esm/tool-names.js.map +1 -1
- package/package.json +1 -1
- package/src/funcs/evalTemplatesV2EvalTemplatesCreate.ts +145 -0
- package/src/funcs/{evalResultsV2EvalResultsList.ts → evalTemplatesV2EvalTemplatesDelete.ts} +23 -19
- package/src/funcs/evalTemplatesV2EvalTemplatesList.ts +124 -0
- package/src/funcs/{evalResultsV2EvalResultsRetrieve.ts → evalTemplatesV2EvalTemplatesRetrieve.ts} +13 -13
- package/src/funcs/evalTemplatesV2EvalTemplatesUpdate.ts +154 -0
- package/src/funcs/{evalsAnalysis.ts → evalsCancel.ts} +13 -13
- package/src/funcs/{evalRunsV2EvalRunsRetrieve.ts → evalsGet.ts} +4 -4
- package/src/funcs/evalsRun.ts +6 -4
- package/src/funcs/evalsV2EvalsBatchCreate.ts +145 -0
- package/src/funcs/evalsV2EvalsCreate.ts +1 -1
- package/src/funcs/evalsV2EvalsUpdate.ts +1 -1
- package/src/funcs/promptsV2PromptRunsList.ts +1 -1
- package/src/funcs/promptsV2PromptRunsRetrieve.ts +1 -1
- package/src/landing-page.ts +1 -1
- package/src/lib/config.ts +3 -3
- package/src/mcp-server/mcp-server.ts +1 -1
- package/src/mcp-server/server.ts +17 -9
- package/src/mcp-server/tools/evalTemplatesV2EvalTemplatesCreate.ts +45 -0
- package/src/mcp-server/tools/evalTemplatesV2EvalTemplatesDelete.ts +45 -0
- package/src/mcp-server/tools/evalTemplatesV2EvalTemplatesList.ts +36 -0
- package/src/mcp-server/tools/{evalResultsV2EvalResultsRetrieve.ts → evalTemplatesV2EvalTemplatesRetrieve.ts} +9 -9
- package/src/mcp-server/tools/evalTemplatesV2EvalTemplatesUpdate.ts +45 -0
- package/src/mcp-server/tools/evalsCancel.ts +43 -0
- package/src/mcp-server/tools/{evalRunsV2EvalRunsRetrieve.ts → evalsGet.ts} +7 -7
- package/src/mcp-server/tools/evalsRun.ts +6 -4
- package/src/mcp-server/tools/evalsV2EvalsBatchCreate.ts +43 -0
- package/src/mcp-server/tools/evalsV2EvalsCreate.ts +1 -1
- package/src/mcp-server/tools/evalsV2EvalsUpdate.ts +1 -1
- package/src/mcp-server/tools/promptsV2PromptRunsList.ts +1 -1
- package/src/mcp-server/tools/promptsV2PromptRunsRetrieve.ts +1 -1
- package/src/models/createevaltemplatebatchrequest.ts +35 -0
- package/src/models/createevaltemplaterequest.ts +151 -0
- package/src/models/createv2evalconfigurationrequest.ts +8 -0
- package/src/models/createv2evalrequest.ts +18 -22
- package/src/models/createv2evalrunrequest.ts +19 -24
- package/src/models/evaltemplate.ts +154 -0
- package/src/models/evaltemplatebatchcreatedeval.ts +26 -0
- package/src/models/evaltemplatebatchitem.ts +54 -0
- package/src/models/evaltemplatebatchresponse.ts +30 -0
- package/src/models/evaltemplatebatchresult.ts +26 -0
- package/src/models/evaltemplateruntime.ts +64 -0
- package/src/models/evaltemplateschedule.ts +44 -0
- package/src/models/evaltemplatesresponse.ts +18 -0
- package/src/models/graderdefinition.ts +34 -0
- package/src/models/publiccompetitorrow.ts +2 -2
- package/src/models/publiccriterionverdict.ts +48 -0
- package/src/models/publiccurrentvisibilityresponse.ts +2 -2
- package/src/models/publicevalrunenvironment.ts +47 -0
- package/src/models/publicevalrunprogress.ts +8 -2
- package/src/models/publicgraderresult.ts +44 -0
- package/src/models/publicpromptvisibilitymetrics.ts +2 -2
- package/src/models/publicvisibilitysnapshot.ts +2 -2
- package/src/models/searchgroundingmode.ts +26 -0
- package/src/models/updateevaltemplaterequest.ts +116 -0
- package/src/models/updatev2evalconfigurationrequest.ts +9 -0
- package/src/models/updatev2evalrequest.ts +20 -21
- package/src/models/v2eval.ts +14 -2
- package/src/models/v2evalconfiguration.ts +8 -0
- package/src/models/v2evalresultsummary.ts +36 -1
- package/src/models/v2evalrun.ts +3 -1
- package/src/models/v2evalrundetail.ts +90 -0
- package/src/models/v2evalrunscancelop.ts +70 -0
- package/src/models/v2evalrunscreateop.ts +43 -5
- package/src/models/v2evalrunsretrieveop.ts +23 -3
- package/src/models/v2evalruntraceevent.ts +3 -1
- package/src/models/v2evalsbatchcreateop.ts +66 -0
- package/src/models/v2evaltemplatescreateop.ts +37 -0
- package/src/models/v2evaltemplatesdeleteop.ts +56 -0
- package/src/models/v2evaltemplatesretrieveop.ts +43 -0
- package/src/models/v2evaltemplatesupdateop.ts +79 -0
- package/src/models/v2inlinemcpdefinition.ts +7 -5
- package/src/models/v2promptrundetail.ts +12 -1
- package/src/models/v2promptrunsummary.ts +3 -1
- package/src/models/v2publictask.ts +9 -7
- package/src/tool-names.ts +31 -15
- package/test/simulations.test.js +30 -3
- package/esm/funcs/evalResultsV2EvalResultsList.d.ts +0 -16
- package/esm/funcs/evalResultsV2EvalResultsList.d.ts.map +0 -1
- package/esm/funcs/evalResultsV2EvalResultsList.js.map +0 -1
- package/esm/funcs/evalResultsV2EvalResultsRetrieve.d.ts +0 -16
- package/esm/funcs/evalResultsV2EvalResultsRetrieve.d.ts.map +0 -1
- package/esm/funcs/evalResultsV2EvalResultsRetrieve.js.map +0 -1
- package/esm/funcs/evalRunsV2EvalRunsRetrieve.d.ts +0 -16
- package/esm/funcs/evalRunsV2EvalRunsRetrieve.d.ts.map +0 -1
- package/esm/funcs/evalRunsV2EvalRunsRetrieve.js.map +0 -1
- package/esm/funcs/evalsAnalysis.d.ts +0 -16
- package/esm/funcs/evalsAnalysis.d.ts.map +0 -1
- package/esm/funcs/evalsAnalysis.js.map +0 -1
- package/esm/mcp-server/tools/evalResultsV2EvalResultsList.d.ts +0 -7
- package/esm/mcp-server/tools/evalResultsV2EvalResultsList.d.ts.map +0 -1
- package/esm/mcp-server/tools/evalResultsV2EvalResultsList.js +0 -35
- package/esm/mcp-server/tools/evalResultsV2EvalResultsList.js.map +0 -1
- package/esm/mcp-server/tools/evalResultsV2EvalResultsRetrieve.d.ts +0 -7
- package/esm/mcp-server/tools/evalResultsV2EvalResultsRetrieve.d.ts.map +0 -1
- package/esm/mcp-server/tools/evalResultsV2EvalResultsRetrieve.js +0 -35
- package/esm/mcp-server/tools/evalResultsV2EvalResultsRetrieve.js.map +0 -1
- package/esm/mcp-server/tools/evalRunsV2EvalRunsRetrieve.d.ts.map +0 -1
- package/esm/mcp-server/tools/evalRunsV2EvalRunsRetrieve.js.map +0 -1
- package/esm/mcp-server/tools/evalsAnalysis.d.ts +0 -7
- package/esm/mcp-server/tools/evalsAnalysis.d.ts.map +0 -1
- package/esm/mcp-server/tools/evalsAnalysis.js +0 -35
- package/esm/mcp-server/tools/evalsAnalysis.js.map +0 -1
- package/esm/models/v2evalresult.d.ts +0 -23
- package/esm/models/v2evalresult.d.ts.map +0 -1
- package/esm/models/v2evalresult.js +0 -24
- package/esm/models/v2evalresult.js.map +0 -1
- package/esm/models/v2evalresultslistop.d.ts +0 -11
- package/esm/models/v2evalresultslistop.d.ts.map +0 -1
- package/esm/models/v2evalresultslistop.js +0 -16
- package/esm/models/v2evalresultslistop.js.map +0 -1
- package/esm/models/v2evalresultsresponse.d.ts +0 -9
- package/esm/models/v2evalresultsresponse.d.ts.map +0 -1
- package/esm/models/v2evalresultsresponse.js +0 -12
- package/esm/models/v2evalresultsresponse.js.map +0 -1
- package/esm/models/v2evalresultsretrieveop.d.ts +0 -10
- package/esm/models/v2evalresultsretrieveop.d.ts.map +0 -1
- package/esm/models/v2evalresultsretrieveop.js +0 -15
- package/esm/models/v2evalresultsretrieveop.js.map +0 -1
- package/esm/models/v2evalrunanalysisresponse.d.ts +0 -12
- package/esm/models/v2evalrunanalysisresponse.d.ts.map +0 -1
- package/esm/models/v2evalrunanalysisresponse.js +0 -15
- package/esm/models/v2evalrunanalysisresponse.js.map +0 -1
- package/esm/models/v2evalrunsanalysisop.d.ts +0 -17
- package/esm/models/v2evalrunsanalysisop.d.ts.map +0 -1
- package/esm/models/v2evalrunsanalysisop.js +0 -19
- package/esm/models/v2evalrunsanalysisop.js.map +0 -1
- package/src/mcp-server/tools/evalResultsV2EvalResultsList.ts +0 -43
- package/src/mcp-server/tools/evalsAnalysis.ts +0 -43
- package/src/models/v2evalresult.ts +0 -46
- package/src/models/v2evalresultslistop.ts +0 -39
- package/src/models/v2evalresultsresponse.ts +0 -23
- package/src/models/v2evalresultsretrieveop.ts +0 -28
- package/src/models/v2evalrunanalysisresponse.ts +0 -32
- package/src/models/v2evalrunsanalysisop.ts +0 -48
|
@@ -20,7 +20,7 @@ export type V2PublicTask = {
|
|
|
20
20
|
execution_prompt: string;
|
|
21
21
|
reasoning?: string | null | undefined;
|
|
22
22
|
evidence?: { [k: string]: any } | undefined;
|
|
23
|
-
score
|
|
23
|
+
score?: number | null | undefined;
|
|
24
24
|
effort: string;
|
|
25
25
|
priority?: string | null | undefined;
|
|
26
26
|
status: string;
|
|
@@ -28,7 +28,7 @@ export type V2PublicTask = {
|
|
|
28
28
|
function_area?: string | null | undefined;
|
|
29
29
|
action_type?: string | null | undefined;
|
|
30
30
|
source?: string | null | undefined;
|
|
31
|
-
source_type
|
|
31
|
+
source_type?: string | null | undefined;
|
|
32
32
|
source_refs?: { [k: string]: any } | undefined;
|
|
33
33
|
metadata?: { [k: string]: any } | undefined;
|
|
34
34
|
created_at?: string | null | undefined;
|
|
@@ -63,7 +63,9 @@ export const V2PublicTask$zodSchema: z.ZodType<V2PublicTask> =
|
|
|
63
63
|
"Sort order within the batch, ascending; values need not start at 1.",
|
|
64
64
|
),
|
|
65
65
|
reasoning: z.string().nullable().optional(),
|
|
66
|
-
score: z.number()
|
|
66
|
+
score: z.number().nullable().optional().describe(
|
|
67
|
+
"Planner impact score; manually created tasks may be unscored.",
|
|
68
|
+
),
|
|
67
69
|
source: z.string().nullable().optional(),
|
|
68
70
|
source_key: z.string().describe(
|
|
69
71
|
"A finding key namespaced by source, such as `api_perf:https://example.com/docs/cli`.",
|
|
@@ -71,8 +73,8 @@ export const V2PublicTask$zodSchema: z.ZodType<V2PublicTask> =
|
|
|
71
73
|
source_refs: z.record(z.string(), z.any()).optional().describe(
|
|
72
74
|
"Source references needed to understand or execute the task.",
|
|
73
75
|
),
|
|
74
|
-
source_type: z.string().describe(
|
|
75
|
-
"A free-form source type such as `api_performance`;
|
|
76
|
+
source_type: z.string().nullable().optional().describe(
|
|
77
|
+
"A free-form source type such as `api_performance`; manually created tasks may not have a source.",
|
|
76
78
|
),
|
|
77
79
|
status: z.string().describe(
|
|
78
80
|
"The workflow status. New tasks start as `new`; `added_to_sheet` and `snoozed` tasks are listed only when `status=all`.",
|
|
@@ -106,11 +108,11 @@ export const V2PublicTask$zodSchemaOutbound: z.ZodType<any> = z.object({
|
|
|
106
108
|
priority: z.nullable(z.string()).optional(),
|
|
107
109
|
rank: z.int(),
|
|
108
110
|
reasoning: z.nullable(z.string()).optional(),
|
|
109
|
-
score: z.number(),
|
|
111
|
+
score: z.nullable(z.number()).optional(),
|
|
110
112
|
source: z.nullable(z.string()).optional(),
|
|
111
113
|
source_key: z.string(),
|
|
112
114
|
source_refs: z.record(z.string(), z.any()).optional(),
|
|
113
|
-
source_type: z.string(),
|
|
115
|
+
source_type: z.nullable(z.string()).optional(),
|
|
114
116
|
status: z.string(),
|
|
115
117
|
task_type: z.string(),
|
|
116
118
|
title: z.string(),
|
package/src/tool-names.ts
CHANGED
|
@@ -50,11 +50,11 @@ export const toolNames: Array<{ name: string; description: string }>= [
|
|
|
50
50
|
},
|
|
51
51
|
{
|
|
52
52
|
"name": "prompts-v2-prompt-runs-list",
|
|
53
|
-
"description": "List Prompt Runs\n\nReturns stored prompt runs without exposing raw runtime or provider configuration.
|
|
53
|
+
"description": "List Prompt Runs\n\nReturns stored prompt runs with limit-and-offset pagination, without exposing raw runtime or provider configuration. Filter by prompt, platform, or date range."
|
|
54
54
|
},
|
|
55
55
|
{
|
|
56
56
|
"name": "prompts-v2-prompt-runs-retrieve",
|
|
57
|
-
"description": "Retrieve Prompt Run\n\nReturns one stored prompt run, including
|
|
57
|
+
"description": "Retrieve Prompt Run\n\nReturns one stored prompt run, including mentions, citations, and any sanitized coding-agent trace captured for the run."
|
|
58
58
|
},
|
|
59
59
|
{
|
|
60
60
|
"name": "prompts-v2-prompts-retrieve",
|
|
@@ -114,7 +114,11 @@ export const toolNames: Array<{ name: string; description: string }>= [
|
|
|
114
114
|
},
|
|
115
115
|
{
|
|
116
116
|
"name": "evals-v2-evals-create",
|
|
117
|
-
"description": "Create Eval\n\nCreates 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."
|
|
117
|
+
"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."
|
|
118
|
+
},
|
|
119
|
+
{
|
|
120
|
+
"name": "evals-v2-evals-batch-create",
|
|
121
|
+
"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."
|
|
118
122
|
},
|
|
119
123
|
{
|
|
120
124
|
"name": "evals-v2-evals-retrieve",
|
|
@@ -122,7 +126,7 @@ export const toolNames: Array<{ name: string; description: string }>= [
|
|
|
122
126
|
},
|
|
123
127
|
{
|
|
124
128
|
"name": "evals-v2-evals-update",
|
|
125
|
-
"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."
|
|
129
|
+
"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."
|
|
126
130
|
},
|
|
127
131
|
{
|
|
128
132
|
"name": "evals-v2-evals-delete",
|
|
@@ -150,31 +154,43 @@ export const toolNames: Array<{ name: string; description: string }>= [
|
|
|
150
154
|
},
|
|
151
155
|
{
|
|
152
156
|
"name": "evals-run",
|
|
153
|
-
"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
|
|
157
|
+
"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."
|
|
154
158
|
},
|
|
155
159
|
{
|
|
156
|
-
"name": "evals-
|
|
157
|
-
"description": "
|
|
160
|
+
"name": "evals-cancel",
|
|
161
|
+
"description": "Cancel an eval run\n\nMarks 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."
|
|
158
162
|
},
|
|
159
163
|
{
|
|
160
164
|
"name": "evals-trace",
|
|
161
165
|
"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."
|
|
162
166
|
},
|
|
163
167
|
{
|
|
164
|
-
"name": "
|
|
165
|
-
"description": "
|
|
168
|
+
"name": "evals-get",
|
|
169
|
+
"description": "Retrieve an eval run\n\nReturns 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."
|
|
170
|
+
},
|
|
171
|
+
{
|
|
172
|
+
"name": "eval-templates-v2-eval-templates-list",
|
|
173
|
+
"description": "List Eval templates\n\nReturns active reusable Eval templates owned by the authenticated organization. Templates contain reusable configuration and never task prompts."
|
|
166
174
|
},
|
|
167
175
|
{
|
|
168
|
-
"name": "eval-
|
|
169
|
-
"description": "
|
|
176
|
+
"name": "eval-templates-v2-eval-templates-create",
|
|
177
|
+
"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."
|
|
170
178
|
},
|
|
171
179
|
{
|
|
172
|
-
"name": "eval-
|
|
173
|
-
"description": "
|
|
180
|
+
"name": "eval-templates-v2-eval-templates-retrieve",
|
|
181
|
+
"description": "Retrieve an Eval template\n\nReturns one active reusable Eval template."
|
|
174
182
|
},
|
|
175
183
|
{
|
|
176
|
-
"name": "eval-
|
|
177
|
-
"description": "
|
|
184
|
+
"name": "eval-templates-v2-eval-templates-update",
|
|
185
|
+
"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."
|
|
186
|
+
},
|
|
187
|
+
{
|
|
188
|
+
"name": "eval-templates-v2-eval-templates-delete",
|
|
189
|
+
"description": "Delete an Eval template\n\nSoft-deletes an Eval template. Evals previously created from it remain independent and continue to run with their copied configuration."
|
|
190
|
+
},
|
|
191
|
+
{
|
|
192
|
+
"name": "eval-runs-v2-eval-runs-list",
|
|
193
|
+
"description": "List Runs\n\nReturns jobs created through `POST /v2/eval-runs`, ordered newest first, with lifecycle state and progress."
|
|
178
194
|
},
|
|
179
195
|
{
|
|
180
196
|
"name": "agents-v2-agents-list",
|
package/test/simulations.test.js
CHANGED
|
@@ -9,6 +9,9 @@ const { CreateV2EvalRequest$zodSchema } = await import(
|
|
|
9
9
|
const { CreateV2EvalRunRequest$zodSchema } = await import(
|
|
10
10
|
"../esm/models/createv2evalrunrequest.js"
|
|
11
11
|
);
|
|
12
|
+
const { UpdateV2EvalRequest$zodSchema } = await import(
|
|
13
|
+
"../esm/models/updatev2evalrequest.js"
|
|
14
|
+
);
|
|
12
15
|
const { tool$evalsV2EvalsCreate } = await import(
|
|
13
16
|
"../esm/mcp-server/tools/evalsV2EvalsCreate.js"
|
|
14
17
|
);
|
|
@@ -16,8 +19,8 @@ const { tool$evalsRun } = await import(
|
|
|
16
19
|
"../esm/mcp-server/tools/evalsRun.js"
|
|
17
20
|
);
|
|
18
21
|
|
|
19
|
-
const savedCli =
|
|
20
|
-
const inlineCli =
|
|
22
|
+
const savedCli = "230f0c71-ed91-4a4a-9932-9cef267411bc";
|
|
23
|
+
const inlineCli = "brew install lightsagehq/tools/lightsage";
|
|
21
24
|
const savedMcp = { id: "550e8400-e29b-41d4-a716-446655440000" };
|
|
22
25
|
const inlineMcp = {
|
|
23
26
|
name: "lightsage",
|
|
@@ -52,12 +55,16 @@ test("create eval accepts the canonical simulation matrix", () => {
|
|
|
52
55
|
agent: ["codex:gpt-5-4", "claude-code:claude-opus-4-8"],
|
|
53
56
|
clis: [savedCli, inlineCli],
|
|
54
57
|
mcps: [savedMcp, inlineMcp],
|
|
58
|
+
setup_commands: ["npm install", "npm run db:migrate"],
|
|
59
|
+
cleanup_commands: ["npm run db:cleanup"],
|
|
55
60
|
runs: 2,
|
|
56
61
|
tags: ["smoke-test"],
|
|
57
62
|
});
|
|
58
63
|
|
|
59
64
|
assert.deepEqual(parsed.clis, [savedCli, inlineCli]);
|
|
60
65
|
assert.deepEqual(parsed.mcps, [savedMcp, inlineMcp]);
|
|
66
|
+
assert.deepEqual(parsed.setup_commands, ["npm install", "npm run db:migrate"]);
|
|
67
|
+
assert.deepEqual(parsed.cleanup_commands, ["npm run db:cleanup"]);
|
|
61
68
|
assert.equal(parsed.runs, 2);
|
|
62
69
|
});
|
|
63
70
|
|
|
@@ -69,6 +76,8 @@ test("direct eval run accepts the same canonical simulation fields", () => {
|
|
|
69
76
|
persona: "Platform engineer",
|
|
70
77
|
clis: [savedCli, inlineCli],
|
|
71
78
|
mcps: [savedMcp, inlineMcp],
|
|
79
|
+
setup_commands: ["npm install"],
|
|
80
|
+
cleanup_commands: ["npm run db:cleanup"],
|
|
72
81
|
env: { LIGHTSAGE_API_KEY: "runtime-only" },
|
|
73
82
|
skills: ["resend/resend-skills"],
|
|
74
83
|
runs: 3,
|
|
@@ -80,11 +89,13 @@ test("direct eval run accepts the same canonical simulation fields", () => {
|
|
|
80
89
|
]);
|
|
81
90
|
assert.deepEqual(parsed.clis, [savedCli, inlineCli]);
|
|
82
91
|
assert.deepEqual(parsed.mcps, [savedMcp, inlineMcp]);
|
|
92
|
+
assert.deepEqual(parsed.setup_commands, ["npm install"]);
|
|
93
|
+
assert.deepEqual(parsed.cleanup_commands, ["npm run db:cleanup"]);
|
|
83
94
|
assert.deepEqual(parsed.skills, ["resend/resend-skills"]);
|
|
84
95
|
assert.equal(parsed.runs, 3);
|
|
85
96
|
});
|
|
86
97
|
|
|
87
|
-
test("simulation tools reject non-object
|
|
98
|
+
test("simulation tools reject non-object MCP entries", () => {
|
|
88
99
|
const result = CreateV2EvalRunRequest$zodSchema.safeParse({
|
|
89
100
|
prompts: ["Use the CLI."],
|
|
90
101
|
judges: ["The workflow succeeds."],
|
|
@@ -96,6 +107,18 @@ test("simulation tools reject non-object CLI and MCP entries", () => {
|
|
|
96
107
|
assert.equal(result.success, false);
|
|
97
108
|
});
|
|
98
109
|
|
|
110
|
+
test("eval updates accept string CLIs and lifecycle commands", () => {
|
|
111
|
+
const parsed = UpdateV2EvalRequest$zodSchema.parse({
|
|
112
|
+
clis: [savedCli, inlineCli],
|
|
113
|
+
setup_commands: ["npm install"],
|
|
114
|
+
cleanup_commands: [],
|
|
115
|
+
});
|
|
116
|
+
|
|
117
|
+
assert.deepEqual(parsed.clis, [savedCli, inlineCli]);
|
|
118
|
+
assert.deepEqual(parsed.setup_commands, ["npm install"]);
|
|
119
|
+
assert.deepEqual(parsed.cleanup_commands, []);
|
|
120
|
+
});
|
|
121
|
+
|
|
99
122
|
test("MCP tools forward canonical eval resources unchanged", async () => {
|
|
100
123
|
const requests = [];
|
|
101
124
|
const server = createSimulationServer(requests);
|
|
@@ -113,6 +136,8 @@ test("MCP tools forward canonical eval resources unchanged", async () => {
|
|
|
113
136
|
agent: ["codex:gpt-5-4"],
|
|
114
137
|
clis: [savedCli, inlineCli],
|
|
115
138
|
mcps: [savedMcp, inlineMcp],
|
|
139
|
+
setup_commands: ["npm install"],
|
|
140
|
+
cleanup_commands: ["npm run db:cleanup"],
|
|
116
141
|
runs: 2,
|
|
117
142
|
tags: ["smoke-test"],
|
|
118
143
|
};
|
|
@@ -122,6 +147,8 @@ test("MCP tools forward canonical eval resources unchanged", async () => {
|
|
|
122
147
|
agent: ["codex:gpt-5-4"],
|
|
123
148
|
clis: [savedCli, inlineCli],
|
|
124
149
|
mcps: [savedMcp, inlineMcp],
|
|
150
|
+
setup_commands: ["npm install"],
|
|
151
|
+
cleanup_commands: ["npm run db:cleanup"],
|
|
125
152
|
env: { LIGHTSAGE_API_KEY: "runtime-only" },
|
|
126
153
|
skills: ["resend/resend-skills"],
|
|
127
154
|
runs: 3,
|
|
@@ -1,16 +0,0 @@
|
|
|
1
|
-
import { LightsageMcpCore } from "../core.js";
|
|
2
|
-
import { RequestOptions } from "../lib/sdks.js";
|
|
3
|
-
import { APIError } from "../models/errors/apierror.js";
|
|
4
|
-
import { ConnectionError, InvalidRequestError, RequestAbortedError, RequestTimeoutError, UnexpectedClientError } from "../models/errors/httpclienterrors.js";
|
|
5
|
-
import { SDKValidationError } from "../models/errors/sdkvalidationerror.js";
|
|
6
|
-
import { V2EvalResultsListRequest } from "../models/v2evalresultslistop.js";
|
|
7
|
-
import { APIPromise } from "../types/async.js";
|
|
8
|
-
import { Result } from "../types/fp.js";
|
|
9
|
-
/**
|
|
10
|
-
* List Results
|
|
11
|
-
*
|
|
12
|
-
* @remarks
|
|
13
|
-
* Returns completed custom-eval results from all supported run flows, ordered newest first. Result IDs are different from eval run IDs.
|
|
14
|
-
*/
|
|
15
|
-
export declare function evalResultsV2EvalResultsList(client$: LightsageMcpCore, request?: V2EvalResultsListRequest | undefined, options?: RequestOptions): APIPromise<Result<Response, APIError | SDKValidationError | UnexpectedClientError | InvalidRequestError | RequestAbortedError | RequestTimeoutError | ConnectionError>>;
|
|
16
|
-
//# sourceMappingURL=evalResultsV2EvalResultsList.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"evalResultsV2EvalResultsList.d.ts","sourceRoot":"","sources":["../../src/funcs/evalResultsV2EvalResultsList.ts"],"names":[],"mappings":"AAKA,OAAO,EAAE,gBAAgB,EAAE,MAAM,YAAY,CAAC;AAI9C,OAAO,EAAE,cAAc,EAAE,MAAM,gBAAgB,CAAC;AAGhD,OAAO,EAAE,QAAQ,EAAE,MAAM,8BAA8B,CAAC;AACxD,OAAO,EACL,eAAe,EACf,mBAAmB,EACnB,mBAAmB,EACnB,mBAAmB,EACnB,qBAAqB,EACtB,MAAM,sCAAsC,CAAC;AAC9C,OAAO,EAAE,kBAAkB,EAAE,MAAM,wCAAwC,CAAC;AAC5E,OAAO,EACL,wBAAwB,EAEzB,MAAM,kCAAkC,CAAC;AAC1C,OAAO,EAAW,UAAU,EAAE,MAAM,mBAAmB,CAAC;AACxD,OAAO,EAAE,MAAM,EAAE,MAAM,gBAAgB,CAAC;AAExC;;;;;GAKG;AACH,wBAAgB,4BAA4B,CAC1C,OAAO,EAAE,gBAAgB,EACzB,OAAO,CAAC,EAAE,wBAAwB,GAAG,SAAS,EAC9C,OAAO,CAAC,EAAE,cAAc,GACvB,UAAU,CACX,MAAM,CACJ,QAAQ,EACN,QAAQ,GACR,kBAAkB,GAClB,qBAAqB,GACrB,mBAAmB,GACnB,mBAAmB,GACnB,mBAAmB,GACnB,eAAe,CAClB,CACF,CAMA"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"evalResultsV2EvalResultsList.js","sourceRoot":"","sources":["../../src/funcs/evalResultsV2EvalResultsList.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAGH,OAAO,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAC;AACtD,OAAO,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAC;AAClD,OAAO,EAAE,SAAS,EAAE,MAAM,mBAAmB,CAAC;AAE9C,OAAO,EAAE,eAAe,EAAE,qBAAqB,EAAE,MAAM,oBAAoB,CAAC;AAC5E,OAAO,EAAE,UAAU,EAAE,MAAM,eAAe,CAAC;AAU3C,OAAO,EAEL,kCAAkC,GACnC,MAAM,kCAAkC,CAAC;AAC1C,OAAO,EAAW,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAGxD;;;;;GAKG;AACH,MAAM,UAAU,4BAA4B,CAC1C,OAAyB,EACzB,OAA8C,EAC9C,OAAwB;IAaxB,OAAO,IAAI,UAAU,CAAC,GAAG,CACvB,OAAO,EACP,OAAO,EACP,OAAO,CACR,CAAC,CAAC;AACL,CAAC;AAED,KAAK,UAAU,GAAG,CAChB,OAAyB,EACzB,OAA8C,EAC9C,OAAwB;IAgBxB,MAAM,OAAO,GAAG,SAAS,CACvB,OAAO,EACP,CAAC,MAAM,EAAE,EAAE,CAAC,kCAAkC,CAAC,QAAQ,EAAE,CAAC,KAAK,CAAC,MAAM,CAAC,EACvE,yBAAyB,CAC1B,CAAC;IACF,IAAI,CAAC,OAAO,CAAC,EAAE,EAAE,CAAC;QAChB,OAAO,CAAC,OAAO,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IAC1C,CAAC;IACD,MAAM,QAAQ,GAAG,OAAO,CAAC,KAAK,CAAC;IAC/B,MAAM,KAAK,GAAG,IAAI,CAAC;IACnB,MAAM,KAAK,GAAG,UAAU,CAAC,kBAAkB,CAAC,EAAE,CAAC;IAC/C,MAAM,MAAM,GAAG,eAAe,CAAC;QAC7B,OAAO,EAAE,QAAQ,EAAE,KAAK;QACxB,OAAO,EAAE,QAAQ,EAAE,KAAK;KACzB,CAAC,CAAC;IAEH,MAAM,QAAQ,GAAG,IAAI,OAAO,CAAC,UAAU,CAAC;QACtC,MAAM,EAAE,kBAAkB;KAC3B,CAAC,CAAC,CAAC;IACJ,MAAM,aAAa,GAAG,MAAM,eAAe,CAAC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;IACvE,MAAM,eAAe,GAAG,qBAAqB,CAAC,aAAa,CAAC,CAAC;IAE7D,MAAM,OAAO,GAAG;QACd,OAAO,EAAE,OAAO,CAAC,QAAQ;QACzB,OAAO,EAAE,OAAO,EAAE,SAAS,IAAI,OAAO,CAAC,QAAQ,IAAI,EAAE;QACrD,WAAW,EAAE,sBAAsB;QACnC,YAAY,EAAE,IAAI;QAClB,gBAAgB,EAAE,eAAe;QACjC,cAAc,EAAE,OAAO,CAAC,QAAQ,CAAC,QAAQ;QACzC,WAAW,EAAE,OAAO,EAAE,OAAO;eACxB,OAAO,CAAC,QAAQ,CAAC,WAAW;eAC5B,EAAE,QAAQ,EAAE,MAAM,EAAE;QACzB,UAAU,EAAE,OAAO,EAAE,UAAU,IAAI;YACjC,KAAK;YACL,KAAK;YACL,KAAK;YACL,KAAK;YACL,KAAK;SACN;KACF,CAAC;IAEF,MAAM,UAAU,GAAG,OAAO,CAAC,cAAc,CAAC,OAAO,EAAE;QACjD,QAAQ,EAAE,eAAe;QACzB,MAAM,EAAE,KAAK;QACb,OAAO,EAAE,OAAO,EAAE,SAAS;QAC3B,IAAI,EAAE,KAAK;QACX,OAAO,EAAE,QAAQ;QACjB,KAAK,EAAE,MAAM;QACb,IAAI,EAAE,KAAK;QACX,SAAS,EAAE,OAAO,CAAC,QAAQ,CAAC,SAAS;QACrC,SAAS,EAAE,OAAO,EAAE,SAAS,IAAI,OAAO,CAAC,QAAQ,CAAC,SAAS;eACtD,CAAC,CAAC;KACR,EAAE,OAAO,CAAC,CAAC;IACZ,IAAI,CAAC,UAAU,CAAC,EAAE,EAAE,CAAC;QACnB,OAAO,CAAC,UAAU,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IAC7C,CAAC;IACD,MAAM,IAAI,GAAG,UAAU,CAAC,KAAK,CAAC;IAE9B,MAAM,QAAQ,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC,IAAI,EAAE;QACvC,OAAO;QACP,UAAU,EAAE,EAAE;QACd,WAAW,EAAE,OAAO,CAAC,WAAW;QAChC,UAAU,EAAE,OAAO,CAAC,UAAU;KAC/B,CAAC,CAAC;IACH,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;QACjB,OAAO,CAAC,QAAQ,EAAE,EAAE,MAAM,EAAE,eAAe,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;IAChE,CAAC;IACD,OAAO,CAAC,QAAQ,EAAE;YAChB,MAAM,EAAE,UAAU;YAClB,SAAS,EAAE,IAAI;YACf,QAAQ,EAAE,QAAQ,CAAC,KAAK;SACzB,CAAC,CAAC;AACL,CAAC"}
|
|
@@ -1,16 +0,0 @@
|
|
|
1
|
-
import { LightsageMcpCore } from "../core.js";
|
|
2
|
-
import { RequestOptions } from "../lib/sdks.js";
|
|
3
|
-
import { APIError } from "../models/errors/apierror.js";
|
|
4
|
-
import { ConnectionError, InvalidRequestError, RequestAbortedError, RequestTimeoutError, UnexpectedClientError } from "../models/errors/httpclienterrors.js";
|
|
5
|
-
import { SDKValidationError } from "../models/errors/sdkvalidationerror.js";
|
|
6
|
-
import { V2EvalResultsRetrieveRequest } from "../models/v2evalresultsretrieveop.js";
|
|
7
|
-
import { APIPromise } from "../types/async.js";
|
|
8
|
-
import { Result } from "../types/fp.js";
|
|
9
|
-
/**
|
|
10
|
-
* Retrieve Result
|
|
11
|
-
*
|
|
12
|
-
* @remarks
|
|
13
|
-
* Returns one completed custom-eval result with its output and diagnostic details.
|
|
14
|
-
*/
|
|
15
|
-
export declare function evalResultsV2EvalResultsRetrieve(client$: LightsageMcpCore, request: V2EvalResultsRetrieveRequest, options?: RequestOptions): APIPromise<Result<Response, APIError | SDKValidationError | UnexpectedClientError | InvalidRequestError | RequestAbortedError | RequestTimeoutError | ConnectionError>>;
|
|
16
|
-
//# sourceMappingURL=evalResultsV2EvalResultsRetrieve.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"evalResultsV2EvalResultsRetrieve.d.ts","sourceRoot":"","sources":["../../src/funcs/evalResultsV2EvalResultsRetrieve.ts"],"names":[],"mappings":"AAKA,OAAO,EAAE,gBAAgB,EAAE,MAAM,YAAY,CAAC;AAI9C,OAAO,EAAE,cAAc,EAAE,MAAM,gBAAgB,CAAC;AAGhD,OAAO,EAAE,QAAQ,EAAE,MAAM,8BAA8B,CAAC;AACxD,OAAO,EACL,eAAe,EACf,mBAAmB,EACnB,mBAAmB,EACnB,mBAAmB,EACnB,qBAAqB,EACtB,MAAM,sCAAsC,CAAC;AAC9C,OAAO,EAAE,kBAAkB,EAAE,MAAM,wCAAwC,CAAC;AAC5E,OAAO,EACL,4BAA4B,EAE7B,MAAM,sCAAsC,CAAC;AAC9C,OAAO,EAAW,UAAU,EAAE,MAAM,mBAAmB,CAAC;AACxD,OAAO,EAAE,MAAM,EAAE,MAAM,gBAAgB,CAAC;AAExC;;;;;GAKG;AACH,wBAAgB,gCAAgC,CAC9C,OAAO,EAAE,gBAAgB,EACzB,OAAO,EAAE,4BAA4B,EACrC,OAAO,CAAC,EAAE,cAAc,GACvB,UAAU,CACX,MAAM,CACJ,QAAQ,EACN,QAAQ,GACR,kBAAkB,GAClB,qBAAqB,GACrB,mBAAmB,GACnB,mBAAmB,GACnB,mBAAmB,GACnB,eAAe,CAClB,CACF,CAMA"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"evalResultsV2EvalResultsRetrieve.js","sourceRoot":"","sources":["../../src/funcs/evalResultsV2EvalResultsRetrieve.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAGH,OAAO,EAAE,YAAY,EAAE,MAAM,qBAAqB,CAAC;AACnD,OAAO,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAC;AAClD,OAAO,EAAE,SAAS,EAAE,MAAM,mBAAmB,CAAC;AAE9C,OAAO,EAAE,eAAe,EAAE,qBAAqB,EAAE,MAAM,oBAAoB,CAAC;AAC5E,OAAO,EAAE,UAAU,EAAE,MAAM,eAAe,CAAC;AAU3C,OAAO,EAEL,sCAAsC,GACvC,MAAM,sCAAsC,CAAC;AAC9C,OAAO,EAAW,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAGxD;;;;;GAKG;AACH,MAAM,UAAU,gCAAgC,CAC9C,OAAyB,EACzB,OAAqC,EACrC,OAAwB;IAaxB,OAAO,IAAI,UAAU,CAAC,GAAG,CACvB,OAAO,EACP,OAAO,EACP,OAAO,CACR,CAAC,CAAC;AACL,CAAC;AAED,KAAK,UAAU,GAAG,CAChB,OAAyB,EACzB,OAAqC,EACrC,OAAwB;IAgBxB,MAAM,OAAO,GAAG,SAAS,CACvB,OAAO,EACP,CAAC,MAAM,EAAE,EAAE,CAAC,sCAAsC,CAAC,KAAK,CAAC,MAAM,CAAC,EAChE,yBAAyB,CAC1B,CAAC;IACF,IAAI,CAAC,OAAO,CAAC,EAAE,EAAE,CAAC;QAChB,OAAO,CAAC,OAAO,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IAC1C,CAAC;IACD,MAAM,QAAQ,GAAG,OAAO,CAAC,KAAK,CAAC;IAC/B,MAAM,KAAK,GAAG,IAAI,CAAC;IAEnB,MAAM,WAAW,GAAG;QAClB,SAAS,EAAE,YAAY,CAAC,WAAW,EAAE,QAAQ,CAAC,SAAS,EAAE;YACvD,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,SAAS;SACxB,CAAC;KACH,CAAC;IACF,MAAM,KAAK,GAAG,UAAU,CAAC,8BAA8B,CAAC,CACtD,WAAW,CACZ,CAAC;IAEF,MAAM,QAAQ,GAAG,IAAI,OAAO,CAAC,UAAU,CAAC;QACtC,MAAM,EAAE,kBAAkB;KAC3B,CAAC,CAAC,CAAC;IACJ,MAAM,aAAa,GAAG,MAAM,eAAe,CAAC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;IACvE,MAAM,eAAe,GAAG,qBAAqB,CAAC,aAAa,CAAC,CAAC;IAE7D,MAAM,OAAO,GAAG;QACd,OAAO,EAAE,OAAO,CAAC,QAAQ;QACzB,OAAO,EAAE,OAAO,EAAE,SAAS,IAAI,OAAO,CAAC,QAAQ,IAAI,EAAE;QACrD,WAAW,EAAE,0BAA0B;QACvC,YAAY,EAAE,IAAI;QAClB,gBAAgB,EAAE,eAAe;QACjC,cAAc,EAAE,OAAO,CAAC,QAAQ,CAAC,QAAQ;QACzC,WAAW,EAAE,OAAO,EAAE,OAAO;eACxB,OAAO,CAAC,QAAQ,CAAC,WAAW;eAC5B,EAAE,QAAQ,EAAE,MAAM,EAAE;QACzB,UAAU,EAAE,OAAO,EAAE,UAAU,IAAI;YACjC,KAAK;YACL,KAAK;YACL,KAAK;YACL,KAAK;YACL,KAAK;SACN;KACF,CAAC;IAEF,MAAM,UAAU,GAAG,OAAO,CAAC,cAAc,CAAC,OAAO,EAAE;QACjD,QAAQ,EAAE,eAAe;QACzB,MAAM,EAAE,KAAK;QACb,OAAO,EAAE,OAAO,EAAE,SAAS;QAC3B,IAAI,EAAE,KAAK;QACX,OAAO,EAAE,QAAQ;QACjB,IAAI,EAAE,KAAK;QACX,SAAS,EAAE,OAAO,CAAC,QAAQ,CAAC,SAAS;QACrC,SAAS,EAAE,OAAO,EAAE,SAAS,IAAI,OAAO,CAAC,QAAQ,CAAC,SAAS;eACtD,CAAC,CAAC;KACR,EAAE,OAAO,CAAC,CAAC;IACZ,IAAI,CAAC,UAAU,CAAC,EAAE,EAAE,CAAC;QACnB,OAAO,CAAC,UAAU,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IAC7C,CAAC;IACD,MAAM,IAAI,GAAG,UAAU,CAAC,KAAK,CAAC;IAE9B,MAAM,QAAQ,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC,IAAI,EAAE;QACvC,OAAO;QACP,UAAU,EAAE,EAAE;QACd,WAAW,EAAE,OAAO,CAAC,WAAW;QAChC,UAAU,EAAE,OAAO,CAAC,UAAU;KAC/B,CAAC,CAAC;IACH,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;QACjB,OAAO,CAAC,QAAQ,EAAE,EAAE,MAAM,EAAE,eAAe,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;IAChE,CAAC;IACD,OAAO,CAAC,QAAQ,EAAE;YAChB,MAAM,EAAE,UAAU;YAClB,SAAS,EAAE,IAAI;YACf,QAAQ,EAAE,QAAQ,CAAC,KAAK;SACzB,CAAC,CAAC;AACL,CAAC"}
|
|
@@ -1,16 +0,0 @@
|
|
|
1
|
-
import { LightsageMcpCore } from "../core.js";
|
|
2
|
-
import { RequestOptions } from "../lib/sdks.js";
|
|
3
|
-
import { APIError } from "../models/errors/apierror.js";
|
|
4
|
-
import { ConnectionError, InvalidRequestError, RequestAbortedError, RequestTimeoutError, UnexpectedClientError } from "../models/errors/httpclienterrors.js";
|
|
5
|
-
import { SDKValidationError } from "../models/errors/sdkvalidationerror.js";
|
|
6
|
-
import { V2EvalRunsRetrieveRequest } from "../models/v2evalrunsretrieveop.js";
|
|
7
|
-
import { APIPromise } from "../types/async.js";
|
|
8
|
-
import { Result } from "../types/fp.js";
|
|
9
|
-
/**
|
|
10
|
-
* Retrieve Run
|
|
11
|
-
*
|
|
12
|
-
* @remarks
|
|
13
|
-
* Returns the status and progress of an eval job. Poll until the status reaches a terminal value.
|
|
14
|
-
*/
|
|
15
|
-
export declare function evalRunsV2EvalRunsRetrieve(client$: LightsageMcpCore, request: V2EvalRunsRetrieveRequest, options?: RequestOptions): APIPromise<Result<Response, APIError | SDKValidationError | UnexpectedClientError | InvalidRequestError | RequestAbortedError | RequestTimeoutError | ConnectionError>>;
|
|
16
|
-
//# sourceMappingURL=evalRunsV2EvalRunsRetrieve.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"evalRunsV2EvalRunsRetrieve.d.ts","sourceRoot":"","sources":["../../src/funcs/evalRunsV2EvalRunsRetrieve.ts"],"names":[],"mappings":"AAKA,OAAO,EAAE,gBAAgB,EAAE,MAAM,YAAY,CAAC;AAI9C,OAAO,EAAE,cAAc,EAAE,MAAM,gBAAgB,CAAC;AAGhD,OAAO,EAAE,QAAQ,EAAE,MAAM,8BAA8B,CAAC;AACxD,OAAO,EACL,eAAe,EACf,mBAAmB,EACnB,mBAAmB,EACnB,mBAAmB,EACnB,qBAAqB,EACtB,MAAM,sCAAsC,CAAC;AAC9C,OAAO,EAAE,kBAAkB,EAAE,MAAM,wCAAwC,CAAC;AAC5E,OAAO,EACL,yBAAyB,EAE1B,MAAM,mCAAmC,CAAC;AAC3C,OAAO,EAAW,UAAU,EAAE,MAAM,mBAAmB,CAAC;AACxD,OAAO,EAAE,MAAM,EAAE,MAAM,gBAAgB,CAAC;AAExC;;;;;GAKG;AACH,wBAAgB,0BAA0B,CACxC,OAAO,EAAE,gBAAgB,EACzB,OAAO,EAAE,yBAAyB,EAClC,OAAO,CAAC,EAAE,cAAc,GACvB,UAAU,CACX,MAAM,CACJ,QAAQ,EACN,QAAQ,GACR,kBAAkB,GAClB,qBAAqB,GACrB,mBAAmB,GACnB,mBAAmB,GACnB,mBAAmB,GACnB,eAAe,CAClB,CACF,CAMA"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"evalRunsV2EvalRunsRetrieve.js","sourceRoot":"","sources":["../../src/funcs/evalRunsV2EvalRunsRetrieve.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAGH,OAAO,EAAE,YAAY,EAAE,MAAM,qBAAqB,CAAC;AACnD,OAAO,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAC;AAClD,OAAO,EAAE,SAAS,EAAE,MAAM,mBAAmB,CAAC;AAE9C,OAAO,EAAE,eAAe,EAAE,qBAAqB,EAAE,MAAM,oBAAoB,CAAC;AAC5E,OAAO,EAAE,UAAU,EAAE,MAAM,eAAe,CAAC;AAU3C,OAAO,EAEL,mCAAmC,GACpC,MAAM,mCAAmC,CAAC;AAC3C,OAAO,EAAW,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAGxD;;;;;GAKG;AACH,MAAM,UAAU,0BAA0B,CACxC,OAAyB,EACzB,OAAkC,EAClC,OAAwB;IAaxB,OAAO,IAAI,UAAU,CAAC,GAAG,CACvB,OAAO,EACP,OAAO,EACP,OAAO,CACR,CAAC,CAAC;AACL,CAAC;AAED,KAAK,UAAU,GAAG,CAChB,OAAyB,EACzB,OAAkC,EAClC,OAAwB;IAgBxB,MAAM,OAAO,GAAG,SAAS,CACvB,OAAO,EACP,CAAC,MAAM,EAAE,EAAE,CAAC,mCAAmC,CAAC,KAAK,CAAC,MAAM,CAAC,EAC7D,yBAAyB,CAC1B,CAAC;IACF,IAAI,CAAC,OAAO,CAAC,EAAE,EAAE,CAAC;QAChB,OAAO,CAAC,OAAO,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IAC1C,CAAC;IACD,MAAM,QAAQ,GAAG,OAAO,CAAC,KAAK,CAAC;IAC/B,MAAM,KAAK,GAAG,IAAI,CAAC;IAEnB,MAAM,WAAW,GAAG;QAClB,MAAM,EAAE,YAAY,CAAC,QAAQ,EAAE,QAAQ,CAAC,MAAM,EAAE;YAC9C,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,SAAS;SACxB,CAAC;KACH,CAAC;IACF,MAAM,KAAK,GAAG,UAAU,CAAC,wBAAwB,CAAC,CAChD,WAAW,CACZ,CAAC;IAEF,MAAM,QAAQ,GAAG,IAAI,OAAO,CAAC,UAAU,CAAC;QACtC,MAAM,EAAE,kBAAkB;KAC3B,CAAC,CAAC,CAAC;IACJ,MAAM,aAAa,GAAG,MAAM,eAAe,CAAC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;IACvE,MAAM,eAAe,GAAG,qBAAqB,CAAC,aAAa,CAAC,CAAC;IAE7D,MAAM,OAAO,GAAG;QACd,OAAO,EAAE,OAAO,CAAC,QAAQ;QACzB,OAAO,EAAE,OAAO,EAAE,SAAS,IAAI,OAAO,CAAC,QAAQ,IAAI,EAAE;QACrD,WAAW,EAAE,uBAAuB;QACpC,YAAY,EAAE,IAAI;QAClB,gBAAgB,EAAE,eAAe;QACjC,cAAc,EAAE,OAAO,CAAC,QAAQ,CAAC,QAAQ;QACzC,WAAW,EAAE,OAAO,EAAE,OAAO;eACxB,OAAO,CAAC,QAAQ,CAAC,WAAW;eAC5B,EAAE,QAAQ,EAAE,MAAM,EAAE;QACzB,UAAU,EAAE,OAAO,EAAE,UAAU,IAAI;YACjC,KAAK;YACL,KAAK;YACL,KAAK;YACL,KAAK;YACL,KAAK;SACN;KACF,CAAC;IAEF,MAAM,UAAU,GAAG,OAAO,CAAC,cAAc,CAAC,OAAO,EAAE;QACjD,QAAQ,EAAE,eAAe;QACzB,MAAM,EAAE,KAAK;QACb,OAAO,EAAE,OAAO,EAAE,SAAS;QAC3B,IAAI,EAAE,KAAK;QACX,OAAO,EAAE,QAAQ;QACjB,IAAI,EAAE,KAAK;QACX,SAAS,EAAE,OAAO,CAAC,QAAQ,CAAC,SAAS;QACrC,SAAS,EAAE,OAAO,EAAE,SAAS,IAAI,OAAO,CAAC,QAAQ,CAAC,SAAS;eACtD,CAAC,CAAC;KACR,EAAE,OAAO,CAAC,CAAC;IACZ,IAAI,CAAC,UAAU,CAAC,EAAE,EAAE,CAAC;QACnB,OAAO,CAAC,UAAU,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IAC7C,CAAC;IACD,MAAM,IAAI,GAAG,UAAU,CAAC,KAAK,CAAC;IAE9B,MAAM,QAAQ,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC,IAAI,EAAE;QACvC,OAAO;QACP,UAAU,EAAE,EAAE;QACd,WAAW,EAAE,OAAO,CAAC,WAAW;QAChC,UAAU,EAAE,OAAO,CAAC,UAAU;KAC/B,CAAC,CAAC;IACH,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;QACjB,OAAO,CAAC,QAAQ,EAAE,EAAE,MAAM,EAAE,eAAe,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;IAChE,CAAC;IACD,OAAO,CAAC,QAAQ,EAAE;YAChB,MAAM,EAAE,UAAU;YAClB,SAAS,EAAE,IAAI;YACf,QAAQ,EAAE,QAAQ,CAAC,KAAK;SACzB,CAAC,CAAC;AACL,CAAC"}
|
|
@@ -1,16 +0,0 @@
|
|
|
1
|
-
import { LightsageMcpCore } from "../core.js";
|
|
2
|
-
import { RequestOptions } from "../lib/sdks.js";
|
|
3
|
-
import { APIError } from "../models/errors/apierror.js";
|
|
4
|
-
import { ConnectionError, InvalidRequestError, RequestAbortedError, RequestTimeoutError, UnexpectedClientError } from "../models/errors/httpclienterrors.js";
|
|
5
|
-
import { SDKValidationError } from "../models/errors/sdkvalidationerror.js";
|
|
6
|
-
import { V2EvalRunsAnalysisRequest } from "../models/v2evalrunsanalysisop.js";
|
|
7
|
-
import { APIPromise } from "../types/async.js";
|
|
8
|
-
import { Result } from "../types/fp.js";
|
|
9
|
-
/**
|
|
10
|
-
* Retrieve eval run analysis
|
|
11
|
-
*
|
|
12
|
-
* @remarks
|
|
13
|
-
* 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.
|
|
14
|
-
*/
|
|
15
|
-
export declare function evalsAnalysis(client$: LightsageMcpCore, request: V2EvalRunsAnalysisRequest, options?: RequestOptions): APIPromise<Result<Response, APIError | SDKValidationError | UnexpectedClientError | InvalidRequestError | RequestAbortedError | RequestTimeoutError | ConnectionError>>;
|
|
16
|
-
//# sourceMappingURL=evalsAnalysis.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"evalsAnalysis.d.ts","sourceRoot":"","sources":["../../src/funcs/evalsAnalysis.ts"],"names":[],"mappings":"AAKA,OAAO,EAAE,gBAAgB,EAAE,MAAM,YAAY,CAAC;AAI9C,OAAO,EAAE,cAAc,EAAE,MAAM,gBAAgB,CAAC;AAGhD,OAAO,EAAE,QAAQ,EAAE,MAAM,8BAA8B,CAAC;AACxD,OAAO,EACL,eAAe,EACf,mBAAmB,EACnB,mBAAmB,EACnB,mBAAmB,EACnB,qBAAqB,EACtB,MAAM,sCAAsC,CAAC;AAC9C,OAAO,EAAE,kBAAkB,EAAE,MAAM,wCAAwC,CAAC;AAC5E,OAAO,EACL,yBAAyB,EAE1B,MAAM,mCAAmC,CAAC;AAC3C,OAAO,EAAW,UAAU,EAAE,MAAM,mBAAmB,CAAC;AACxD,OAAO,EAAE,MAAM,EAAE,MAAM,gBAAgB,CAAC;AAExC;;;;;GAKG;AACH,wBAAgB,aAAa,CAC3B,OAAO,EAAE,gBAAgB,EACzB,OAAO,EAAE,yBAAyB,EAClC,OAAO,CAAC,EAAE,cAAc,GACvB,UAAU,CACX,MAAM,CACJ,QAAQ,EACN,QAAQ,GACR,kBAAkB,GAClB,qBAAqB,GACrB,mBAAmB,GACnB,mBAAmB,GACnB,mBAAmB,GACnB,eAAe,CAClB,CACF,CAMA"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"evalsAnalysis.js","sourceRoot":"","sources":["../../src/funcs/evalsAnalysis.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAGH,OAAO,EAAE,YAAY,EAAE,MAAM,qBAAqB,CAAC;AACnD,OAAO,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAC;AAClD,OAAO,EAAE,SAAS,EAAE,MAAM,mBAAmB,CAAC;AAE9C,OAAO,EAAE,eAAe,EAAE,qBAAqB,EAAE,MAAM,oBAAoB,CAAC;AAC5E,OAAO,EAAE,UAAU,EAAE,MAAM,eAAe,CAAC;AAU3C,OAAO,EAEL,mCAAmC,GACpC,MAAM,mCAAmC,CAAC;AAC3C,OAAO,EAAW,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAGxD;;;;;GAKG;AACH,MAAM,UAAU,aAAa,CAC3B,OAAyB,EACzB,OAAkC,EAClC,OAAwB;IAaxB,OAAO,IAAI,UAAU,CAAC,GAAG,CACvB,OAAO,EACP,OAAO,EACP,OAAO,CACR,CAAC,CAAC;AACL,CAAC;AAED,KAAK,UAAU,GAAG,CAChB,OAAyB,EACzB,OAAkC,EAClC,OAAwB;IAgBxB,MAAM,OAAO,GAAG,SAAS,CACvB,OAAO,EACP,CAAC,MAAM,EAAE,EAAE,CAAC,mCAAmC,CAAC,KAAK,CAAC,MAAM,CAAC,EAC7D,yBAAyB,CAC1B,CAAC;IACF,IAAI,CAAC,OAAO,CAAC,EAAE,EAAE,CAAC;QAChB,OAAO,CAAC,OAAO,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IAC1C,CAAC;IACD,MAAM,QAAQ,GAAG,OAAO,CAAC,KAAK,CAAC;IAC/B,MAAM,KAAK,GAAG,IAAI,CAAC;IAEnB,MAAM,WAAW,GAAG;QAClB,MAAM,EAAE,YAAY,CAAC,QAAQ,EAAE,QAAQ,CAAC,MAAM,EAAE;YAC9C,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,SAAS;SACxB,CAAC;KACH,CAAC;IACF,MAAM,KAAK,GAAG,UAAU,CAAC,iCAAiC,CAAC,CACzD,WAAW,CACZ,CAAC;IAEF,MAAM,QAAQ,GAAG,IAAI,OAAO,CAAC,UAAU,CAAC;QACtC,MAAM,EAAE,kBAAkB;KAC3B,CAAC,CAAC,CAAC;IACJ,MAAM,aAAa,GAAG,MAAM,eAAe,CAAC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;IACvE,MAAM,eAAe,GAAG,qBAAqB,CAAC,aAAa,CAAC,CAAC;IAE7D,MAAM,OAAO,GAAG;QACd,OAAO,EAAE,OAAO,CAAC,QAAQ;QACzB,OAAO,EAAE,OAAO,EAAE,SAAS,IAAI,OAAO,CAAC,QAAQ,IAAI,EAAE;QACrD,WAAW,EAAE,uBAAuB;QACpC,YAAY,EAAE,IAAI;QAClB,gBAAgB,EAAE,eAAe;QACjC,cAAc,EAAE,OAAO,CAAC,QAAQ,CAAC,QAAQ;QACzC,WAAW,EAAE,OAAO,EAAE,OAAO;eACxB,OAAO,CAAC,QAAQ,CAAC,WAAW;eAC5B,EAAE,QAAQ,EAAE,MAAM,EAAE;QACzB,UAAU,EAAE,OAAO,EAAE,UAAU,IAAI;YACjC,KAAK;YACL,KAAK;YACL,KAAK;YACL,KAAK;YACL,KAAK;SACN;KACF,CAAC;IAEF,MAAM,UAAU,GAAG,OAAO,CAAC,cAAc,CAAC,OAAO,EAAE;QACjD,QAAQ,EAAE,eAAe;QACzB,MAAM,EAAE,KAAK;QACb,OAAO,EAAE,OAAO,EAAE,SAAS;QAC3B,IAAI,EAAE,KAAK;QACX,OAAO,EAAE,QAAQ;QACjB,IAAI,EAAE,KAAK;QACX,SAAS,EAAE,OAAO,CAAC,QAAQ,CAAC,SAAS;QACrC,SAAS,EAAE,OAAO,EAAE,SAAS,IAAI,OAAO,CAAC,QAAQ,CAAC,SAAS;eACtD,CAAC,CAAC;KACR,EAAE,OAAO,CAAC,CAAC;IACZ,IAAI,CAAC,UAAU,CAAC,EAAE,EAAE,CAAC;QACnB,OAAO,CAAC,UAAU,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IAC7C,CAAC;IACD,MAAM,IAAI,GAAG,UAAU,CAAC,KAAK,CAAC;IAE9B,MAAM,QAAQ,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC,IAAI,EAAE;QACvC,OAAO;QACP,UAAU,EAAE,EAAE;QACd,WAAW,EAAE,OAAO,CAAC,WAAW;QAChC,UAAU,EAAE,OAAO,CAAC,UAAU;KAC/B,CAAC,CAAC;IACH,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;QACjB,OAAO,CAAC,QAAQ,EAAE,EAAE,MAAM,EAAE,eAAe,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;IAChE,CAAC;IACD,OAAO,CAAC,QAAQ,EAAE;YAChB,MAAM,EAAE,UAAU;YAClB,SAAS,EAAE,IAAI;YACf,QAAQ,EAAE,QAAQ,CAAC,KAAK;SACzB,CAAC,CAAC;AACL,CAAC"}
|
|
@@ -1,7 +0,0 @@
|
|
|
1
|
-
import { ToolDefinition } from "../tools.js";
|
|
2
|
-
declare const args: {
|
|
3
|
-
request: import("zod").ZodOptional<import("zod").ZodType<import("../../models/v2evalresultslistop.js").V2EvalResultsListRequest, unknown, import("zod/v4/core").$ZodTypeInternals<import("../../models/v2evalresultslistop.js").V2EvalResultsListRequest, unknown>>>;
|
|
4
|
-
};
|
|
5
|
-
export declare const tool$evalResultsV2EvalResultsList: ToolDefinition<typeof args>;
|
|
6
|
-
export {};
|
|
7
|
-
//# sourceMappingURL=evalResultsV2EvalResultsList.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"evalResultsV2EvalResultsList.d.ts","sourceRoot":"","sources":["../../../src/mcp-server/tools/evalResultsV2EvalResultsList.ts"],"names":[],"mappings":"AAOA,OAAO,EAAgB,cAAc,EAAE,MAAM,aAAa,CAAC;AAE3D,QAAA,MAAM,IAAI;;CAET,CAAC;AAEF,eAAO,MAAM,iCAAiC,EAAE,cAAc,CAAC,OAAO,IAAI,CA6BzE,CAAC"}
|
|
@@ -1,35 +0,0 @@
|
|
|
1
|
-
/*
|
|
2
|
-
* Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
|
|
3
|
-
* @generated-id: cf085b52d599
|
|
4
|
-
*/
|
|
5
|
-
import { evalResultsV2EvalResultsList } from "../../funcs/evalResultsV2EvalResultsList.js";
|
|
6
|
-
import { V2EvalResultsListRequest$zodSchema } from "../../models/v2evalresultslistop.js";
|
|
7
|
-
import { formatResult } from "../tools.js";
|
|
8
|
-
const args = {
|
|
9
|
-
request: V2EvalResultsListRequest$zodSchema.optional(),
|
|
10
|
-
};
|
|
11
|
-
export const tool$evalResultsV2EvalResultsList = {
|
|
12
|
-
name: "eval-results-v2-eval-results-list",
|
|
13
|
-
description: `List Results
|
|
14
|
-
|
|
15
|
-
Returns completed custom-eval results from all supported run flows, ordered newest first. Result IDs are different from eval run IDs.`,
|
|
16
|
-
annotations: {
|
|
17
|
-
"title": "",
|
|
18
|
-
"destructiveHint": false,
|
|
19
|
-
"idempotentHint": false,
|
|
20
|
-
"openWorldHint": false,
|
|
21
|
-
"readOnlyHint": true,
|
|
22
|
-
},
|
|
23
|
-
args,
|
|
24
|
-
tool: async (client, args, ctx) => {
|
|
25
|
-
const [result] = await evalResultsV2EvalResultsList(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
|
|
26
|
-
if (!result.ok) {
|
|
27
|
-
return {
|
|
28
|
-
content: [{ type: "text", text: result.error.message }],
|
|
29
|
-
isError: true,
|
|
30
|
-
};
|
|
31
|
-
}
|
|
32
|
-
return formatResult(result.value);
|
|
33
|
-
},
|
|
34
|
-
};
|
|
35
|
-
//# sourceMappingURL=evalResultsV2EvalResultsList.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"evalResultsV2EvalResultsList.js","sourceRoot":"","sources":["../../../src/mcp-server/tools/evalResultsV2EvalResultsList.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,EAAE,4BAA4B,EAAE,MAAM,6CAA6C,CAAC;AAC3F,OAAO,EAAE,kCAAkC,EAAE,MAAM,qCAAqC,CAAC;AACzF,OAAO,EAAE,YAAY,EAAkB,MAAM,aAAa,CAAC;AAE3D,MAAM,IAAI,GAAG;IACX,OAAO,EAAE,kCAAkC,CAAC,QAAQ,EAAE;CACvD,CAAC;AAEF,MAAM,CAAC,MAAM,iCAAiC,GAAgC;IAC5E,IAAI,EAAE,mCAAmC;IACzC,WAAW,EAAE;;sIAEuH;IACpI,WAAW,EAAE;QACX,OAAO,EAAE,EAAE;QACX,iBAAiB,EAAE,KAAK;QACxB,gBAAgB,EAAE,KAAK;QACvB,eAAe,EAAE,KAAK;QACtB,cAAc,EAAE,IAAI;KACrB;IACD,IAAI;IACJ,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE;QAChC,MAAM,CAAC,MAAM,CAAC,GAAG,MAAM,4BAA4B,CACjD,MAAM,EACN,IAAI,CAAC,OAAO,EACZ,EAAE,YAAY,EAAE,EAAE,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,EAAE,CACzC,CAAC,QAAQ,EAAE,CAAC;QAEb,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;YACf,OAAO;gBACL,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;gBACvD,OAAO,EAAE,IAAI;aACd,CAAC;QACJ,CAAC;QAED,OAAO,YAAY,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IACpC,CAAC;CACF,CAAC"}
|
|
@@ -1,7 +0,0 @@
|
|
|
1
|
-
import { ToolDefinition } from "../tools.js";
|
|
2
|
-
declare const args: {
|
|
3
|
-
request: import("zod").ZodType<import("../../models/v2evalresultsretrieveop.js").V2EvalResultsRetrieveRequest, unknown, import("zod/v4/core").$ZodTypeInternals<import("../../models/v2evalresultsretrieveop.js").V2EvalResultsRetrieveRequest, unknown>>;
|
|
4
|
-
};
|
|
5
|
-
export declare const tool$evalResultsV2EvalResultsRetrieve: ToolDefinition<typeof args>;
|
|
6
|
-
export {};
|
|
7
|
-
//# sourceMappingURL=evalResultsV2EvalResultsRetrieve.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"evalResultsV2EvalResultsRetrieve.d.ts","sourceRoot":"","sources":["../../../src/mcp-server/tools/evalResultsV2EvalResultsRetrieve.ts"],"names":[],"mappings":"AAOA,OAAO,EAAgB,cAAc,EAAE,MAAM,aAAa,CAAC;AAE3D,QAAA,MAAM,IAAI;;CAET,CAAC;AAEF,eAAO,MAAM,qCAAqC,EAAE,cAAc,CAChE,OAAO,IAAI,CA8BZ,CAAC"}
|
|
@@ -1,35 +0,0 @@
|
|
|
1
|
-
/*
|
|
2
|
-
* Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
|
|
3
|
-
* @generated-id: ca0f00d47dca
|
|
4
|
-
*/
|
|
5
|
-
import { evalResultsV2EvalResultsRetrieve } from "../../funcs/evalResultsV2EvalResultsRetrieve.js";
|
|
6
|
-
import { V2EvalResultsRetrieveRequest$zodSchema } from "../../models/v2evalresultsretrieveop.js";
|
|
7
|
-
import { formatResult } from "../tools.js";
|
|
8
|
-
const args = {
|
|
9
|
-
request: V2EvalResultsRetrieveRequest$zodSchema,
|
|
10
|
-
};
|
|
11
|
-
export const tool$evalResultsV2EvalResultsRetrieve = {
|
|
12
|
-
name: "eval-results-v2-eval-results-retrieve",
|
|
13
|
-
description: `Retrieve Result
|
|
14
|
-
|
|
15
|
-
Returns one completed custom-eval result with its output and diagnostic details.`,
|
|
16
|
-
annotations: {
|
|
17
|
-
"title": "",
|
|
18
|
-
"destructiveHint": false,
|
|
19
|
-
"idempotentHint": false,
|
|
20
|
-
"openWorldHint": false,
|
|
21
|
-
"readOnlyHint": true,
|
|
22
|
-
},
|
|
23
|
-
args,
|
|
24
|
-
tool: async (client, args, ctx) => {
|
|
25
|
-
const [result] = await evalResultsV2EvalResultsRetrieve(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
|
|
26
|
-
if (!result.ok) {
|
|
27
|
-
return {
|
|
28
|
-
content: [{ type: "text", text: result.error.message }],
|
|
29
|
-
isError: true,
|
|
30
|
-
};
|
|
31
|
-
}
|
|
32
|
-
return formatResult(result.value);
|
|
33
|
-
},
|
|
34
|
-
};
|
|
35
|
-
//# sourceMappingURL=evalResultsV2EvalResultsRetrieve.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"evalResultsV2EvalResultsRetrieve.js","sourceRoot":"","sources":["../../../src/mcp-server/tools/evalResultsV2EvalResultsRetrieve.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,EAAE,gCAAgC,EAAE,MAAM,iDAAiD,CAAC;AACnG,OAAO,EAAE,sCAAsC,EAAE,MAAM,yCAAyC,CAAC;AACjG,OAAO,EAAE,YAAY,EAAkB,MAAM,aAAa,CAAC;AAE3D,MAAM,IAAI,GAAG;IACX,OAAO,EAAE,sCAAsC;CAChD,CAAC;AAEF,MAAM,CAAC,MAAM,qCAAqC,GAE9C;IACF,IAAI,EAAE,uCAAuC;IAC7C,WAAW,EAAE;;iFAEkE;IAC/E,WAAW,EAAE;QACX,OAAO,EAAE,EAAE;QACX,iBAAiB,EAAE,KAAK;QACxB,gBAAgB,EAAE,KAAK;QACvB,eAAe,EAAE,KAAK;QACtB,cAAc,EAAE,IAAI;KACrB;IACD,IAAI;IACJ,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE;QAChC,MAAM,CAAC,MAAM,CAAC,GAAG,MAAM,gCAAgC,CACrD,MAAM,EACN,IAAI,CAAC,OAAO,EACZ,EAAE,YAAY,EAAE,EAAE,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,EAAE,CACzC,CAAC,QAAQ,EAAE,CAAC;QAEb,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;YACf,OAAO;gBACL,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;gBACvD,OAAO,EAAE,IAAI;aACd,CAAC;QACJ,CAAC;QAED,OAAO,YAAY,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IACpC,CAAC;CACF,CAAC"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"evalRunsV2EvalRunsRetrieve.d.ts","sourceRoot":"","sources":["../../../src/mcp-server/tools/evalRunsV2EvalRunsRetrieve.ts"],"names":[],"mappings":"AAOA,OAAO,EAAgB,cAAc,EAAE,MAAM,aAAa,CAAC;AAE3D,QAAA,MAAM,IAAI;;CAET,CAAC;AAEF,eAAO,MAAM,+BAA+B,EAAE,cAAc,CAAC,OAAO,IAAI,CA6BvE,CAAC"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"evalRunsV2EvalRunsRetrieve.js","sourceRoot":"","sources":["../../../src/mcp-server/tools/evalRunsV2EvalRunsRetrieve.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,EAAE,0BAA0B,EAAE,MAAM,2CAA2C,CAAC;AACvF,OAAO,EAAE,mCAAmC,EAAE,MAAM,sCAAsC,CAAC;AAC3F,OAAO,EAAE,YAAY,EAAkB,MAAM,aAAa,CAAC;AAE3D,MAAM,IAAI,GAAG;IACX,OAAO,EAAE,mCAAmC;CAC7C,CAAC;AAEF,MAAM,CAAC,MAAM,+BAA+B,GAAgC;IAC1E,IAAI,EAAE,iCAAiC;IACvC,WAAW,EAAE;;gGAEiF;IAC9F,WAAW,EAAE;QACX,OAAO,EAAE,EAAE;QACX,iBAAiB,EAAE,KAAK;QACxB,gBAAgB,EAAE,KAAK;QACvB,eAAe,EAAE,KAAK;QACtB,cAAc,EAAE,IAAI;KACrB;IACD,IAAI;IACJ,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE;QAChC,MAAM,CAAC,MAAM,CAAC,GAAG,MAAM,0BAA0B,CAC/C,MAAM,EACN,IAAI,CAAC,OAAO,EACZ,EAAE,YAAY,EAAE,EAAE,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,EAAE,CACzC,CAAC,QAAQ,EAAE,CAAC;QAEb,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;YACf,OAAO;gBACL,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;gBACvD,OAAO,EAAE,IAAI;aACd,CAAC;QACJ,CAAC;QAED,OAAO,YAAY,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IACpC,CAAC;CACF,CAAC"}
|
|
@@ -1,7 +0,0 @@
|
|
|
1
|
-
import { ToolDefinition } from "../tools.js";
|
|
2
|
-
declare const args: {
|
|
3
|
-
request: import("zod").ZodType<import("../../models/v2evalrunsanalysisop.js").V2EvalRunsAnalysisRequest, unknown, import("zod/v4/core").$ZodTypeInternals<import("../../models/v2evalrunsanalysisop.js").V2EvalRunsAnalysisRequest, unknown>>;
|
|
4
|
-
};
|
|
5
|
-
export declare const tool$evalsAnalysis: ToolDefinition<typeof args>;
|
|
6
|
-
export {};
|
|
7
|
-
//# sourceMappingURL=evalsAnalysis.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"evalsAnalysis.d.ts","sourceRoot":"","sources":["../../../src/mcp-server/tools/evalsAnalysis.ts"],"names":[],"mappings":"AAOA,OAAO,EAAgB,cAAc,EAAE,MAAM,aAAa,CAAC;AAE3D,QAAA,MAAM,IAAI;;CAET,CAAC;AAEF,eAAO,MAAM,kBAAkB,EAAE,cAAc,CAAC,OAAO,IAAI,CA6B1D,CAAC"}
|
|
@@ -1,35 +0,0 @@
|
|
|
1
|
-
/*
|
|
2
|
-
* Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
|
|
3
|
-
* @generated-id: 323a92ae7245
|
|
4
|
-
*/
|
|
5
|
-
import { evalsAnalysis } from "../../funcs/evalsAnalysis.js";
|
|
6
|
-
import { V2EvalRunsAnalysisRequest$zodSchema } from "../../models/v2evalrunsanalysisop.js";
|
|
7
|
-
import { formatResult } from "../tools.js";
|
|
8
|
-
const args = {
|
|
9
|
-
request: V2EvalRunsAnalysisRequest$zodSchema,
|
|
10
|
-
};
|
|
11
|
-
export const tool$evalsAnalysis = {
|
|
12
|
-
name: "evals-analysis",
|
|
13
|
-
description: `Retrieve eval run analysis
|
|
14
|
-
|
|
15
|
-
Returns the persisted analysis produced for each completed attempt in one eval run. The analysis includes the verdict, outcome explanation, findings, bottleneck, and analysis sections. Active runs return the analysis available so far together with current run status and progress.`,
|
|
16
|
-
annotations: {
|
|
17
|
-
"title": "",
|
|
18
|
-
"destructiveHint": false,
|
|
19
|
-
"idempotentHint": false,
|
|
20
|
-
"openWorldHint": false,
|
|
21
|
-
"readOnlyHint": true,
|
|
22
|
-
},
|
|
23
|
-
args,
|
|
24
|
-
tool: async (client, args, ctx) => {
|
|
25
|
-
const [result] = await evalsAnalysis(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
|
|
26
|
-
if (!result.ok) {
|
|
27
|
-
return {
|
|
28
|
-
content: [{ type: "text", text: result.error.message }],
|
|
29
|
-
isError: true,
|
|
30
|
-
};
|
|
31
|
-
}
|
|
32
|
-
return formatResult(result.value);
|
|
33
|
-
},
|
|
34
|
-
};
|
|
35
|
-
//# sourceMappingURL=evalsAnalysis.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"evalsAnalysis.js","sourceRoot":"","sources":["../../../src/mcp-server/tools/evalsAnalysis.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,EAAE,aAAa,EAAE,MAAM,8BAA8B,CAAC;AAC7D,OAAO,EAAE,mCAAmC,EAAE,MAAM,sCAAsC,CAAC;AAC3F,OAAO,EAAE,YAAY,EAAkB,MAAM,aAAa,CAAC;AAE3D,MAAM,IAAI,GAAG;IACX,OAAO,EAAE,mCAAmC;CAC7C,CAAC;AAEF,MAAM,CAAC,MAAM,kBAAkB,GAAgC;IAC7D,IAAI,EAAE,gBAAgB;IACtB,WAAW,EAAE;;yRAE0Q;IACvR,WAAW,EAAE;QACX,OAAO,EAAE,EAAE;QACX,iBAAiB,EAAE,KAAK;QACxB,gBAAgB,EAAE,KAAK;QACvB,eAAe,EAAE,KAAK;QACtB,cAAc,EAAE,IAAI;KACrB;IACD,IAAI;IACJ,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE;QAChC,MAAM,CAAC,MAAM,CAAC,GAAG,MAAM,aAAa,CAClC,MAAM,EACN,IAAI,CAAC,OAAO,EACZ,EAAE,YAAY,EAAE,EAAE,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,EAAE,CACzC,CAAC,QAAQ,EAAE,CAAC;QAEb,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;YACf,OAAO;gBACL,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;gBACvD,OAAO,EAAE,IAAI;aACd,CAAC;QACJ,CAAC;QAED,OAAO,YAAY,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IACpC,CAAC;CACF,CAAC"}
|
|
@@ -1,23 +0,0 @@
|
|
|
1
|
-
import * as z from "zod";
|
|
2
|
-
export type V2EvalResult = {
|
|
3
|
-
id: string;
|
|
4
|
-
eval_id?: string | null | undefined;
|
|
5
|
-
agent?: string | null | undefined;
|
|
6
|
-
prompt?: string | null | undefined;
|
|
7
|
-
prompt_index?: number | undefined;
|
|
8
|
-
repeat_index?: number | undefined;
|
|
9
|
-
passed?: boolean | null | undefined;
|
|
10
|
-
score?: number | null | undefined;
|
|
11
|
-
failure_reasons?: Array<string> | null | undefined;
|
|
12
|
-
latency_ms?: number | null | undefined;
|
|
13
|
-
created_at?: string | null | undefined;
|
|
14
|
-
output?: string | null | undefined;
|
|
15
|
-
execution_stdout?: string | null | undefined;
|
|
16
|
-
execution_stderr?: string | null | undefined;
|
|
17
|
-
exit_code?: number | null | undefined;
|
|
18
|
-
tool_calls?: Array<{
|
|
19
|
-
[k: string]: any;
|
|
20
|
-
}> | null | undefined;
|
|
21
|
-
};
|
|
22
|
-
export declare const V2EvalResult$zodSchema: z.ZodType<V2EvalResult>;
|
|
23
|
-
//# sourceMappingURL=v2evalresult.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"v2evalresult.d.ts","sourceRoot":"","sources":["../../src/models/v2evalresult.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB,MAAM,MAAM,YAAY,GAAG;IACzB,EAAE,EAAE,MAAM,CAAC;IACX,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACpC,KAAK,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IAClC,MAAM,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACnC,YAAY,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAClC,YAAY,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAClC,MAAM,CAAC,EAAE,OAAO,GAAG,IAAI,GAAG,SAAS,CAAC;IACpC,KAAK,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IAClC,eAAe,CAAC,EAAE,KAAK,CAAC,MAAM,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACnD,UAAU,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACvC,UAAU,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACvC,MAAM,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACnC,gBAAgB,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IAC7C,gBAAgB,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IAC7C,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACtC,UAAU,CAAC,EAAE,KAAK,CAAC;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;CAC7D,CAAC;AAEF,eAAO,MAAM,sBAAsB,EAAE,CAAC,CAAC,OAAO,CAAC,YAAY,CAmBzD,CAAC"}
|
|
@@ -1,24 +0,0 @@
|
|
|
1
|
-
/*
|
|
2
|
-
* Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
|
|
3
|
-
* @generated-id: af5fe155cd88
|
|
4
|
-
*/
|
|
5
|
-
import * as z from "zod";
|
|
6
|
-
export const V2EvalResult$zodSchema = z.object({
|
|
7
|
-
agent: z.string().nullable().optional().describe("Runnable agent combining the harness and model."),
|
|
8
|
-
created_at: z.string().nullable().optional(),
|
|
9
|
-
eval_id: z.string().nullable().optional(),
|
|
10
|
-
execution_stderr: z.string().nullable().optional(),
|
|
11
|
-
execution_stdout: z.string().nullable().optional(),
|
|
12
|
-
exit_code: z.int().nullable().optional(),
|
|
13
|
-
failure_reasons: z.array(z.string()).nullable().optional(),
|
|
14
|
-
id: z.string().describe("Result ID, not an eval job ID."),
|
|
15
|
-
latency_ms: z.int().nullable().optional(),
|
|
16
|
-
output: z.string().nullable().optional(),
|
|
17
|
-
passed: z.boolean().nullable().optional(),
|
|
18
|
-
prompt: z.string().nullable().optional(),
|
|
19
|
-
prompt_index: z.int().default(0),
|
|
20
|
-
repeat_index: z.int().default(0),
|
|
21
|
-
score: z.number().nullable().optional().describe("Score from 0 to 1."),
|
|
22
|
-
tool_calls: z.array(z.record(z.string(), z.any())).nullable().optional(),
|
|
23
|
-
});
|
|
24
|
-
//# sourceMappingURL=v2evalresult.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"v2evalresult.js","sourceRoot":"","sources":["../../src/models/v2evalresult.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAqBzB,MAAM,CAAC,MAAM,sBAAsB,GAA4B,CAAC,CAAC,MAAM,CAAC;IACtE,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC9C,iDAAiD,CAClD;IACD,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAC5C,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACzC,gBAAgB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAClD,gBAAgB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAClD,SAAS,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACxC,eAAe,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAC1D,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,gCAAgC,CAAC;IACzD,UAAU,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACzC,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACxC,MAAM,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACzC,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACxC,YAAY,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC;IAChC,YAAY,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC;IAChC,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAAC,oBAAoB,CAAC;IACtE,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;CACzE,CAAC,CAAC"}
|