@lightsage/mcp 0.2.1 → 0.3.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (125) hide show
  1. package/README.md +2 -2
  2. package/bin/mcp-server.js +544 -276
  3. package/bin/mcp-server.js.map +29 -22
  4. package/esm/funcs/evalResultsV2EvalResultsList.d.ts +1 -1
  5. package/esm/funcs/evalResultsV2EvalResultsList.js +1 -1
  6. package/esm/funcs/evalsAnalysis.d.ts +16 -0
  7. package/esm/funcs/evalsAnalysis.d.ts.map +1 -0
  8. package/esm/funcs/evalsAnalysis.js +88 -0
  9. package/esm/funcs/evalsAnalysis.js.map +1 -0
  10. package/esm/funcs/evalsRun.d.ts +9 -2
  11. package/esm/funcs/evalsRun.d.ts.map +1 -1
  12. package/esm/funcs/evalsRun.js +9 -2
  13. package/esm/funcs/evalsRun.js.map +1 -1
  14. package/esm/funcs/evalsTrace.d.ts +16 -0
  15. package/esm/funcs/evalsTrace.d.ts.map +1 -0
  16. package/esm/funcs/{evalsMonitor.js → evalsTrace.js} +19 -21
  17. package/esm/funcs/evalsTrace.js.map +1 -0
  18. package/esm/funcs/evalsV2EvalsCreate.d.ts +1 -1
  19. package/esm/funcs/evalsV2EvalsCreate.js +1 -1
  20. package/esm/funcs/evalsV2EvalsUpdate.d.ts +1 -1
  21. package/esm/funcs/evalsV2EvalsUpdate.js +1 -1
  22. package/esm/landing-page.js +1 -1
  23. package/esm/lib/config.d.ts +2 -2
  24. package/esm/lib/config.js +2 -2
  25. package/esm/mcp-server/mcp-server.js +1 -1
  26. package/esm/mcp-server/server.d.ts.map +1 -1
  27. package/esm/mcp-server/server.js +5 -3
  28. package/esm/mcp-server/server.js.map +1 -1
  29. package/esm/mcp-server/tools/evalResultsV2EvalResultsList.js +1 -1
  30. package/esm/mcp-server/tools/evalResultsV2EvalResultsList.js.map +1 -1
  31. package/esm/mcp-server/tools/evalsAnalysis.d.ts +7 -0
  32. package/esm/mcp-server/tools/evalsAnalysis.d.ts.map +1 -0
  33. package/esm/mcp-server/tools/evalsAnalysis.js +35 -0
  34. package/esm/mcp-server/tools/evalsAnalysis.js.map +1 -0
  35. package/esm/mcp-server/tools/evalsRun.d.ts.map +1 -1
  36. package/esm/mcp-server/tools/evalsRun.js +9 -2
  37. package/esm/mcp-server/tools/evalsRun.js.map +1 -1
  38. package/esm/mcp-server/tools/evalsTrace.d.ts +7 -0
  39. package/esm/mcp-server/tools/evalsTrace.d.ts.map +1 -0
  40. package/esm/mcp-server/tools/evalsTrace.js +35 -0
  41. package/esm/mcp-server/tools/evalsTrace.js.map +1 -0
  42. package/esm/mcp-server/tools/evalsV2EvalsCreate.js +1 -1
  43. package/esm/mcp-server/tools/evalsV2EvalsCreate.js.map +1 -1
  44. package/esm/mcp-server/tools/evalsV2EvalsUpdate.js +1 -1
  45. package/esm/mcp-server/tools/evalsV2EvalsUpdate.js.map +1 -1
  46. package/esm/models/createv2evalrequest.d.ts +1 -2
  47. package/esm/models/createv2evalrequest.d.ts.map +1 -1
  48. package/esm/models/createv2evalrequest.js +1 -2
  49. package/esm/models/createv2evalrequest.js.map +1 -1
  50. package/esm/models/createv2evalrunrequest.d.ts +3 -0
  51. package/esm/models/createv2evalrunrequest.d.ts.map +1 -1
  52. package/esm/models/createv2evalrunrequest.js +9 -9
  53. package/esm/models/createv2evalrunrequest.js.map +1 -1
  54. package/esm/models/updatev2evalrequest.d.ts +1 -2
  55. package/esm/models/updatev2evalrequest.d.ts.map +1 -1
  56. package/esm/models/updatev2evalrequest.js +1 -2
  57. package/esm/models/updatev2evalrequest.js.map +1 -1
  58. package/esm/models/v2evalrunanalysis.d.ts +18 -0
  59. package/esm/models/v2evalrunanalysis.d.ts.map +1 -0
  60. package/esm/models/v2evalrunanalysis.js +20 -0
  61. package/esm/models/v2evalrunanalysis.js.map +1 -0
  62. package/esm/models/v2evalrunanalysisresponse.d.ts +12 -0
  63. package/esm/models/v2evalrunanalysisresponse.d.ts.map +1 -0
  64. package/esm/models/v2evalrunanalysisresponse.js +15 -0
  65. package/esm/models/v2evalrunanalysisresponse.js.map +1 -0
  66. package/esm/models/v2evalrunsanalysisop.d.ts +17 -0
  67. package/esm/models/v2evalrunsanalysisop.d.ts.map +1 -0
  68. package/esm/models/v2evalrunsanalysisop.js +19 -0
  69. package/esm/models/v2evalrunsanalysisop.js.map +1 -0
  70. package/esm/models/v2evalrunstraceop.d.ts +19 -0
  71. package/esm/models/v2evalrunstraceop.d.ts.map +1 -0
  72. package/esm/models/v2evalrunstraceop.js +21 -0
  73. package/esm/models/v2evalrunstraceop.js.map +1 -0
  74. package/esm/models/v2evalruntraceevent.d.ts +20 -0
  75. package/esm/models/v2evalruntraceevent.d.ts.map +1 -0
  76. package/esm/models/v2evalruntraceevent.js +22 -0
  77. package/esm/models/v2evalruntraceevent.js.map +1 -0
  78. package/esm/models/v2evalruntraceresponse.d.ts +14 -0
  79. package/esm/models/v2evalruntraceresponse.d.ts.map +1 -0
  80. package/esm/models/v2evalruntraceresponse.js +17 -0
  81. package/esm/models/v2evalruntraceresponse.js.map +1 -0
  82. package/esm/models/v2listtasksresponse.d.ts +1 -1
  83. package/esm/models/v2taskslistop.d.ts +1 -1
  84. package/esm/tool-names.d.ts.map +1 -1
  85. package/esm/tool-names.js +10 -6
  86. package/esm/tool-names.js.map +1 -1
  87. package/package.json +1 -1
  88. package/src/funcs/evalResultsV2EvalResultsList.ts +1 -1
  89. package/src/funcs/evalsAnalysis.ts +153 -0
  90. package/src/funcs/evalsRun.ts +9 -2
  91. package/src/funcs/{evalsMonitor.ts → evalsTrace.ts} +26 -25
  92. package/src/funcs/evalsV2EvalsCreate.ts +1 -1
  93. package/src/funcs/evalsV2EvalsUpdate.ts +1 -1
  94. package/src/landing-page.ts +1 -1
  95. package/src/lib/config.ts +2 -2
  96. package/src/mcp-server/mcp-server.ts +1 -1
  97. package/src/mcp-server/server.ts +5 -3
  98. package/src/mcp-server/tools/evalResultsV2EvalResultsList.ts +1 -1
  99. package/src/mcp-server/tools/evalsAnalysis.ts +43 -0
  100. package/src/mcp-server/tools/evalsRun.ts +9 -2
  101. package/src/mcp-server/tools/{evalsMonitor.ts → evalsTrace.ts} +9 -9
  102. package/src/mcp-server/tools/evalsV2EvalsCreate.ts +1 -1
  103. package/src/mcp-server/tools/evalsV2EvalsUpdate.ts +1 -1
  104. package/src/models/createv2evalrequest.ts +2 -3
  105. package/src/models/createv2evalrunrequest.ts +16 -9
  106. package/src/models/updatev2evalrequest.ts +4 -3
  107. package/src/models/v2evalrunanalysis.ts +51 -0
  108. package/src/models/v2evalrunanalysisresponse.ts +32 -0
  109. package/src/models/v2evalrunsanalysisop.ts +48 -0
  110. package/src/models/v2evalrunstraceop.ts +58 -0
  111. package/src/models/v2evalruntraceevent.ts +43 -0
  112. package/src/models/v2evalruntraceresponse.ts +38 -0
  113. package/src/tool-names.ts +10 -6
  114. package/esm/funcs/evalsMonitor.d.ts +0 -20
  115. package/esm/funcs/evalsMonitor.d.ts.map +0 -1
  116. package/esm/funcs/evalsMonitor.js.map +0 -1
  117. package/esm/mcp-server/tools/evalsMonitor.d.ts +0 -7
  118. package/esm/mcp-server/tools/evalsMonitor.d.ts.map +0 -1
  119. package/esm/mcp-server/tools/evalsMonitor.js +0 -35
  120. package/esm/mcp-server/tools/evalsMonitor.js.map +0 -1
  121. package/esm/models/v2evalrunsmonitorop.d.ts +0 -28
  122. package/esm/models/v2evalrunsmonitorop.d.ts.map +0 -1
  123. package/esm/models/v2evalrunsmonitorop.js +0 -29
  124. package/esm/models/v2evalrunsmonitorop.js.map +0 -1
  125. package/src/models/v2evalrunsmonitorop.ts +0 -62
@@ -0,0 +1,35 @@
1
+ /*
2
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
3
+ * @generated-id: d4ce799d0d36
4
+ */
5
+ import { evalsTrace } from "../../funcs/evalsTrace.js";
6
+ import { V2EvalRunsTraceRequest$zodSchema } from "../../models/v2evalrunstraceop.js";
7
+ import { formatResult } from "../tools.js";
8
+ const args = {
9
+ request: V2EvalRunsTraceRequest$zodSchema,
10
+ };
11
+ export const tool$evalsTrace = {
12
+ name: "evals-trace",
13
+ description: `Retrieve an eval run trace
14
+
15
+ Returns a finite page of sanitized execution events for one eval run. Use \`next_after_id\` as \`after_id\` to retrieve the next page. Use the monitor operation instead when following a run live.`,
16
+ annotations: {
17
+ "title": "",
18
+ "destructiveHint": false,
19
+ "idempotentHint": false,
20
+ "openWorldHint": false,
21
+ "readOnlyHint": true,
22
+ },
23
+ args,
24
+ tool: async (client, args, ctx) => {
25
+ const [result] = await evalsTrace(client, args.request, { fetchOptions: { signal: ctx.signal } }).$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=evalsTrace.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"evalsTrace.js","sourceRoot":"","sources":["../../../src/mcp-server/tools/evalsTrace.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,EAAE,UAAU,EAAE,MAAM,2BAA2B,CAAC;AACvD,OAAO,EAAE,gCAAgC,EAAE,MAAM,mCAAmC,CAAC;AACrF,OAAO,EAAE,YAAY,EAAkB,MAAM,aAAa,CAAC;AAE3D,MAAM,IAAI,GAAG;IACX,OAAO,EAAE,gCAAgC;CAC1C,CAAC;AAEF,MAAM,CAAC,MAAM,eAAe,GAAgC;IAC1D,IAAI,EAAE,aAAa;IACnB,WAAW,EAAE;;oMAEqL;IAClM,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,UAAU,CAC/B,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"}
@@ -12,7 +12,7 @@ export const tool$evalsV2EvalsCreate = {
12
12
  name: "evals-v2-evals-create",
13
13
  description: `Create Eval
14
14
 
15
- Creates an eval with a task prompt, subject, and judge. The eval can then be run through eval runs.`,
15
+ Creates an eval with a task prompt, subject, and plain-text success criterion in \`judge\`. The eval can then be run through eval runs.`,
16
16
  annotations: {
17
17
  "title": "",
18
18
  "destructiveHint": false,
@@ -1 +1 @@
1
- {"version":3,"file":"evalsV2EvalsCreate.js","sourceRoot":"","sources":["../../../src/mcp-server/tools/evalsV2EvalsCreate.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,EAAE,kBAAkB,EAAE,MAAM,mCAAmC,CAAC;AACvE,OAAO,EAAE,6BAA6B,EAAE,MAAM,qCAAqC,CAAC;AACpF,OAAO,EAAE,YAAY,EAAkB,MAAM,aAAa,CAAC;AAE3D,MAAM,IAAI,GAAG;IACX,OAAO,EAAE,6BAA6B;CACvC,CAAC;AAEF,MAAM,CAAC,MAAM,uBAAuB,GAAgC;IAClE,IAAI,EAAE,uBAAuB;IAC7B,WAAW,EAAE;;oGAEqF;IAClG,WAAW,EAAE;QACX,OAAO,EAAE,EAAE;QACX,iBAAiB,EAAE,KAAK;QACxB,gBAAgB,EAAE,KAAK;QACvB,eAAe,EAAE,KAAK;QACtB,cAAc,EAAE,KAAK;KACtB;IACD,IAAI;IACJ,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE;QAChC,MAAM,CAAC,MAAM,CAAC,GAAG,MAAM,kBAAkB,CACvC,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
+ {"version":3,"file":"evalsV2EvalsCreate.js","sourceRoot":"","sources":["../../../src/mcp-server/tools/evalsV2EvalsCreate.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,EAAE,kBAAkB,EAAE,MAAM,mCAAmC,CAAC;AACvE,OAAO,EAAE,6BAA6B,EAAE,MAAM,qCAAqC,CAAC;AACpF,OAAO,EAAE,YAAY,EAAkB,MAAM,aAAa,CAAC;AAE3D,MAAM,IAAI,GAAG;IACX,OAAO,EAAE,6BAA6B;CACvC,CAAC;AAEF,MAAM,CAAC,MAAM,uBAAuB,GAAgC;IAClE,IAAI,EAAE,uBAAuB;IAC7B,WAAW,EAAE;;wIAEyH;IACtI,WAAW,EAAE;QACX,OAAO,EAAE,EAAE;QACX,iBAAiB,EAAE,KAAK;QACxB,gBAAgB,EAAE,KAAK;QACvB,eAAe,EAAE,KAAK;QACtB,cAAc,EAAE,KAAK;KACtB;IACD,IAAI;IACJ,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE;QAChC,MAAM,CAAC,MAAM,CAAC,GAAG,MAAM,kBAAkB,CACvC,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"}
@@ -12,7 +12,7 @@ export const tool$evalsV2EvalsUpdate = {
12
12
  name: "evals-v2-evals-update",
13
13
  description: `Update Eval
14
14
 
15
- Updates an eval's task prompt, subject, judge, or name. Omitted fields remain unchanged.`,
15
+ Updates an eval's task prompt, subject, plain-text \`judge\`, or name. Omitted fields remain unchanged.`,
16
16
  annotations: {
17
17
  "title": "",
18
18
  "destructiveHint": false,
@@ -1 +1 @@
1
- {"version":3,"file":"evalsV2EvalsUpdate.js","sourceRoot":"","sources":["../../../src/mcp-server/tools/evalsV2EvalsUpdate.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,EAAE,kBAAkB,EAAE,MAAM,mCAAmC,CAAC;AACvE,OAAO,EAAE,8BAA8B,EAAE,MAAM,iCAAiC,CAAC;AACjF,OAAO,EAAE,YAAY,EAAkB,MAAM,aAAa,CAAC;AAE3D,MAAM,IAAI,GAAG;IACX,OAAO,EAAE,8BAA8B;CACxC,CAAC;AAEF,MAAM,CAAC,MAAM,uBAAuB,GAAgC;IAClE,IAAI,EAAE,uBAAuB;IAC7B,WAAW,EAAE;;yFAE0E;IACvF,WAAW,EAAE;QACX,OAAO,EAAE,EAAE;QACX,iBAAiB,EAAE,KAAK;QACxB,gBAAgB,EAAE,KAAK;QACvB,eAAe,EAAE,KAAK;QACtB,cAAc,EAAE,KAAK;KACtB;IACD,IAAI;IACJ,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE;QAChC,MAAM,CAAC,MAAM,CAAC,GAAG,MAAM,kBAAkB,CACvC,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
+ {"version":3,"file":"evalsV2EvalsUpdate.js","sourceRoot":"","sources":["../../../src/mcp-server/tools/evalsV2EvalsUpdate.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,EAAE,kBAAkB,EAAE,MAAM,mCAAmC,CAAC;AACvE,OAAO,EAAE,8BAA8B,EAAE,MAAM,iCAAiC,CAAC;AACjF,OAAO,EAAE,YAAY,EAAkB,MAAM,aAAa,CAAC;AAE3D,MAAM,IAAI,GAAG;IACX,OAAO,EAAE,8BAA8B;CACxC,CAAC;AAEF,MAAM,CAAC,MAAM,uBAAuB,GAAgC;IAClE,IAAI,EAAE,uBAAuB;IAC7B,WAAW,EAAE;;wGAEyF;IACtG,WAAW,EAAE;QACX,OAAO,EAAE,EAAE;QACX,iBAAiB,EAAE,KAAK;QACxB,gBAAgB,EAAE,KAAK;QACvB,eAAe,EAAE,KAAK;QACtB,cAAc,EAAE,KAAK;KACtB;IACD,IAAI;IACJ,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE;QAChC,MAAM,CAAC,MAAM,CAAC,GAAG,MAAM,kBAAkB,CACvC,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"}
@@ -1,6 +1,5 @@
1
1
  import * as z from "zod";
2
2
  import { ClosedEnum } from "../types/enums.js";
3
- import { V2EvalJudge } from "./v2evaljudge.js";
4
3
  export declare const CreateV2EvalRequestType: {
5
4
  readonly Workflow: "workflow";
6
5
  readonly Mcp: "mcp";
@@ -18,7 +17,7 @@ export type CreateV2EvalRequest = {
18
17
  name: string;
19
18
  type: CreateV2EvalRequestType;
20
19
  prompt: string;
21
- judge: V2EvalJudge;
20
+ judge: string;
22
21
  mcp_server_id?: string | undefined;
23
22
  cli_id?: string | undefined;
24
23
  skill_id?: string | undefined;
@@ -1 +1 @@
1
- {"version":3,"file":"createv2evalrequest.d.ts","sourceRoot":"","sources":["../../src/models/createv2evalrequest.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAC/C,OAAO,EAAE,WAAW,EAAyB,MAAM,kBAAkB,CAAC;AAEtE,eAAO,MAAM,uBAAuB;;;;;CAK1B,CAAC;AACX,MAAM,MAAM,uBAAuB,GAAG,UAAU,CAC9C,OAAO,uBAAuB,CAC/B,CAAC;AAEF,eAAO,MAAM,iCAAiC;;;;;EAK5C,CAAC;AAEH,MAAM,MAAM,mBAAmB,GAAG;IAChC,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,uBAAuB,CAAC;IAC9B,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,EAAE,WAAW,CAAC;IACnB,aAAa,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IACnC,MAAM,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC5B,QAAQ,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;CAC/B,CAAC;AAEF,eAAO,MAAM,6BAA6B,EAAE,CAAC,CAAC,OAAO,CAAC,mBAAmB,CASrE,CAAC"}
1
+ {"version":3,"file":"createv2evalrequest.d.ts","sourceRoot":"","sources":["../../src/models/createv2evalrequest.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAE/C,eAAO,MAAM,uBAAuB;;;;;CAK1B,CAAC;AACX,MAAM,MAAM,uBAAuB,GAAG,UAAU,CAC9C,OAAO,uBAAuB,CAC/B,CAAC;AAEF,eAAO,MAAM,iCAAiC;;;;;EAK5C,CAAC;AAEH,MAAM,MAAM,mBAAmB,GAAG;IAChC,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,uBAAuB,CAAC;IAC9B,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,EAAE,MAAM,CAAC;IACd,aAAa,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IACnC,MAAM,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC5B,QAAQ,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;CAC/B,CAAC;AAEF,eAAO,MAAM,6BAA6B,EAAE,CAAC,CAAC,OAAO,CAAC,mBAAmB,CASrE,CAAC"}
@@ -3,7 +3,6 @@
3
3
  * @generated-id: f1ede886cc6f
4
4
  */
5
5
  import * as z from "zod";
6
- import { V2EvalJudge$zodSchema } from "./v2evaljudge.js";
7
6
  export const CreateV2EvalRequestType = {
8
7
  Workflow: "workflow",
9
8
  Mcp: "mcp",
@@ -19,7 +18,7 @@ export const CreateV2EvalRequestType$zodSchema = z.enum([
19
18
  export const CreateV2EvalRequest$zodSchema = z
20
19
  .object({
21
20
  cli_id: z.string().optional(),
22
- judge: V2EvalJudge$zodSchema,
21
+ judge: z.string().describe("Plain-text outcome that counts as success."),
23
22
  mcp_server_id: z.string().optional(),
24
23
  name: z.string(),
25
24
  prompt: z.string(),
@@ -1 +1 @@
1
- {"version":3,"file":"createv2evalrequest.js","sourceRoot":"","sources":["../../src/models/createv2evalrequest.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB,OAAO,EAAe,qBAAqB,EAAE,MAAM,kBAAkB,CAAC;AAEtE,MAAM,CAAC,MAAM,uBAAuB,GAAG;IACrC,QAAQ,EAAE,UAAU;IACpB,GAAG,EAAE,KAAK;IACV,GAAG,EAAE,KAAK;IACV,KAAK,EAAE,OAAO;CACN,CAAC;AAKX,MAAM,CAAC,MAAM,iCAAiC,GAAG,CAAC,CAAC,IAAI,CAAC;IACtD,UAAU;IACV,KAAK;IACL,KAAK;IACL,OAAO;CACR,CAAC,CAAC;AAYH,MAAM,CAAC,MAAM,6BAA6B,GAAmC,CAAC;KAC3E,MAAM,CAAC;IACN,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC7B,KAAK,EAAE,qBAAqB;IAC5B,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IACpC,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE;IAChB,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE;IAClB,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC/B,IAAI,EAAE,iCAAiC;CACxC,CAAC,CAAC"}
1
+ {"version":3,"file":"createv2evalrequest.js","sourceRoot":"","sources":["../../src/models/createv2evalrequest.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAGzB,MAAM,CAAC,MAAM,uBAAuB,GAAG;IACrC,QAAQ,EAAE,UAAU;IACpB,GAAG,EAAE,KAAK;IACV,GAAG,EAAE,KAAK;IACV,KAAK,EAAE,OAAO;CACN,CAAC;AAKX,MAAM,CAAC,MAAM,iCAAiC,GAAG,CAAC,CAAC,IAAI,CAAC;IACtD,UAAU;IACV,KAAK;IACL,KAAK;IACL,OAAO;CACR,CAAC,CAAC;AAYH,MAAM,CAAC,MAAM,6BAA6B,GAAmC,CAAC;KAC3E,MAAM,CAAC;IACN,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC7B,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,4CAA4C,CAAC;IACxE,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IACpC,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE;IAChB,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE;IAClB,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC/B,IAAI,EAAE,iCAAiC;CACxC,CAAC,CAAC"}
@@ -1,4 +1,7 @@
1
1
  import * as z from "zod";
2
+ /**
3
+ * Start an eval run using one of two flat request forms. Saved runs require `eval_id`, `configuration_id`, and `agent_id`. Direct create-and-run requests require `prompt`, at least one item in `judges`, and `agent_id`. Do not combine saved-run and direct-run fields; no separate mode field is required.
4
+ */
2
5
  export type CreateV2EvalRunRequest = {
3
6
  eval_id?: string | undefined;
4
7
  configuration_id?: string | undefined;
@@ -1 +1 @@
1
- {"version":3,"file":"createv2evalrunrequest.d.ts","sourceRoot":"","sources":["../../src/models/createv2evalrunrequest.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB,MAAM,MAAM,sBAAsB,GAAG;IACnC,OAAO,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC7B,gBAAgB,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IACtC,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC5B,MAAM,CAAC,EAAE,KAAK,CAAC,MAAM,CAAC,GAAG,SAAS,CAAC;IACnC,UAAU,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAChC,GAAG,CAAC,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,MAAM,CAAA;KAAE,GAAG,SAAS,CAAC;IAC1C,MAAM,CAAC,EAAE,KAAK,CAAC,MAAM,CAAC,GAAG,SAAS,CAAC;CACpC,CAAC;AAEF,eAAO,MAAM,gCAAgC,EAAE,CAAC,CAAC,OAAO,CACtD,sBAAsB,CAwBtB,CAAC"}
1
+ {"version":3,"file":"createv2evalrunrequest.d.ts","sourceRoot":"","sources":["../../src/models/createv2evalrunrequest.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB;;GAEG;AACH,MAAM,MAAM,sBAAsB,GAAG;IACnC,OAAO,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC7B,gBAAgB,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IACtC,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC5B,MAAM,CAAC,EAAE,KAAK,CAAC,MAAM,CAAC,GAAG,SAAS,CAAC;IACnC,UAAU,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAChC,GAAG,CAAC,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,MAAM,CAAA;KAAE,GAAG,SAAS,CAAC;IAC1C,MAAM,CAAC,EAAE,KAAK,CAAC,MAAM,CAAC,GAAG,SAAS,CAAC;CACpC,CAAC;AAEF,eAAO,MAAM,gCAAgC,EAAE,CAAC,CAAC,OAAO,CACtD,sBAAsB,CA4BvB,CAAC"}
@@ -4,13 +4,13 @@
4
4
  */
5
5
  import * as z from "zod";
6
6
  export const CreateV2EvalRunRequest$zodSchema = z.object({
7
- agent_id: z.string().describe("Runnable agent ID from `GET /v2/agents`."),
8
- configuration_id: z.string().optional().describe("Saved runtime configuration. Required when `eval_id` is provided."),
9
- env: z.record(z.string(), z.string()).optional().describe("Write-only environment variables for a direct run. Values are stored only in the queued job's private configuration."),
10
- eval_id: z.string().optional().describe("Existing eval definition to run. Omit to create and run an eval directly."),
11
- judges: z.array(z.string()).optional().describe("Success criteria for a direct create-and-run request. CLI clients may expose this string array as a repeatable `--judge` flag."),
12
- prompt: z.string().optional().describe("Task prompt for a direct create-and-run request."),
13
- repository: z.string().optional().describe("Repository for a direct run: a saved repository ID, built-in alias such as `nextjs-starter`, or a public GitHub HTTPS URL."),
14
- skills: z.array(z.string()).optional().describe("Public skill packages to install before the agent starts, using GitHub `owner/repository` references such as `resend/resend-skills`."),
15
- });
7
+ agent_id: z.string().describe("Required in both request forms. Runnable agent ID from `GET /v2/agents`."),
8
+ configuration_id: z.string().optional().describe("Saved-run form only. Saved runtime configuration; required when `eval_id` is provided."),
9
+ env: z.record(z.string(), z.string()).optional().describe("Direct create-and-run form only. Write-only environment variables; values are stored only in the queued job's private configuration."),
10
+ eval_id: z.string().optional().describe("Saved-run form only. Existing eval definition to run. Omit this field to use the direct create-and-run form."),
11
+ judges: z.array(z.string()).optional().describe("Direct create-and-run form only. Provide at least one success criterion. CLI clients may expose this string array as a repeatable `--judge` flag."),
12
+ prompt: z.string().optional().describe("Direct create-and-run form only. Task prompt; required when `eval_id` is omitted."),
13
+ repository: z.string().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."),
14
+ skills: z.array(z.string()).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."),
15
+ }).describe("Start an eval run using one of two flat request forms. Saved runs require `eval_id`, `configuration_id`, and `agent_id`. Direct create-and-run requests require `prompt`, at least one item in `judges`, and `agent_id`. Do not combine saved-run and direct-run fields; no separate mode field is required.");
16
16
  //# sourceMappingURL=createv2evalrunrequest.js.map
@@ -1 +1 @@
1
- {"version":3,"file":"createv2evalrunrequest.js","sourceRoot":"","sources":["../../src/models/createv2evalrunrequest.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAazB,MAAM,CAAC,MAAM,gCAAgC,GAEzC,CAAC,CAAC,MAAM,CAAC;IACX,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,0CAA0C,CAAC;IACzE,gBAAgB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC9C,mEAAmE,CACpE;IACD,GAAG,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACvD,sHAAsH,CACvH;IACD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACrC,2EAA2E,CAC5E;IACD,MAAM,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC7C,gIAAgI,CACjI;IACD,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACpC,kDAAkD,CACnD;IACD,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACxC,4HAA4H,CAC7H;IACD,MAAM,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC7C,sIAAsI,CACvI;CACF,CAAC,CAAC"}
1
+ {"version":3,"file":"createv2evalrunrequest.js","sourceRoot":"","sources":["../../src/models/createv2evalrunrequest.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAgBzB,MAAM,CAAC,MAAM,gCAAgC,GAEzC,CAAC,CAAC,MAAM,CAAC;IACX,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAC3B,0EAA0E,CAC3E;IACD,gBAAgB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC9C,wFAAwF,CACzF;IACD,GAAG,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACvD,sIAAsI,CACvI;IACD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACrC,8GAA8G,CAC/G;IACD,MAAM,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC7C,mJAAmJ,CACpJ;IACD,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACpC,mFAAmF,CACpF;IACD,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACxC,gIAAgI,CACjI;IACD,MAAM,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC7C,iPAAiP,CAClP;CACF,CAAC,CAAC,QAAQ,CACT,8SAA8S,CAC/S,CAAC"}
@@ -1,6 +1,5 @@
1
1
  import * as z from "zod";
2
2
  import { ClosedEnum } from "../types/enums.js";
3
- import { V2EvalJudge } from "./v2evaljudge.js";
4
3
  export declare const UpdateV2EvalRequestType: {
5
4
  readonly Workflow: "workflow";
6
5
  readonly Mcp: "mcp";
@@ -18,7 +17,7 @@ export type UpdateV2EvalRequest = {
18
17
  name?: string | undefined;
19
18
  type?: UpdateV2EvalRequestType | undefined;
20
19
  prompt?: string | undefined;
21
- judge?: V2EvalJudge | null | undefined;
20
+ judge?: string | undefined;
22
21
  mcp_server_id?: string | undefined;
23
22
  cli_id?: string | undefined;
24
23
  skill_id?: string | undefined;
@@ -1 +1 @@
1
- {"version":3,"file":"updatev2evalrequest.d.ts","sourceRoot":"","sources":["../../src/models/updatev2evalrequest.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAC/C,OAAO,EAAE,WAAW,EAAyB,MAAM,kBAAkB,CAAC;AAEtE,eAAO,MAAM,uBAAuB;;;;;CAK1B,CAAC;AACX,MAAM,MAAM,uBAAuB,GAAG,UAAU,CAC9C,OAAO,uBAAuB,CAC/B,CAAC;AAEF,eAAO,MAAM,iCAAiC;;;;;EAK5C,CAAC;AAEH,MAAM,MAAM,mBAAmB,GAAG;IAChC,IAAI,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC1B,IAAI,CAAC,EAAE,uBAAuB,GAAG,SAAS,CAAC;IAC3C,MAAM,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC5B,KAAK,CAAC,EAAE,WAAW,GAAG,IAAI,GAAG,SAAS,CAAC;IACvC,aAAa,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IACnC,MAAM,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC5B,QAAQ,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;CAC/B,CAAC;AAEF,eAAO,MAAM,6BAA6B,EAAE,CAAC,CAAC,OAAO,CAAC,mBAAmB,CASrE,CAAC"}
1
+ {"version":3,"file":"updatev2evalrequest.d.ts","sourceRoot":"","sources":["../../src/models/updatev2evalrequest.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAE/C,eAAO,MAAM,uBAAuB;;;;;CAK1B,CAAC;AACX,MAAM,MAAM,uBAAuB,GAAG,UAAU,CAC9C,OAAO,uBAAuB,CAC/B,CAAC;AAEF,eAAO,MAAM,iCAAiC;;;;;EAK5C,CAAC;AAEH,MAAM,MAAM,mBAAmB,GAAG;IAChC,IAAI,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC1B,IAAI,CAAC,EAAE,uBAAuB,GAAG,SAAS,CAAC;IAC3C,MAAM,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC5B,KAAK,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC3B,aAAa,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IACnC,MAAM,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC5B,QAAQ,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;CAC/B,CAAC;AAEF,eAAO,MAAM,6BAA6B,EAAE,CAAC,CAAC,OAAO,CAAC,mBAAmB,CAWrE,CAAC"}
@@ -3,7 +3,6 @@
3
3
  * @generated-id: 1f9fe5c659f2
4
4
  */
5
5
  import * as z from "zod";
6
- import { V2EvalJudge$zodSchema } from "./v2evaljudge.js";
7
6
  export const UpdateV2EvalRequestType = {
8
7
  Workflow: "workflow",
9
8
  Mcp: "mcp",
@@ -19,7 +18,7 @@ export const UpdateV2EvalRequestType$zodSchema = z.enum([
19
18
  export const UpdateV2EvalRequest$zodSchema = z
20
19
  .object({
21
20
  cli_id: z.string().optional(),
22
- judge: V2EvalJudge$zodSchema.nullable().optional(),
21
+ judge: z.string().optional().describe("Plain-text outcome that counts as success."),
23
22
  mcp_server_id: z.string().optional(),
24
23
  name: z.string().optional(),
25
24
  prompt: z.string().optional(),
@@ -1 +1 @@
1
- {"version":3,"file":"updatev2evalrequest.js","sourceRoot":"","sources":["../../src/models/updatev2evalrequest.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB,OAAO,EAAe,qBAAqB,EAAE,MAAM,kBAAkB,CAAC;AAEtE,MAAM,CAAC,MAAM,uBAAuB,GAAG;IACrC,QAAQ,EAAE,UAAU;IACpB,GAAG,EAAE,KAAK;IACV,GAAG,EAAE,KAAK;IACV,KAAK,EAAE,OAAO;CACN,CAAC;AAKX,MAAM,CAAC,MAAM,iCAAiC,GAAG,CAAC,CAAC,IAAI,CAAC;IACtD,UAAU;IACV,KAAK;IACL,KAAK;IACL,OAAO;CACR,CAAC,CAAC;AAYH,MAAM,CAAC,MAAM,6BAA6B,GAAmC,CAAC;KAC3E,MAAM,CAAC;IACN,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC7B,KAAK,EAAE,qBAAqB,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAClD,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IACpC,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC3B,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC7B,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC/B,IAAI,EAAE,iCAAiC,CAAC,QAAQ,EAAE;CACnD,CAAC,CAAC"}
1
+ {"version":3,"file":"updatev2evalrequest.js","sourceRoot":"","sources":["../../src/models/updatev2evalrequest.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAGzB,MAAM,CAAC,MAAM,uBAAuB,GAAG;IACrC,QAAQ,EAAE,UAAU;IACpB,GAAG,EAAE,KAAK;IACV,GAAG,EAAE,KAAK;IACV,KAAK,EAAE,OAAO;CACN,CAAC;AAKX,MAAM,CAAC,MAAM,iCAAiC,GAAG,CAAC,CAAC,IAAI,CAAC;IACtD,UAAU;IACV,KAAK;IACL,KAAK;IACL,OAAO;CACR,CAAC,CAAC;AAYH,MAAM,CAAC,MAAM,6BAA6B,GAAmC,CAAC;KAC3E,MAAM,CAAC;IACN,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC7B,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACnC,4CAA4C,CAC7C;IACD,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IACpC,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC3B,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC7B,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC/B,IAAI,EAAE,iCAAiC,CAAC,QAAQ,EAAE;CACnD,CAAC,CAAC"}
@@ -0,0 +1,18 @@
1
+ import * as z from "zod";
2
+ export type V2EvalRunAnalysis = {
3
+ id: string;
4
+ result_id: string;
5
+ attempt_id: string;
6
+ eval_id?: string | null | undefined;
7
+ agent_id?: string | null | undefined;
8
+ variant_key?: string | undefined;
9
+ repeat_index?: number | undefined;
10
+ analysis_version?: string | null | undefined;
11
+ analysis?: {
12
+ [k: string]: any;
13
+ } | undefined;
14
+ created_at?: string | null | undefined;
15
+ updated_at?: string | null | undefined;
16
+ };
17
+ export declare const V2EvalRunAnalysis$zodSchema: z.ZodType<V2EvalRunAnalysis>;
18
+ //# sourceMappingURL=v2evalrunanalysis.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"v2evalrunanalysis.d.ts","sourceRoot":"","sources":["../../src/models/v2evalrunanalysis.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB,MAAM,MAAM,iBAAiB,GAAG;IAC9B,EAAE,EAAE,MAAM,CAAC;IACX,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACpC,QAAQ,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACrC,WAAW,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IACjC,YAAY,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAClC,gBAAgB,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IAC7C,QAAQ,CAAC,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,GAAG,SAAS,CAAC;IAC5C,UAAU,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACvC,UAAU,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;CACxC,CAAC;AAEF,eAAO,MAAM,2BAA2B,EAAE,CAAC,CAAC,OAAO,CAAC,iBAAiB,CA6BjE,CAAC"}
@@ -0,0 +1,20 @@
1
+ /*
2
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
3
+ * @generated-id: ce9f97cba8e0
4
+ */
5
+ import * as z from "zod";
6
+ export const V2EvalRunAnalysis$zodSchema = z
7
+ .object({
8
+ agent_id: z.string().nullable().optional().describe("Runnable agent ID used for this concrete attempt."),
9
+ analysis: z.record(z.string(), z.any()).optional().describe("Persisted eval_run_analysis payload, including verdict, outcome, findings, bottleneck, and analysis sections."),
10
+ analysis_version: z.string().nullable().optional().describe("Version of the analysis pipeline that produced this payload."),
11
+ attempt_id: z.string().describe("Child eval-attempt ID used to correlate the analysis with trace events."),
12
+ created_at: z.string().nullable().optional(),
13
+ eval_id: z.string().nullable().optional().describe("Public eval definition ID, not the backing executable eval ID."),
14
+ id: z.string().describe("Persisted eval analysis ID."),
15
+ repeat_index: z.int().default(0).describe("Zero-based repeat index within the submitted run."),
16
+ result_id: z.string().describe("Concrete eval result ID analyzed for this attempt."),
17
+ updated_at: z.string().nullable().optional(),
18
+ variant_key: z.string().default("default").describe("Runtime variant that produced this result."),
19
+ });
20
+ //# sourceMappingURL=v2evalrunanalysis.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"v2evalrunanalysis.js","sourceRoot":"","sources":["../../src/models/v2evalrunanalysis.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAgBzB,MAAM,CAAC,MAAM,2BAA2B,GAAiC,CAAC;KACvE,MAAM,CAAC;IACN,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACjD,mDAAmD,CACpD;IACD,QAAQ,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACzD,+GAA+G,CAChH;IACD,gBAAgB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACzD,8DAA8D,CAC/D;IACD,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAC7B,yEAAyE,CAC1E;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,CAAC,QAAQ,CAChD,gEAAgE,CACjE;IACD,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,6BAA6B,CAAC;IACtD,YAAY,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,QAAQ,CACvC,mDAAmD,CACpD;IACD,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAC5B,oDAAoD,CACrD;IACD,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAC5C,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,QAAQ,CACjD,4CAA4C,CAC7C;CACF,CAAC,CAAC"}
@@ -0,0 +1,12 @@
1
+ import * as z from "zod";
2
+ import { PublicEvalRunProgress } from "./publicevalrunprogress.js";
3
+ import { V2EvalRunAnalysis } from "./v2evalrunanalysis.js";
4
+ export type V2EvalRunAnalysisResponse = {
5
+ run_id: string;
6
+ status: string;
7
+ progress?: PublicEvalRunProgress | undefined;
8
+ error?: string | null | undefined;
9
+ data?: Array<V2EvalRunAnalysis> | undefined;
10
+ };
11
+ export declare const V2EvalRunAnalysisResponse$zodSchema: z.ZodType<V2EvalRunAnalysisResponse>;
12
+ //# sourceMappingURL=v2evalrunanalysisresponse.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"v2evalrunanalysisresponse.d.ts","sourceRoot":"","sources":["../../src/models/v2evalrunanalysisresponse.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EACL,qBAAqB,EAEtB,MAAM,4BAA4B,CAAC;AACpC,OAAO,EACL,iBAAiB,EAElB,MAAM,wBAAwB,CAAC;AAEhC,MAAM,MAAM,yBAAyB,GAAG;IACtC,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,CAAC,EAAE,qBAAqB,GAAG,SAAS,CAAC;IAC7C,KAAK,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IAClC,IAAI,CAAC,EAAE,KAAK,CAAC,iBAAiB,CAAC,GAAG,SAAS,CAAC;CAC7C,CAAC;AAEF,eAAO,MAAM,mCAAmC,EAAE,CAAC,CAAC,OAAO,CACzD,yBAAyB,CAOzB,CAAC"}
@@ -0,0 +1,15 @@
1
+ /*
2
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
3
+ * @generated-id: 16648f81799b
4
+ */
5
+ import * as z from "zod";
6
+ import { PublicEvalRunProgress$zodSchema, } from "./publicevalrunprogress.js";
7
+ import { V2EvalRunAnalysis$zodSchema, } from "./v2evalrunanalysis.js";
8
+ export const V2EvalRunAnalysisResponse$zodSchema = z.object({
9
+ data: z.array(V2EvalRunAnalysis$zodSchema).optional(),
10
+ error: z.string().nullable().optional(),
11
+ progress: PublicEvalRunProgress$zodSchema.optional(),
12
+ run_id: z.string(),
13
+ status: z.string(),
14
+ });
15
+ //# sourceMappingURL=v2evalrunanalysisresponse.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"v2evalrunanalysisresponse.js","sourceRoot":"","sources":["../../src/models/v2evalrunanalysisresponse.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAEL,+BAA+B,GAChC,MAAM,4BAA4B,CAAC;AACpC,OAAO,EAEL,2BAA2B,GAC5B,MAAM,wBAAwB,CAAC;AAUhC,MAAM,CAAC,MAAM,mCAAmC,GAE5C,CAAC,CAAC,MAAM,CAAC;IACX,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,2BAA2B,CAAC,CAAC,QAAQ,EAAE;IACrD,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACvC,QAAQ,EAAE,+BAA+B,CAAC,QAAQ,EAAE;IACpD,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE;IAClB,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE;CACnB,CAAC,CAAC"}
@@ -0,0 +1,17 @@
1
+ import * as z from "zod";
2
+ import { HTTPValidationError } from "./httpvalidationerror.js";
3
+ import { V2EvalRunAnalysisResponse } from "./v2evalrunanalysisresponse.js";
4
+ export type V2EvalRunsAnalysisRequest = {
5
+ run_id: string;
6
+ };
7
+ export declare const V2EvalRunsAnalysisRequest$zodSchema: z.ZodType<V2EvalRunsAnalysisRequest>;
8
+ /**
9
+ * Eval run not found. Use the run ID returned when you started the eval.
10
+ */
11
+ export type V2EvalRunsAnalysisResponseBody = {
12
+ detail: string;
13
+ };
14
+ export declare const V2EvalRunsAnalysisResponseBody$zodSchema: z.ZodType<V2EvalRunsAnalysisResponseBody>;
15
+ export type V2EvalRunsAnalysisResponse = V2EvalRunAnalysisResponse | V2EvalRunsAnalysisResponseBody | HTTPValidationError;
16
+ export declare const V2EvalRunsAnalysisResponse$zodSchema: z.ZodType<V2EvalRunsAnalysisResponse>;
17
+ //# sourceMappingURL=v2evalrunsanalysisop.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"v2evalrunsanalysisop.d.ts","sourceRoot":"","sources":["../../src/models/v2evalrunsanalysisop.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EACL,mBAAmB,EAEpB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EACL,yBAAyB,EAE1B,MAAM,gCAAgC,CAAC;AAExC,MAAM,MAAM,yBAAyB,GAAG;IAAE,MAAM,EAAE,MAAM,CAAA;CAAE,CAAC;AAE3D,eAAO,MAAM,mCAAmC,EAAE,CAAC,CAAC,OAAO,CACzD,yBAAyB,CAGzB,CAAC;AAEH;;GAEG;AACH,MAAM,MAAM,8BAA8B,GAAG;IAAE,MAAM,EAAE,MAAM,CAAA;CAAE,CAAC;AAEhE,eAAO,MAAM,wCAAwC,EAAE,CAAC,CAAC,OAAO,CAC9D,8BAA8B,CAK/B,CAAC;AAEF,MAAM,MAAM,0BAA0B,GAClC,yBAAyB,GACzB,8BAA8B,GAC9B,mBAAmB,CAAC;AAExB,eAAO,MAAM,oCAAoC,EAAE,CAAC,CAAC,OAAO,CAC1D,0BAA0B,CAK1B,CAAC"}
@@ -0,0 +1,19 @@
1
+ /*
2
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
3
+ * @generated-id: 4c00eef582d7
4
+ */
5
+ import * as z from "zod";
6
+ import { HTTPValidationError$zodSchema, } from "./httpvalidationerror.js";
7
+ import { V2EvalRunAnalysisResponse$zodSchema, } from "./v2evalrunanalysisresponse.js";
8
+ export const V2EvalRunsAnalysisRequest$zodSchema = z.object({
9
+ run_id: z.string(),
10
+ });
11
+ export const V2EvalRunsAnalysisResponseBody$zodSchema = z.object({
12
+ detail: z.string(),
13
+ }).describe("Eval run not found. Use the run ID returned when you started the eval.");
14
+ export const V2EvalRunsAnalysisResponse$zodSchema = z.union([
15
+ V2EvalRunAnalysisResponse$zodSchema,
16
+ z.lazy(() => V2EvalRunsAnalysisResponseBody$zodSchema),
17
+ HTTPValidationError$zodSchema,
18
+ ]);
19
+ //# sourceMappingURL=v2evalrunsanalysisop.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"v2evalrunsanalysisop.js","sourceRoot":"","sources":["../../src/models/v2evalrunsanalysisop.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAEL,6BAA6B,GAC9B,MAAM,0BAA0B,CAAC;AAClC,OAAO,EAEL,mCAAmC,GACpC,MAAM,gCAAgC,CAAC;AAIxC,MAAM,CAAC,MAAM,mCAAmC,GAE5C,CAAC,CAAC,MAAM,CAAC;IACX,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE;CACnB,CAAC,CAAC;AAOH,MAAM,CAAC,MAAM,wCAAwC,GAEjD,CAAC,CAAC,MAAM,CAAC;IACX,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE;CACnB,CAAC,CAAC,QAAQ,CACT,wEAAwE,CACzE,CAAC;AAOF,MAAM,CAAC,MAAM,oCAAoC,GAE7C,CAAC,CAAC,KAAK,CAAC;IACV,mCAAmC;IACnC,CAAC,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC,wCAAwC,CAAC;IACtD,6BAA6B;CAC9B,CAAC,CAAC"}
@@ -0,0 +1,19 @@
1
+ import * as z from "zod";
2
+ import { HTTPValidationError } from "./httpvalidationerror.js";
3
+ import { V2EvalRunTraceResponse } from "./v2evalruntraceresponse.js";
4
+ export type V2EvalRunsTraceRequest = {
5
+ run_id: string;
6
+ after_id?: number | undefined;
7
+ limit?: number | undefined;
8
+ };
9
+ export declare const V2EvalRunsTraceRequest$zodSchema: z.ZodType<V2EvalRunsTraceRequest>;
10
+ /**
11
+ * Eval run not found. Use the run ID returned when you started the eval.
12
+ */
13
+ export type V2EvalRunsTraceResponseBody = {
14
+ detail: string;
15
+ };
16
+ export declare const V2EvalRunsTraceResponseBody$zodSchema: z.ZodType<V2EvalRunsTraceResponseBody>;
17
+ export type V2EvalRunsTraceResponse = V2EvalRunTraceResponse | V2EvalRunsTraceResponseBody | HTTPValidationError;
18
+ export declare const V2EvalRunsTraceResponse$zodSchema: z.ZodType<V2EvalRunsTraceResponse>;
19
+ //# sourceMappingURL=v2evalrunstraceop.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"v2evalrunstraceop.d.ts","sourceRoot":"","sources":["../../src/models/v2evalrunstraceop.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EACL,mBAAmB,EAEpB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EACL,sBAAsB,EAEvB,MAAM,6BAA6B,CAAC;AAErC,MAAM,MAAM,sBAAsB,GAAG;IACnC,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC9B,KAAK,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;CAC5B,CAAC;AAEF,eAAO,MAAM,gCAAgC,EAAE,CAAC,CAAC,OAAO,CACtD,sBAAsB,CAStB,CAAC;AAEH;;GAEG;AACH,MAAM,MAAM,2BAA2B,GAAG;IAAE,MAAM,EAAE,MAAM,CAAA;CAAE,CAAC;AAE7D,eAAO,MAAM,qCAAqC,EAAE,CAAC,CAAC,OAAO,CAC3D,2BAA2B,CAK5B,CAAC;AAEF,MAAM,MAAM,uBAAuB,GAC/B,sBAAsB,GACtB,2BAA2B,GAC3B,mBAAmB,CAAC;AAExB,eAAO,MAAM,iCAAiC,EAAE,CAAC,CAAC,OAAO,CACvD,uBAAuB,CAKvB,CAAC"}
@@ -0,0 +1,21 @@
1
+ /*
2
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
3
+ * @generated-id: f7cdcd3e7f8d
4
+ */
5
+ import * as z from "zod";
6
+ import { HTTPValidationError$zodSchema, } from "./httpvalidationerror.js";
7
+ import { V2EvalRunTraceResponse$zodSchema, } from "./v2evalruntraceresponse.js";
8
+ export const V2EvalRunsTraceRequest$zodSchema = z.object({
9
+ after_id: z.int().default(0).describe("Return events with IDs greater than this cursor."),
10
+ limit: z.int().default(500).describe("Maximum number of trace events to return."),
11
+ run_id: z.string(),
12
+ });
13
+ export const V2EvalRunsTraceResponseBody$zodSchema = z.object({
14
+ detail: z.string(),
15
+ }).describe("Eval run not found. Use the run ID returned when you started the eval.");
16
+ export const V2EvalRunsTraceResponse$zodSchema = z.union([
17
+ V2EvalRunTraceResponse$zodSchema,
18
+ z.lazy(() => V2EvalRunsTraceResponseBody$zodSchema),
19
+ HTTPValidationError$zodSchema,
20
+ ]);
21
+ //# sourceMappingURL=v2evalrunstraceop.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"v2evalrunstraceop.js","sourceRoot":"","sources":["../../src/models/v2evalrunstraceop.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAEL,6BAA6B,GAC9B,MAAM,0BAA0B,CAAC;AAClC,OAAO,EAEL,gCAAgC,GACjC,MAAM,6BAA6B,CAAC;AAQrC,MAAM,CAAC,MAAM,gCAAgC,GAEzC,CAAC,CAAC,MAAM,CAAC;IACX,QAAQ,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,QAAQ,CACnC,kDAAkD,CACnD;IACD,KAAK,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,QAAQ,CAClC,2CAA2C,CAC5C;IACD,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE;CACnB,CAAC,CAAC;AAOH,MAAM,CAAC,MAAM,qCAAqC,GAE9C,CAAC,CAAC,MAAM,CAAC;IACX,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE;CACnB,CAAC,CAAC,QAAQ,CACT,wEAAwE,CACzE,CAAC;AAOF,MAAM,CAAC,MAAM,iCAAiC,GAE1C,CAAC,CAAC,KAAK,CAAC;IACV,gCAAgC;IAChC,CAAC,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC,qCAAqC,CAAC;IACnD,6BAA6B;CAC9B,CAAC,CAAC"}
@@ -0,0 +1,20 @@
1
+ import * as z from "zod";
2
+ export type V2EvalRunTraceEvent = {
3
+ run_id: string;
4
+ eval_id?: string | null | undefined;
5
+ eval_name?: string | null | undefined;
6
+ configuration_id?: string | null | undefined;
7
+ agent_id?: string | null | undefined;
8
+ agent_name?: string | null | undefined;
9
+ child_job_id: string;
10
+ event_id: number;
11
+ timestamp?: string | null | undefined;
12
+ category: string;
13
+ type: string;
14
+ message: string;
15
+ data?: {
16
+ [k: string]: any;
17
+ } | undefined;
18
+ };
19
+ export declare const V2EvalRunTraceEvent$zodSchema: z.ZodType<V2EvalRunTraceEvent>;
20
+ //# sourceMappingURL=v2evalruntraceevent.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"v2evalruntraceevent.d.ts","sourceRoot":"","sources":["../../src/models/v2evalruntraceevent.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB,MAAM,MAAM,mBAAmB,GAAG;IAChC,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACpC,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACtC,gBAAgB,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IAC7C,QAAQ,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACrC,UAAU,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACvC,YAAY,EAAE,MAAM,CAAC;IACrB,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IACtC,QAAQ,EAAE,MAAM,CAAC;IACjB,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,GAAG,SAAS,CAAC;CACzC,CAAC;AAEF,eAAO,MAAM,6BAA6B,EAAE,CAAC,CAAC,OAAO,CAAC,mBAAmB,CAmBrE,CAAC"}
@@ -0,0 +1,22 @@
1
+ /*
2
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
3
+ * @generated-id: cf33bec99838
4
+ */
5
+ import * as z from "zod";
6
+ export const V2EvalRunTraceEvent$zodSchema = z
7
+ .object({
8
+ agent_id: z.string().nullable().optional(),
9
+ agent_name: z.string().nullable().optional(),
10
+ category: z.string(),
11
+ child_job_id: z.string().describe("Child eval-attempt ID associated with the event."),
12
+ configuration_id: z.string().nullable().optional(),
13
+ data: z.record(z.string(), z.any()).optional().describe("Sanitized event payload."),
14
+ eval_id: z.string().nullable().optional(),
15
+ eval_name: z.string().nullable().optional(),
16
+ event_id: z.int(),
17
+ message: z.string(),
18
+ run_id: z.string(),
19
+ timestamp: z.string().nullable().optional(),
20
+ type: z.string(),
21
+ });
22
+ //# sourceMappingURL=v2evalruntraceevent.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"v2evalruntraceevent.js","sourceRoot":"","sources":["../../src/models/v2evalruntraceevent.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAkBzB,MAAM,CAAC,MAAM,6BAA6B,GAAmC,CAAC;KAC3E,MAAM,CAAC;IACN,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAC1C,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAC5C,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE;IACpB,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAC/B,kDAAkD,CACnD;IACD,gBAAgB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAClD,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACrD,0BAA0B,CAC3B;IACD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACzC,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAC3C,QAAQ,EAAE,CAAC,CAAC,GAAG,EAAE;IACjB,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE;IACnB,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE;IAClB,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IAC3C,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE;CACjB,CAAC,CAAC"}
@@ -0,0 +1,14 @@
1
+ import * as z from "zod";
2
+ import { PublicEvalRunProgress } from "./publicevalrunprogress.js";
3
+ import { V2EvalRunTraceEvent } from "./v2evalruntraceevent.js";
4
+ export type V2EvalRunTraceResponse = {
5
+ run_id: string;
6
+ status: string;
7
+ progress?: PublicEvalRunProgress | undefined;
8
+ error?: string | null | undefined;
9
+ data?: Array<V2EvalRunTraceEvent> | undefined;
10
+ next_after_id: number;
11
+ has_more: boolean;
12
+ };
13
+ export declare const V2EvalRunTraceResponse$zodSchema: z.ZodType<V2EvalRunTraceResponse>;
14
+ //# sourceMappingURL=v2evalruntraceresponse.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"v2evalruntraceresponse.d.ts","sourceRoot":"","sources":["../../src/models/v2evalruntraceresponse.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EACL,qBAAqB,EAEtB,MAAM,4BAA4B,CAAC;AACpC,OAAO,EACL,mBAAmB,EAEpB,MAAM,0BAA0B,CAAC;AAElC,MAAM,MAAM,sBAAsB,GAAG;IACnC,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,CAAC,EAAE,qBAAqB,GAAG,SAAS,CAAC;IAC7C,KAAK,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IAClC,IAAI,CAAC,EAAE,KAAK,CAAC,mBAAmB,CAAC,GAAG,SAAS,CAAC;IAC9C,aAAa,EAAE,MAAM,CAAC;IACtB,QAAQ,EAAE,OAAO,CAAC;CACnB,CAAC;AAEF,eAAO,MAAM,gCAAgC,EAAE,CAAC,CAAC,OAAO,CACtD,sBAAsB,CAWtB,CAAC"}
@@ -0,0 +1,17 @@
1
+ /*
2
+ * Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
3
+ * @generated-id: 25d2b266ce5f
4
+ */
5
+ import * as z from "zod";
6
+ import { PublicEvalRunProgress$zodSchema, } from "./publicevalrunprogress.js";
7
+ import { V2EvalRunTraceEvent$zodSchema, } from "./v2evalruntraceevent.js";
8
+ export const V2EvalRunTraceResponse$zodSchema = z.object({
9
+ data: z.array(V2EvalRunTraceEvent$zodSchema).optional(),
10
+ error: z.string().nullable().optional(),
11
+ has_more: z.boolean(),
12
+ next_after_id: z.int().describe("Cursor to pass as `after_id` when requesting the next page."),
13
+ progress: PublicEvalRunProgress$zodSchema.optional(),
14
+ run_id: z.string(),
15
+ status: z.string(),
16
+ });
17
+ //# sourceMappingURL=v2evalruntraceresponse.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"v2evalruntraceresponse.js","sourceRoot":"","sources":["../../src/models/v2evalruntraceresponse.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAEL,+BAA+B,GAChC,MAAM,4BAA4B,CAAC;AACpC,OAAO,EAEL,6BAA6B,GAC9B,MAAM,0BAA0B,CAAC;AAYlC,MAAM,CAAC,MAAM,gCAAgC,GAEzC,CAAC,CAAC,MAAM,CAAC;IACX,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,6BAA6B,CAAC,CAAC,QAAQ,EAAE;IACvD,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACvC,QAAQ,EAAE,CAAC,CAAC,OAAO,EAAE;IACrB,aAAa,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,QAAQ,CAC7B,6DAA6D,CAC9D;IACD,QAAQ,EAAE,+BAA+B,CAAC,QAAQ,EAAE;IACpD,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE;IAClB,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE;CACnB,CAAC,CAAC"}
@@ -20,13 +20,13 @@ export declare const V2ListTasksResponseStatus: {
20
20
  export type V2ListTasksResponseStatus = ClosedEnum<typeof V2ListTasksResponseStatus>;
21
21
  export declare const V2ListTasksResponseStatus$zodSchema: z.ZodEnum<{
22
22
  completed: "completed";
23
- all: "all";
24
23
  active: "active";
25
24
  todo: "todo";
26
25
  new: "new";
27
26
  in_progress: "in_progress";
28
27
  done: "done";
29
28
  dismissed: "dismissed";
29
+ all: "all";
30
30
  }>;
31
31
  export type V2ListTasksResponse = {
32
32
  action_date: string;
@@ -15,13 +15,13 @@ export declare const V2TasksListStatus: {
15
15
  export type V2TasksListStatus = ClosedEnum<typeof V2TasksListStatus>;
16
16
  export declare const V2TasksListStatus$zodSchema: z.ZodEnum<{
17
17
  completed: "completed";
18
- all: "all";
19
18
  active: "active";
20
19
  todo: "todo";
21
20
  new: "new";
22
21
  in_progress: "in_progress";
23
22
  done: "done";
24
23
  dismissed: "dismissed";
24
+ all: "all";
25
25
  }>;
26
26
  export type V2TasksListRequest = {
27
27
  action_date?: string | undefined;
@@ -1 +1 @@
1
- {"version":3,"file":"tool-names.d.ts","sourceRoot":"","sources":["../src/tool-names.ts"],"names":[],"mappings":"AACA,eAAO,MAAM,SAAS,EAAE,KAAK,CAAC;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,WAAW,EAAE,MAAM,CAAA;CAAE,CAqNlE,CAAC"}
1
+ {"version":3,"file":"tool-names.d.ts","sourceRoot":"","sources":["../src/tool-names.ts"],"names":[],"mappings":"AACA,eAAO,MAAM,SAAS,EAAE,KAAK,CAAC;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,WAAW,EAAE,MAAM,CAAA;CAAE,CAyNlE,CAAC"}
package/esm/tool-names.js CHANGED
@@ -114,7 +114,7 @@ export const toolNames = [
114
114
  },
115
115
  {
116
116
  "name": "evals-v2-evals-create",
117
- "description": "Create Eval\n\nCreates an eval with a task prompt, subject, and judge. The eval can then be run through eval runs."
117
+ "description": "Create Eval\n\nCreates an eval with a task prompt, subject, and plain-text success criterion in `judge`. The eval can then be run through eval runs."
118
118
  },
119
119
  {
120
120
  "name": "evals-v2-evals-retrieve",
@@ -122,7 +122,7 @@ export const toolNames = [
122
122
  },
123
123
  {
124
124
  "name": "evals-v2-evals-update",
125
- "description": "Update Eval\n\nUpdates an eval's task prompt, subject, judge, or name. Omitted fields remain unchanged."
125
+ "description": "Update Eval\n\nUpdates an eval's task prompt, subject, plain-text `judge`, or name. Omitted fields remain unchanged."
126
126
  },
127
127
  {
128
128
  "name": "evals-v2-evals-delete",
@@ -150,11 +150,15 @@ export const toolNames = [
150
150
  },
151
151
  {
152
152
  "name": "evals-run",
153
- "description": "Create 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. Choose exactly one request mode. For saved mode, provide `eval_id`, `configuration_id`, and `agent_id`. For direct mode, omit `eval_id` and provide `prompt`, `judges`, and `agent_id`. Direct runs may also provide a repository, write-only environment variables, and public skill packages."
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`, `configuration_id`, and `agent_id`.\n- **Direct create-and-run:** provide `prompt`, `judges`, and `agent_id`. You may also provide a repository, write-only environment variables, and public skill packages.\n\nDo not combine saved-run and direct-run fields. No separate `mode` field is required."
154
154
  },
155
155
  {
156
- "name": "evals-monitor",
157
- "description": "Monitor Runs\n\nStreams an immediate snapshot followed by live status and sanitized activity for one public v2 eval run or the active runs captured when the connection opens. Exactly one of `run_id` or `all=true` is required."
156
+ "name": "evals-analysis",
157
+ "description": "Retrieve eval run analysis\n\nReturns 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."
158
+ },
159
+ {
160
+ "name": "evals-trace",
161
+ "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."
158
162
  },
159
163
  {
160
164
  "name": "eval-runs-v2-eval-runs-list",
@@ -166,7 +170,7 @@ export const toolNames = [
166
170
  },
167
171
  {
168
172
  "name": "eval-results-v2-eval-results-list",
169
- "description": "List Results\n\nReturns completed custom-eval results from onboarding, v1, and v2 run flows, ordered newest first. Result IDs are different from eval run IDs."
173
+ "description": "List Results\n\nReturns completed custom-eval results from all supported run flows, ordered newest first. Result IDs are different from eval run IDs."
170
174
  },
171
175
  {
172
176
  "name": "eval-results-v2-eval-results-retrieve",