@jean.gnc/harness-kit 0.14.4 → 0.14.5

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 (63) hide show
  1. package/dist/cli.js +1 -1
  2. package/dist/cli.js.map +1 -1
  3. package/dist/eval/claude-env.d.ts +3 -0
  4. package/dist/eval/claude-env.d.ts.map +1 -0
  5. package/dist/eval/claude-env.js +12 -0
  6. package/dist/eval/claude-env.js.map +1 -0
  7. package/dist/eval/{grade-deterministic.d.ts → grade/deterministic.d.ts} +3 -3
  8. package/dist/eval/grade/deterministic.d.ts.map +1 -0
  9. package/dist/eval/{grade-deterministic.js → grade/deterministic.js} +3 -3
  10. package/dist/eval/grade/deterministic.js.map +1 -0
  11. package/dist/eval/grade/index.d.ts +4 -0
  12. package/dist/eval/grade/index.d.ts.map +1 -0
  13. package/dist/eval/grade/index.js +4 -0
  14. package/dist/eval/grade/index.js.map +1 -0
  15. package/dist/eval/grade/results.d.ts +5 -0
  16. package/dist/eval/grade/results.d.ts.map +1 -0
  17. package/dist/eval/{grade.js → grade/results.js} +4 -4
  18. package/dist/eval/grade/results.js.map +1 -0
  19. package/dist/eval/{grade-judge.d.ts → grade/rubric.d.ts} +3 -3
  20. package/dist/eval/grade/rubric.d.ts.map +1 -0
  21. package/dist/eval/{grade-judge.js → grade/rubric.js} +1 -1
  22. package/dist/eval/grade/rubric.js.map +1 -0
  23. package/dist/eval/index.d.ts +1 -1
  24. package/dist/eval/index.d.ts.map +1 -1
  25. package/dist/eval/index.js +8 -19
  26. package/dist/eval/index.js.map +1 -1
  27. package/dist/eval/judge/claude-cli.d.ts +9 -0
  28. package/dist/eval/judge/claude-cli.d.ts.map +1 -0
  29. package/dist/eval/judge/claude-cli.js +38 -0
  30. package/dist/eval/judge/claude-cli.js.map +1 -0
  31. package/dist/eval/judge/index.d.ts +5 -0
  32. package/dist/eval/judge/index.d.ts.map +1 -0
  33. package/dist/eval/judge/index.js +5 -0
  34. package/dist/eval/judge/index.js.map +1 -0
  35. package/dist/eval/judge/select.d.ts +9 -0
  36. package/dist/eval/judge/select.d.ts.map +1 -0
  37. package/dist/eval/judge/select.js +15 -0
  38. package/dist/eval/judge/select.js.map +1 -0
  39. package/dist/eval/judge/subagent.d.ts +13 -0
  40. package/dist/eval/judge/subagent.d.ts.map +1 -0
  41. package/dist/eval/judge/subagent.js +65 -0
  42. package/dist/eval/judge/subagent.js.map +1 -0
  43. package/dist/eval/{judge.d.ts → judge/verdict.d.ts} +2 -7
  44. package/dist/eval/judge/verdict.d.ts.map +1 -0
  45. package/dist/eval/judge/verdict.js +15 -0
  46. package/dist/eval/judge/verdict.js.map +1 -0
  47. package/dist/eval/runner.d.ts.map +1 -1
  48. package/dist/eval/runner.js +2 -6
  49. package/dist/eval/runner.js.map +1 -1
  50. package/dist/eval/score.d.ts +1 -2
  51. package/dist/eval/score.d.ts.map +1 -1
  52. package/dist/eval/score.js.map +1 -1
  53. package/package.json +1 -2
  54. package/dist/eval/grade-deterministic.d.ts.map +0 -1
  55. package/dist/eval/grade-deterministic.js.map +0 -1
  56. package/dist/eval/grade-judge.d.ts.map +0 -1
  57. package/dist/eval/grade-judge.js.map +0 -1
  58. package/dist/eval/grade.d.ts +0 -5
  59. package/dist/eval/grade.d.ts.map +0 -1
  60. package/dist/eval/grade.js.map +0 -1
  61. package/dist/eval/judge.d.ts.map +0 -1
  62. package/dist/eval/judge.js +0 -55
  63. package/dist/eval/judge.js.map +0 -1
package/dist/cli.js CHANGED
@@ -233,7 +233,7 @@ const evalCmd = defineCommand({
233
233
  },
234
234
  "judge-model": {
235
235
  type: "string",
236
- description: "model for the solving-tier LLM judge (default: claude-sonnet-4-5)",
236
+ description: "model for the solving-tier LLM judge (default: claude-haiku-4-5)",
237
237
  },
238
238
  "solving-timeout": {
239
239
  type: "string",
package/dist/cli.js.map CHANGED
@@ -1 +1 @@
1
- {"version":3,"file":"cli.js","sourceRoot":"","sources":["../src/cli.ts"],"names":[],"mappings":";AACA,OAAO,EAAE,YAAY,EAAE,MAAM,SAAS,CAAC;AACvC,OAAO,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;AAEzC,OAAO,EAAE,aAAa,EAAE,OAAO,EAAE,MAAM,OAAO,CAAC;AAC/C,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,OAAO,EAAE,KAAK,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAC;AACpD,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAEpC,OAAO,EAAE,OAAO,EAAE,MAAM,cAAc,CAAC;AACvC,OAAO,EAAE,KAAK,EAAyC,MAAM,kBAAkB,CAAC;AAChF,OAAO,EAAE,aAAa,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,iBAAiB,CAAC;AACjE,OAAO,EAAE,KAAK,EAAa,MAAM,kBAAkB,CAAC;AACpD,OAAO,EAAE,WAAW,EAAE,MAAM,iBAAiB,CAAC;AAC9C,OAAO,EAAE,iBAAiB,EAAE,OAAO,EAAE,SAAS,EAAE,MAAM,EAAE,MAAM,oBAAoB,CAAC;AACnF,OAAO,EAAE,gBAAgB,EAAE,MAAM,mBAAmB,CAAC;AACrD,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AACtD,OAAO,EAAE,cAAc,EAAE,qBAAqB,EAAE,MAAM,sBAAsB,CAAC;AAE7E,MAAM,iBAAiB,GAAG,CAAC,CAAC,MAAM,CAAC,EAAE,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;AAEnE,MAAM,OAAO,GAAG,aAAa,CAAC,IAAI,GAAG,CAAC,iBAAiB,EAAE,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;AAC3E,MAAM,GAAG,GAAG,iBAAiB,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC;AAE/E,MAAM,UAAU,GAAG,aAAa,CAAC;IAC/B,IAAI,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,WAAW,EAAE,kCAAkC,EAAE;IAC1E,IAAI,EAAE;QACJ,GAAG,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,OAAO,EAAE,WAAW,EAAE,aAAa,EAAE;QACrE,GAAG,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,QAAQ,EAAE,WAAW,EAAE,aAAa,EAAE;QACtE,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,GAAG,EAAE,WAAW,EAAE,sCAAsC,EAAE;QAC3F,MAAM,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,OAAO,EAAE,KAAK,EAAE,WAAW,EAAE,sBAAsB,EAAE;KACjF;IACD,GAAG,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,EAAE,EAAE;QACtB,MAAM,OAAO,CAAC;YACZ,OAAO,EAAE,IAAI,CAAC,GAAG;YACjB,OAAO,EAAE,IAAI,CAAC,GAAG;YACjB,QAAQ,EAAE,IAAI,CAAC,IAAI;YACnB,MAAM,EAAE,IAAI,CAAC,MAAM;SACpB,CAAC,CAAC;IACL,CAAC;CACF,CAAC,CAAC;AAEH,MAAM,WAAW,GAAG;IAClB,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,QAAQ,EAAE,WAAW,EAAE,WAAW,EAAE;IACrE,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,GAAG,EAAE,WAAW,EAAE,sCAAsC,EAAE;IAC3F,MAAM,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,OAAO,EAAE,KAAK,EAAE,WAAW,EAAE,sBAAsB,EAAE;IAChF,SAAS,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,OAAO,EAAE,KAAK,EAAE,WAAW,EAAE,6BAA6B,EAAE;CAClF,CAAC;AAEX,MAAM,UAAU,GAAG,aAAa,CAAC;IAC/B,IAAI,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,WAAW,EAAE,qDAAqD,EAAE;IAC7F,IAAI,EAAE;QACJ,GAAG,WAAW;QACd,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,OAAO,EAAE,WAAW,EAAE,gCAAgC,EAAE;KAC1F;IACD,GAAG,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,EAAE,EAAE;QACtB,MAAM,IAAI,GAAG,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACzC,MAAM,OAAO,GAAG,MAAM,qBAAqB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACvD,MAAM,OAAO,CAAC;YACZ,QAAQ,EAAE,IAAI,CAAC,IAAI;YACnB,QAAQ,EAAE,IAAI,CAAC,IAAI;YACnB,OAAO;YACP,IAAI;YACJ,MAAM,EAAE,IAAI,CAAC,MAAM;YACnB,MAAM,EAAE,IAAI,CAAC,SAAS,CAAC;SACxB,CAAC,CAAC;IACL,CAAC;CACF,CAAC,CAAC;AAEH,MAAM,YAAY,GAAG,aAAa,CAAC;IACjC,IAAI,EAAE,EAAE,IAAI,EAAE,WAAW,EAAE,WAAW,EAAE,8CAA8C,EAAE;IACxF,IAAI,EAAE,WAAW;IACjB,GAAG,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,EAAE,EAAE;QACtB,MAAM,OAAO,GAAG,MAAM,qBAAqB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACvD,MAAM,SAAS,CAAC;YACd,QAAQ,EAAE,IAAI,CAAC,IAAI;YACnB,QAAQ,EAAE,IAAI,CAAC,IAAI;YACnB,OAAO;YACP,MAAM,EAAE,IAAI,CAAC,MAAM;YACnB,MAAM,EAAE,IAAI,CAAC,SAAS,CAAC;SACxB,CAAC,CAAC;IACL,CAAC;CACF,CAAC,CAAC;AAEH,MAAM,SAAS,GAAG,aAAa,CAAC;IAC9B,IAAI,EAAE;QACJ,IAAI,EAAE,QAAQ;QACd,WAAW,EAAE,gFAAgF;KAC9F;IACD,IAAI,EAAE;QACJ,GAAG,WAAW;QACd,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,OAAO,EAAE,WAAW,EAAE,gCAAgC,EAAE;KAC1F;IACD,GAAG,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,EAAE,EAAE;QACtB,MAAM,IAAI,GAAG,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACzC,MAAM,OAAO,GAAG,MAAM,qBAAqB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACvD,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC;YAC1B,QAAQ,EAAE,IAAI,CAAC,IAAI;YACnB,QAAQ,EAAE,IAAI,CAAC,IAAI;YACnB,OAAO;YACP,IAAI;YACJ,MAAM,EAAE,IAAI,CAAC,MAAM;YACnB,MAAM,EAAE,IAAI,CAAC,SAAS,CAAC;SACxB,CAAC,CAAC;QACH,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;YACf,OAAO,CAAC,KAAK,CAAC,iBAAiB,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC;YAC/C,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,CAAC;IACH,CAAC;CACF,CAAC,CAAC;AAEH,MAAM,OAAO,GAAG,aAAa,CAAC;IAC5B,IAAI,EAAE;QACJ,IAAI,EAAE,MAAM;QACZ,WAAW,EAAE,2EAA2E;KACzF;IACD,IAAI,EAAE;QACJ,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,GAAG,EAAE,WAAW,EAAE,4BAA4B,EAAE;QACjF,WAAW,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,QAAQ,EAAE,IAAI,EAAE,WAAW,EAAE,kBAAkB,EAAE;QAChF,OAAO,EAAE;YACP,IAAI,EAAE,QAAQ;YACd,QAAQ,EAAE,IAAI;YACd,WAAW,EAAE,mCAAmC,cAAc,EAAE;iBAC7D,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;iBAClB,IAAI,CAAC,IAAI,CAAC,GAAG;SACjB;QACD,MAAM,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,OAAO,EAAE,KAAK,EAAE,WAAW,EAAE,sBAAsB,EAAE;KACjF;IACD,GAAG,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,EAAE,EAAE;QACtB,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO;aACzB,KAAK,CAAC,GAAG,CAAC;aACV,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;aACpB,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;QAC/B,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;YAAE,MAAM,IAAI,KAAK,CAAC,yCAAyC,CAAC,CAAC;QACrF,cAAc,CAAC,OAAO,EAAE,cAAc,EAAE,CAAC,CAAC;QAC1C,MAAM,WAAW,CAAC;YAChB,QAAQ,EAAE,IAAI,CAAC,IAAI;YACnB,WAAW,EAAE,IAAI,CAAC,WAAW;YAC7B,OAAO;YACP,MAAM,EAAE,IAAI,CAAC,MAAM;SACpB,CAAC,CAAC;IACL,CAAC;CACF,CAAC,CAAC;AAEH,MAAM,QAAQ,GAAG,aAAa,CAAC;IAC7B,IAAI,EAAE;QACJ,IAAI,EAAE,OAAO;QACb,WAAW,EAAE,4BAA4B;KAC1C;IACD,IAAI,EAAE;QACJ,GAAG,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,OAAO,EAAE,WAAW,EAAE,aAAa,EAAE;QACrE,MAAM,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,OAAO,EAAE,KAAK,EAAE,WAAW,EAAE,2BAA2B,EAAE;KACtF;IACD,GAAG,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,EAAE,EAAE;QACtB,MAAM,MAAM,GAAG,MAAM,KAAK,CAAC,EAAE,OAAO,EAAE,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC;QAClD,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE,CAAC;YACjB,MAAM,SAAS,GAAG,MAAM,CAAC,cAAc,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC,MAAM,IAAI,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YAC7F,MAAM,KAAK,GAAG,MAAM,CAAC,cAAc,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC;YAC9E,OAAO,CAAC,GAAG,CACT,WAAW,KAAK,kBAAkB,MAAM,CAAC,cAAc,CAAC,MAAM,aAAa,SAAS,GAAG,CACxF,CAAC;YACF,OAAO,CAAC,GAAG,CAAC,WAAW,MAAM,CAAC,YAAY,eAAe,CAAC,CAAC;YAC3D,IAAI,MAAM,CAAC,QAAQ,CAAC,MAAM,GAAG,CAAC;gBAAE,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;QAClD,CAAC;QACD,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YAChC,OAAO,CAAC,GAAG,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC,CAAC;QAChC,CAAC;QACD,IAAI,CAAC,IAAI,CAAC,MAAM,IAAI,MAAM,CAAC,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC/C,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;YAChB,OAAO,CAAC,GAAG,CAAC,GAAG,MAAM,CAAC,QAAQ,CAAC,MAAM,WAAW,CAAC,CAAC;QACpD,CAAC;QACD,IAAI,CAAC,IAAI,CAAC,MAAM,IAAI,MAAM,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC;YAAE,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;QAClE,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,UAAU,EAAE,CAAC;YAClC,OAAO,CAAC,GAAG,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC,CAAC;QAClC,CAAC;QACD,IAAI,MAAM,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACjC,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE,CAAC;gBACjB,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;gBAChB,OAAO,CAAC,GAAG,CAAC,GAAG,MAAM,CAAC,UAAU,CAAC,MAAM,aAAa,CAAC,CAAC;YACxD,CAAC;YACD,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,CAAC;IACH,CAAC;CACF,CAAC,CAAC;AAEH,SAAS,eAAe,CAAC,CAAqB;IAC5C,OAAO,GAAG,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,MAAM,OAAO,CAAC,CAAC,KAAK,QAAQ,CAAC,CAAC,OAAO,EAAE,CAAC;AAC1E,CAAC;AAED,SAAS,aAAa,CAAC,CAAU;IAC/B,OAAO,GAAG,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,MAAM,cAAc,CAAC,CAAC,OAAO,EAAE,CAAC;AAClE,CAAC;AAED,MAAM,OAAO,GAAG,aAAa,CAAC;IAC5B,IAAI,EAAE;QACJ,IAAI,EAAE,MAAM;QACZ,WAAW,EAAE,qEAAqE;KACnF;IACD,IAAI,EAAE;QACJ,GAAG,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,QAAQ,EAAE,WAAW,EAAE,aAAa,EAAE;QACtE,MAAM,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,OAAO,EAAE,KAAK,EAAE,WAAW,EAAE,2BAA2B,EAAE;KACtF;IACD,GAAG,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,EAAE,EAAE;QACtB,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,EAAE,OAAO,EAAE,IAAI,CAAC,GAAG,EAAE,MAAM,EAAE,IAAI,CAAC,MAAM,EAAE,CAAC,CAAC;QACtE,IAAI,MAAM,CAAC,UAAU,GAAG,CAAC;YAAE,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAC7C,CAAC;CACF,CAAC,CAAC;AAEH,SAAS,SAAS,CAAC,KAAa;IAC9B,IAAK,KAA2B,CAAC,QAAQ,CAAC,KAAK,CAAC;QAAE,OAAO,KAAa,CAAC;IACvE,MAAM,IAAI,KAAK,CAAC,iBAAiB,KAAK,aAAa,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;AACzE,CAAC;AAED,SAAS,gBAAgB,CAAC,KAAa,EAAE,IAAY;IACnD,MAAM,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC;IAC7B,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,IAAI,MAAM,GAAG,CAAC,EAAE,CAAC;QAC5C,MAAM,IAAI,KAAK,CAAC,KAAK,IAAI,qCAAqC,KAAK,GAAG,CAAC,CAAC;IAC1E,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,MAAM,OAAO,GAAG,aAAa,CAAC;IAC5B,IAAI,EAAE;QACJ,IAAI,EAAE,MAAM;QACZ,WAAW,EAAE,6EAA6E;KAC3F;IACD,IAAI,EAAE;QACJ,KAAK,EAAE;YACL,IAAI,EAAE,QAAQ;YACd,OAAO,EAAE,eAAe;YACxB,WAAW,EAAE,8BAA8B;SAC5C;QACD,GAAG,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,GAAG,EAAE,WAAW,EAAE,0CAA0C,EAAE;QAC9F,KAAK,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,gCAAgC,EAAE;QACxE,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,gCAAgC,EAAE;QACvE,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,uBAAuB,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,EAAE;QACjF,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,4CAA4C,EAAE;QACnF,WAAW,EAAE;YACX,IAAI,EAAE,QAAQ;YACd,OAAO,EAAE,GAAG;YACZ,WAAW,EAAE,+EAA+E;SAC7F;QACD,KAAK,EAAE;YACL,IAAI,EAAE,QAAQ;YACd,WAAW,EAAE,wDAAwD;SACtE;QACD,aAAa,EAAE;YACb,IAAI,EAAE,QAAQ;YACd,WAAW,EAAE,mEAAmE;SACjF;QACD,iBAAiB,EAAE;YACjB,IAAI,EAAE,QAAQ;YACd,WAAW,EAAE,iEAAiE;SAC/E;QACD,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,6CAA6C,EAAE;KACrF;IACD,GAAG,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,EAAE,EAAE;QACtB,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC;YAC3B,QAAQ,EAAE,IAAI,CAAC,KAAK;YACpB,GAAG,EAAE,IAAI,CAAC,GAAG;YACb,WAAW,EAAE,gBAAgB,CAAC,IAAI,CAAC,WAAW,EAAE,aAAa,CAAC;YAC9D,GAAG,CAAC,IAAI,CAAC,KAAK,KAAK,SAAS,IAAI,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE,CAAC;YACtD,GAAG,CAAC,IAAI,CAAC,IAAI,KAAK,SAAS,IAAI,EAAE,MAAM,EAAE,IAAI,CAAC,IAAI,EAAE,CAAC;YACrD,GAAG,CAAC,IAAI,CAAC,IAAI,KAAK,SAAS,IAAI,EAAE,IAAI,EAAE,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;YAC9D,GAAG,CAAC,IAAI,CAAC,IAAI,KAAK,SAAS,IAAI,EAAE,IAAI,EAAE,gBAAgB,CAAC,IAAI,CAAC,IAAI,EAAE,MAAM,CAAC,EAAE,CAAC;YAC7E,GAAG,CAAC,IAAI,CAAC,KAAK,KAAK,SAAS,IAAI,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE,CAAC;YACtD,GAAG,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,SAAS,IAAI,EAAE,UAAU,EAAE,IAAI,CAAC,aAAa,CAAC,EAAE,CAAC;YAC7E,GAAG,CAAC,IAAI,CAAC,iBAAiB,CAAC,KAAK,SAAS,IAAI;gBAC3C,gBAAgB,EAAE,gBAAgB,CAAC,IAAI,CAAC,iBAAiB,CAAC,EAAE,iBAAiB,CAAC,GAAG,IAAI;aACtF,CAAC;SACH,CAAC,CAAC;QAEH,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;YACf,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,KAAK;gBAAE,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,KAAK,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC;YACvE,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,CAAC;QAED,OAAO,CAAC,GAAG,CAAC,aAAa,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC;QACzC,IAAI,IAAI,CAAC,IAAI,EAAE,CAAC;YACd,MAAM,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;YACrD,MAAM,SAAS,CAAC,IAAI,CAAC,IAAI,EAAE,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,IAAI,CAAC,CAAC;QAC1D,CAAC;QACD,IAAI,MAAM,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC;YAAE,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAC/C,CAAC;CACF,CAAC,CAAC;AAEH,MAAM,IAAI,GAAG,aAAa,CAAC;IACzB,IAAI,EAAE;QACJ,IAAI,EAAE,SAAS;QACf,OAAO,EAAE,GAAG,CAAC,OAAO;QACpB,WAAW,EACT,yFAAyF;KAC5F;IACD,WAAW,EAAE;QACX,KAAK,EAAE,QAAQ;QACf,OAAO,EAAE,UAAU;QACnB,IAAI,EAAE,OAAO;QACb,IAAI,EAAE,OAAO;QACb,OAAO,EAAE,UAAU;QACnB,IAAI,EAAE,OAAO;QACb,SAAS,EAAE,YAAY;QACvB,MAAM,EAAE,SAAS;KAClB;CACF,CAAC,CAAC;AAEH,MAAM,OAAO,CAAC,IAAI,CAAC,CAAC"}
1
+ {"version":3,"file":"cli.js","sourceRoot":"","sources":["../src/cli.ts"],"names":[],"mappings":";AACA,OAAO,EAAE,YAAY,EAAE,MAAM,SAAS,CAAC;AACvC,OAAO,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;AAEzC,OAAO,EAAE,aAAa,EAAE,OAAO,EAAE,MAAM,OAAO,CAAC;AAC/C,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,OAAO,EAAE,KAAK,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAC;AACpD,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAEpC,OAAO,EAAE,OAAO,EAAE,MAAM,cAAc,CAAC;AACvC,OAAO,EAAE,KAAK,EAAyC,MAAM,kBAAkB,CAAC;AAChF,OAAO,EAAE,aAAa,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,iBAAiB,CAAC;AACjE,OAAO,EAAE,KAAK,EAAa,MAAM,kBAAkB,CAAC;AACpD,OAAO,EAAE,WAAW,EAAE,MAAM,iBAAiB,CAAC;AAC9C,OAAO,EAAE,iBAAiB,EAAE,OAAO,EAAE,SAAS,EAAE,MAAM,EAAE,MAAM,oBAAoB,CAAC;AACnF,OAAO,EAAE,gBAAgB,EAAE,MAAM,mBAAmB,CAAC;AACrD,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AACtD,OAAO,EAAE,cAAc,EAAE,qBAAqB,EAAE,MAAM,sBAAsB,CAAC;AAE7E,MAAM,iBAAiB,GAAG,CAAC,CAAC,MAAM,CAAC,EAAE,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;AAEnE,MAAM,OAAO,GAAG,aAAa,CAAC,IAAI,GAAG,CAAC,iBAAiB,EAAE,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;AAC3E,MAAM,GAAG,GAAG,iBAAiB,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC;AAE/E,MAAM,UAAU,GAAG,aAAa,CAAC;IAC/B,IAAI,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,WAAW,EAAE,kCAAkC,EAAE;IAC1E,IAAI,EAAE;QACJ,GAAG,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,OAAO,EAAE,WAAW,EAAE,aAAa,EAAE;QACrE,GAAG,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,QAAQ,EAAE,WAAW,EAAE,aAAa,EAAE;QACtE,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,GAAG,EAAE,WAAW,EAAE,sCAAsC,EAAE;QAC3F,MAAM,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,OAAO,EAAE,KAAK,EAAE,WAAW,EAAE,sBAAsB,EAAE;KACjF;IACD,GAAG,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,EAAE,EAAE;QACtB,MAAM,OAAO,CAAC;YACZ,OAAO,EAAE,IAAI,CAAC,GAAG;YACjB,OAAO,EAAE,IAAI,CAAC,GAAG;YACjB,QAAQ,EAAE,IAAI,CAAC,IAAI;YACnB,MAAM,EAAE,IAAI,CAAC,MAAM;SACpB,CAAC,CAAC;IACL,CAAC;CACF,CAAC,CAAC;AAEH,MAAM,WAAW,GAAG;IAClB,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,QAAQ,EAAE,WAAW,EAAE,WAAW,EAAE;IACrE,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,GAAG,EAAE,WAAW,EAAE,sCAAsC,EAAE;IAC3F,MAAM,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,OAAO,EAAE,KAAK,EAAE,WAAW,EAAE,sBAAsB,EAAE;IAChF,SAAS,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,OAAO,EAAE,KAAK,EAAE,WAAW,EAAE,6BAA6B,EAAE;CAClF,CAAC;AAEX,MAAM,UAAU,GAAG,aAAa,CAAC;IAC/B,IAAI,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,WAAW,EAAE,qDAAqD,EAAE;IAC7F,IAAI,EAAE;QACJ,GAAG,WAAW;QACd,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,OAAO,EAAE,WAAW,EAAE,gCAAgC,EAAE;KAC1F;IACD,GAAG,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,EAAE,EAAE;QACtB,MAAM,IAAI,GAAG,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACzC,MAAM,OAAO,GAAG,MAAM,qBAAqB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACvD,MAAM,OAAO,CAAC;YACZ,QAAQ,EAAE,IAAI,CAAC,IAAI;YACnB,QAAQ,EAAE,IAAI,CAAC,IAAI;YACnB,OAAO;YACP,IAAI;YACJ,MAAM,EAAE,IAAI,CAAC,MAAM;YACnB,MAAM,EAAE,IAAI,CAAC,SAAS,CAAC;SACxB,CAAC,CAAC;IACL,CAAC;CACF,CAAC,CAAC;AAEH,MAAM,YAAY,GAAG,aAAa,CAAC;IACjC,IAAI,EAAE,EAAE,IAAI,EAAE,WAAW,EAAE,WAAW,EAAE,8CAA8C,EAAE;IACxF,IAAI,EAAE,WAAW;IACjB,GAAG,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,EAAE,EAAE;QACtB,MAAM,OAAO,GAAG,MAAM,qBAAqB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACvD,MAAM,SAAS,CAAC;YACd,QAAQ,EAAE,IAAI,CAAC,IAAI;YACnB,QAAQ,EAAE,IAAI,CAAC,IAAI;YACnB,OAAO;YACP,MAAM,EAAE,IAAI,CAAC,MAAM;YACnB,MAAM,EAAE,IAAI,CAAC,SAAS,CAAC;SACxB,CAAC,CAAC;IACL,CAAC;CACF,CAAC,CAAC;AAEH,MAAM,SAAS,GAAG,aAAa,CAAC;IAC9B,IAAI,EAAE;QACJ,IAAI,EAAE,QAAQ;QACd,WAAW,EAAE,gFAAgF;KAC9F;IACD,IAAI,EAAE;QACJ,GAAG,WAAW;QACd,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,OAAO,EAAE,WAAW,EAAE,gCAAgC,EAAE;KAC1F;IACD,GAAG,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,EAAE,EAAE;QACtB,MAAM,IAAI,GAAG,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACzC,MAAM,OAAO,GAAG,MAAM,qBAAqB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACvD,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC;YAC1B,QAAQ,EAAE,IAAI,CAAC,IAAI;YACnB,QAAQ,EAAE,IAAI,CAAC,IAAI;YACnB,OAAO;YACP,IAAI;YACJ,MAAM,EAAE,IAAI,CAAC,MAAM;YACnB,MAAM,EAAE,IAAI,CAAC,SAAS,CAAC;SACxB,CAAC,CAAC;QACH,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;YACf,OAAO,CAAC,KAAK,CAAC,iBAAiB,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC;YAC/C,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,CAAC;IACH,CAAC;CACF,CAAC,CAAC;AAEH,MAAM,OAAO,GAAG,aAAa,CAAC;IAC5B,IAAI,EAAE;QACJ,IAAI,EAAE,MAAM;QACZ,WAAW,EAAE,2EAA2E;KACzF;IACD,IAAI,EAAE;QACJ,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,GAAG,EAAE,WAAW,EAAE,4BAA4B,EAAE;QACjF,WAAW,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,QAAQ,EAAE,IAAI,EAAE,WAAW,EAAE,kBAAkB,EAAE;QAChF,OAAO,EAAE;YACP,IAAI,EAAE,QAAQ;YACd,QAAQ,EAAE,IAAI;YACd,WAAW,EAAE,mCAAmC,cAAc,EAAE;iBAC7D,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;iBAClB,IAAI,CAAC,IAAI,CAAC,GAAG;SACjB;QACD,MAAM,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,OAAO,EAAE,KAAK,EAAE,WAAW,EAAE,sBAAsB,EAAE;KACjF;IACD,GAAG,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,EAAE,EAAE;QACtB,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO;aACzB,KAAK,CAAC,GAAG,CAAC;aACV,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;aACpB,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;QAC/B,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;YAAE,MAAM,IAAI,KAAK,CAAC,yCAAyC,CAAC,CAAC;QACrF,cAAc,CAAC,OAAO,EAAE,cAAc,EAAE,CAAC,CAAC;QAC1C,MAAM,WAAW,CAAC;YAChB,QAAQ,EAAE,IAAI,CAAC,IAAI;YACnB,WAAW,EAAE,IAAI,CAAC,WAAW;YAC7B,OAAO;YACP,MAAM,EAAE,IAAI,CAAC,MAAM;SACpB,CAAC,CAAC;IACL,CAAC;CACF,CAAC,CAAC;AAEH,MAAM,QAAQ,GAAG,aAAa,CAAC;IAC7B,IAAI,EAAE;QACJ,IAAI,EAAE,OAAO;QACb,WAAW,EAAE,4BAA4B;KAC1C;IACD,IAAI,EAAE;QACJ,GAAG,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,OAAO,EAAE,WAAW,EAAE,aAAa,EAAE;QACrE,MAAM,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,OAAO,EAAE,KAAK,EAAE,WAAW,EAAE,2BAA2B,EAAE;KACtF;IACD,GAAG,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,EAAE,EAAE;QACtB,MAAM,MAAM,GAAG,MAAM,KAAK,CAAC,EAAE,OAAO,EAAE,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC;QAClD,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE,CAAC;YACjB,MAAM,SAAS,GAAG,MAAM,CAAC,cAAc,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC,MAAM,IAAI,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YAC7F,MAAM,KAAK,GAAG,MAAM,CAAC,cAAc,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC;YAC9E,OAAO,CAAC,GAAG,CACT,WAAW,KAAK,kBAAkB,MAAM,CAAC,cAAc,CAAC,MAAM,aAAa,SAAS,GAAG,CACxF,CAAC;YACF,OAAO,CAAC,GAAG,CAAC,WAAW,MAAM,CAAC,YAAY,eAAe,CAAC,CAAC;YAC3D,IAAI,MAAM,CAAC,QAAQ,CAAC,MAAM,GAAG,CAAC;gBAAE,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;QAClD,CAAC;QACD,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YAChC,OAAO,CAAC,GAAG,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC,CAAC;QAChC,CAAC;QACD,IAAI,CAAC,IAAI,CAAC,MAAM,IAAI,MAAM,CAAC,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC/C,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;YAChB,OAAO,CAAC,GAAG,CAAC,GAAG,MAAM,CAAC,QAAQ,CAAC,MAAM,WAAW,CAAC,CAAC;QACpD,CAAC;QACD,IAAI,CAAC,IAAI,CAAC,MAAM,IAAI,MAAM,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC;YAAE,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;QAClE,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,UAAU,EAAE,CAAC;YAClC,OAAO,CAAC,GAAG,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC,CAAC;QAClC,CAAC;QACD,IAAI,MAAM,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACjC,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE,CAAC;gBACjB,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;gBAChB,OAAO,CAAC,GAAG,CAAC,GAAG,MAAM,CAAC,UAAU,CAAC,MAAM,aAAa,CAAC,CAAC;YACxD,CAAC;YACD,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,CAAC;IACH,CAAC;CACF,CAAC,CAAC;AAEH,SAAS,eAAe,CAAC,CAAqB;IAC5C,OAAO,GAAG,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,MAAM,OAAO,CAAC,CAAC,KAAK,QAAQ,CAAC,CAAC,OAAO,EAAE,CAAC;AAC1E,CAAC;AAED,SAAS,aAAa,CAAC,CAAU;IAC/B,OAAO,GAAG,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,MAAM,cAAc,CAAC,CAAC,OAAO,EAAE,CAAC;AAClE,CAAC;AAED,MAAM,OAAO,GAAG,aAAa,CAAC;IAC5B,IAAI,EAAE;QACJ,IAAI,EAAE,MAAM;QACZ,WAAW,EAAE,qEAAqE;KACnF;IACD,IAAI,EAAE;QACJ,GAAG,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,QAAQ,EAAE,WAAW,EAAE,aAAa,EAAE;QACtE,MAAM,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,OAAO,EAAE,KAAK,EAAE,WAAW,EAAE,2BAA2B,EAAE;KACtF;IACD,GAAG,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,EAAE,EAAE;QACtB,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,EAAE,OAAO,EAAE,IAAI,CAAC,GAAG,EAAE,MAAM,EAAE,IAAI,CAAC,MAAM,EAAE,CAAC,CAAC;QACtE,IAAI,MAAM,CAAC,UAAU,GAAG,CAAC;YAAE,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAC7C,CAAC;CACF,CAAC,CAAC;AAEH,SAAS,SAAS,CAAC,KAAa;IAC9B,IAAK,KAA2B,CAAC,QAAQ,CAAC,KAAK,CAAC;QAAE,OAAO,KAAa,CAAC;IACvE,MAAM,IAAI,KAAK,CAAC,iBAAiB,KAAK,aAAa,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;AACzE,CAAC;AAED,SAAS,gBAAgB,CAAC,KAAa,EAAE,IAAY;IACnD,MAAM,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC;IAC7B,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,IAAI,MAAM,GAAG,CAAC,EAAE,CAAC;QAC5C,MAAM,IAAI,KAAK,CAAC,KAAK,IAAI,qCAAqC,KAAK,GAAG,CAAC,CAAC;IAC1E,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,MAAM,OAAO,GAAG,aAAa,CAAC;IAC5B,IAAI,EAAE;QACJ,IAAI,EAAE,MAAM;QACZ,WAAW,EAAE,6EAA6E;KAC3F;IACD,IAAI,EAAE;QACJ,KAAK,EAAE;YACL,IAAI,EAAE,QAAQ;YACd,OAAO,EAAE,eAAe;YACxB,WAAW,EAAE,8BAA8B;SAC5C;QACD,GAAG,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,GAAG,EAAE,WAAW,EAAE,0CAA0C,EAAE;QAC9F,KAAK,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,gCAAgC,EAAE;QACxE,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,gCAAgC,EAAE;QACvE,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,uBAAuB,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,EAAE;QACjF,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,4CAA4C,EAAE;QACnF,WAAW,EAAE;YACX,IAAI,EAAE,QAAQ;YACd,OAAO,EAAE,GAAG;YACZ,WAAW,EAAE,+EAA+E;SAC7F;QACD,KAAK,EAAE;YACL,IAAI,EAAE,QAAQ;YACd,WAAW,EAAE,wDAAwD;SACtE;QACD,aAAa,EAAE;YACb,IAAI,EAAE,QAAQ;YACd,WAAW,EAAE,kEAAkE;SAChF;QACD,iBAAiB,EAAE;YACjB,IAAI,EAAE,QAAQ;YACd,WAAW,EAAE,iEAAiE;SAC/E;QACD,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,6CAA6C,EAAE;KACrF;IACD,GAAG,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,EAAE,EAAE;QACtB,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC;YAC3B,QAAQ,EAAE,IAAI,CAAC,KAAK;YACpB,GAAG,EAAE,IAAI,CAAC,GAAG;YACb,WAAW,EAAE,gBAAgB,CAAC,IAAI,CAAC,WAAW,EAAE,aAAa,CAAC;YAC9D,GAAG,CAAC,IAAI,CAAC,KAAK,KAAK,SAAS,IAAI,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE,CAAC;YACtD,GAAG,CAAC,IAAI,CAAC,IAAI,KAAK,SAAS,IAAI,EAAE,MAAM,EAAE,IAAI,CAAC,IAAI,EAAE,CAAC;YACrD,GAAG,CAAC,IAAI,CAAC,IAAI,KAAK,SAAS,IAAI,EAAE,IAAI,EAAE,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;YAC9D,GAAG,CAAC,IAAI,CAAC,IAAI,KAAK,SAAS,IAAI,EAAE,IAAI,EAAE,gBAAgB,CAAC,IAAI,CAAC,IAAI,EAAE,MAAM,CAAC,EAAE,CAAC;YAC7E,GAAG,CAAC,IAAI,CAAC,KAAK,KAAK,SAAS,IAAI,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE,CAAC;YACtD,GAAG,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,SAAS,IAAI,EAAE,UAAU,EAAE,IAAI,CAAC,aAAa,CAAC,EAAE,CAAC;YAC7E,GAAG,CAAC,IAAI,CAAC,iBAAiB,CAAC,KAAK,SAAS,IAAI;gBAC3C,gBAAgB,EAAE,gBAAgB,CAAC,IAAI,CAAC,iBAAiB,CAAC,EAAE,iBAAiB,CAAC,GAAG,IAAI;aACtF,CAAC;SACH,CAAC,CAAC;QAEH,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;YACf,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,KAAK;gBAAE,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,KAAK,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC;YACvE,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,CAAC;QAED,OAAO,CAAC,GAAG,CAAC,aAAa,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC;QACzC,IAAI,IAAI,CAAC,IAAI,EAAE,CAAC;YACd,MAAM,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;YACrD,MAAM,SAAS,CAAC,IAAI,CAAC,IAAI,EAAE,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,IAAI,CAAC,CAAC;QAC1D,CAAC;QACD,IAAI,MAAM,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC;YAAE,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAC/C,CAAC;CACF,CAAC,CAAC;AAEH,MAAM,IAAI,GAAG,aAAa,CAAC;IACzB,IAAI,EAAE;QACJ,IAAI,EAAE,SAAS;QACf,OAAO,EAAE,GAAG,CAAC,OAAO;QACpB,WAAW,EACT,yFAAyF;KAC5F;IACD,WAAW,EAAE;QACX,KAAK,EAAE,QAAQ;QACf,OAAO,EAAE,UAAU;QACnB,IAAI,EAAE,OAAO;QACb,IAAI,EAAE,OAAO;QACb,OAAO,EAAE,UAAU;QACnB,IAAI,EAAE,OAAO;QACb,SAAS,EAAE,YAAY;QACvB,MAAM,EAAE,SAAS;KAClB;CACF,CAAC,CAAC;AAEH,MAAM,OAAO,CAAC,IAAI,CAAC,CAAC"}
@@ -0,0 +1,3 @@
1
+ export declare function scrubEnv(base: NodeJS.ProcessEnv, keys: readonly string[]): NodeJS.ProcessEnv;
2
+ export declare const SUBPROCESS_ENV_OVERRIDES: readonly ["ANTHROPIC_API_KEY", "ANTHROPIC_AUTH_TOKEN", "CLAUDECODE"];
3
+ //# sourceMappingURL=claude-env.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"claude-env.d.ts","sourceRoot":"","sources":["../../src/eval/claude-env.ts"],"names":[],"mappings":"AAAA,wBAAgB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC,UAAU,EAAE,IAAI,EAAE,SAAS,MAAM,EAAE,GAAG,MAAM,CAAC,UAAU,CAI5F;AAED,eAAO,MAAM,wBAAwB,sEAI3B,CAAC"}
@@ -0,0 +1,12 @@
1
+ export function scrubEnv(base, keys) {
2
+ const env = { ...base };
3
+ for (const key of keys)
4
+ delete env[key];
5
+ return env;
6
+ }
7
+ export const SUBPROCESS_ENV_OVERRIDES = [
8
+ "ANTHROPIC_API_KEY",
9
+ "ANTHROPIC_AUTH_TOKEN",
10
+ "CLAUDECODE",
11
+ ];
12
+ //# sourceMappingURL=claude-env.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"claude-env.js","sourceRoot":"","sources":["../../src/eval/claude-env.ts"],"names":[],"mappings":"AAAA,MAAM,UAAU,QAAQ,CAAC,IAAuB,EAAE,IAAuB;IACvE,MAAM,GAAG,GAAG,EAAE,GAAG,IAAI,EAAE,CAAC;IACxB,KAAK,MAAM,GAAG,IAAI,IAAI;QAAE,OAAO,GAAG,CAAC,GAAG,CAAC,CAAC;IACxC,OAAO,GAAG,CAAC;AACb,CAAC;AAED,MAAM,CAAC,MAAM,wBAAwB,GAAG;IACtC,mBAAmB;IACnB,sBAAsB;IACtB,YAAY;CACJ,CAAC"}
@@ -1,9 +1,9 @@
1
- import type { SolvingCapture } from "./capture.js";
2
- import type { Assertion } from "./schema.js";
1
+ import type { SolvingCapture } from "../capture.js";
2
+ import type { Assertion } from "../schema.js";
3
3
  export interface AssertionResult {
4
4
  readonly assertion: Assertion;
5
5
  readonly pass: boolean;
6
6
  readonly evidence: string;
7
7
  }
8
8
  export declare function gradeAssertions(assertions: readonly Assertion[], capture: SolvingCapture): AssertionResult[];
9
- //# sourceMappingURL=grade-deterministic.d.ts.map
9
+ //# sourceMappingURL=deterministic.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"deterministic.d.ts","sourceRoot":"","sources":["../../../src/eval/grade/deterministic.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,eAAe,CAAC;AAEpD,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,cAAc,CAAC;AAE9C,MAAM,WAAW,eAAe;IAC9B,QAAQ,CAAC,SAAS,EAAE,SAAS,CAAC;IAC9B,QAAQ,CAAC,IAAI,EAAE,OAAO,CAAC;IACvB,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;CAC3B;AAED,wBAAgB,eAAe,CAC7B,UAAU,EAAE,SAAS,SAAS,EAAE,EAChC,OAAO,EAAE,cAAc,GACtB,eAAe,EAAE,CAEnB"}
@@ -1,5 +1,5 @@
1
- import { err, ok } from "../result.js";
2
- import { asRecord } from "./detect.js";
1
+ import { err, ok } from "../../result.js";
2
+ import { asRecord } from "../detect.js";
3
3
  export function gradeAssertions(assertions, capture) {
4
4
  return assertions.map((assertion) => grade(assertion, capture));
5
5
  }
@@ -84,4 +84,4 @@ function pass(assertion, evidence) {
84
84
  function fail(assertion, evidence) {
85
85
  return { assertion, pass: false, evidence };
86
86
  }
87
- //# sourceMappingURL=grade-deterministic.js.map
87
+ //# sourceMappingURL=deterministic.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"deterministic.js","sourceRoot":"","sources":["../../../src/eval/grade/deterministic.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,GAAG,EAAE,EAAE,EAAe,MAAM,iBAAiB,CAAC;AAEvD,OAAO,EAAE,QAAQ,EAAE,MAAM,cAAc,CAAC;AASxC,MAAM,UAAU,eAAe,CAC7B,UAAgC,EAChC,OAAuB;IAEvB,OAAO,UAAU,CAAC,GAAG,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,KAAK,CAAC,SAAS,EAAE,OAAO,CAAC,CAAC,CAAC;AAClE,CAAC;AAED,SAAS,KAAK,CAAC,SAAoB,EAAE,OAAuB;IAC1D,QAAQ,SAAS,CAAC,IAAI,EAAE,CAAC;QACvB,KAAK,eAAe;YAClB,OAAO,WAAW,CAAC,SAAS,EAAE,OAAO,CAAC,UAAU,EAAE;gBAChD,WAAW,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC,SAAS,EAAE,YAAY,SAAS,CAAC,OAAO,GAAG,CAAC;gBACpE,UAAU,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC,SAAS,EAAE,iBAAiB,SAAS,CAAC,OAAO,GAAG,CAAC;aACzE,CAAC,CAAC;QAEL,KAAK,gBAAgB;YACnB,OAAO,WAAW,CAAC,SAAS,EAAE,OAAO,CAAC,UAAU,EAAE;gBAChD,WAAW,EAAE,CAAC,GAAG,EAAE,EAAE,CAAC,IAAI,CAAC,SAAS,EAAE,oBAAoB,GAAG,GAAG,CAAC;gBACjE,UAAU,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC,SAAS,EAAE,YAAY,SAAS,CAAC,OAAO,GAAG,CAAC;aACpE,CAAC,CAAC;QAEL,KAAK,UAAU;YACb,OAAO,KAAK,CAAC,OAAO,CAAC,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,CAAC;gBAC5C,CAAC,CAAC,IAAI,CAAC,SAAS,EAAE,QAAQ,SAAS,CAAC,IAAI,EAAE,CAAC;gBAC3C,CAAC,CAAC,IAAI,CAAC,SAAS,EAAE,gBAAgB,KAAK,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAEpE,KAAK,eAAe;YAClB,OAAO,KAAK,CAAC,OAAO,CAAC,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,CAAC;gBAC5C,CAAC,CAAC,IAAI,CAAC,SAAS,EAAE,qBAAqB,SAAS,CAAC,IAAI,EAAE,CAAC;gBACxD,CAAC,CAAC,IAAI,CAAC,SAAS,EAAE,eAAe,SAAS,CAAC,IAAI,EAAE,CAAC,CAAC;QAEvD,KAAK,WAAW;YACd,OAAO,cAAc,CAAC,SAAS,EAAE,OAAO,CAAC,CAAC;IAC9C,CAAC;AACH,CAAC;AAOD,SAAS,WAAW,CAClB,SAA2E,EAC3E,IAAY,EACZ,OAAqB;IAErB,MAAM,OAAO,GAAG,cAAc,CAAC,SAAS,CAAC,OAAO,EAAE,SAAS,CAAC,KAAK,CAAC,CAAC;IACnE,IAAI,CAAC,OAAO,CAAC,EAAE;QAAE,OAAO,IAAI,CAAC,SAAS,EAAE,OAAO,CAAC,KAAK,CAAC,CAAC;IACvD,MAAM,GAAG,GAAG,OAAO,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAChC,OAAO,GAAG,KAAK,IAAI,CAAC,CAAC,CAAC,OAAO,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC;AACxE,CAAC;AAED,SAAS,cAAc,CACrB,SAAoD,EACpD,OAAuB;IAEvB,IAAI,CAAC,OAAO,CAAC,OAAO,EAAE,SAAS,CAAC,IAAI,CAAC,EAAE,CAAC;QACtC,OAAO,IAAI,CACT,SAAS,EACT,GAAG,SAAS,CAAC,IAAI,yBAAyB,YAAY,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAC/E,CAAC;IACJ,CAAC;IACD,IAAI,SAAS,CAAC,cAAc,KAAK,SAAS,EAAE,CAAC;QAC3C,OAAO,IAAI,CAAC,SAAS,EAAE,SAAS,SAAS,CAAC,IAAI,EAAE,CAAC,CAAC;IACpD,CAAC;IACD,MAAM,OAAO,GAAG,OAAO,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,SAAS,CAAC,IAAI,CAAC,EAAE,OAAO,IAAI,EAAE,CAAC;IACrF,MAAM,OAAO,GAAG,cAAc,CAAC,SAAS,CAAC,cAAc,EAAE,SAAS,CAAC,KAAK,CAAC,CAAC;IAC1E,IAAI,CAAC,OAAO,CAAC,EAAE;QAAE,OAAO,IAAI,CAAC,SAAS,EAAE,OAAO,CAAC,KAAK,CAAC,CAAC;IACvD,OAAO,OAAO,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,IAAI;QACpC,CAAC,CAAC,IAAI,CAAC,SAAS,EAAE,GAAG,SAAS,CAAC,IAAI,kBAAkB,CAAC;QACtD,CAAC,CAAC,IAAI,CAAC,SAAS,EAAE,GAAG,SAAS,CAAC,IAAI,2BAA2B,SAAS,CAAC,cAAc,GAAG,CAAC,CAAC;AAC/F,CAAC;AAID,SAAS,cAAc,CAAC,OAAe,EAAE,KAAc;IACrD,IAAI,CAAC,KAAK,EAAE,CAAC;QACX,OAAO,EAAE,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC;IACjE,CAAC;IACD,IAAI,QAAgB,CAAC;IACrB,IAAI,CAAC;QACH,QAAQ,GAAG,IAAI,MAAM,CAAC,OAAO,CAAC,CAAC;IACjC,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,OAAO,GAAG,CAAC,kBAAkB,OAAO,MAAO,KAAe,CAAC,OAAO,EAAE,CAAC,CAAC;IACxE,CAAC;IACD,OAAO,EAAE,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC;AACxD,CAAC;AAED,SAAS,OAAO,CAAC,OAAuB,EAAE,IAAY;IACpD,OAAO,YAAY,CAAC,OAAO,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;AAC9C,CAAC;AAED,SAAS,YAAY,CAAC,OAAuB;IAC3C,OAAO,OAAO,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,EAAE,CACzC,WAAW,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,WAAW,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CACtD,CAAC;AACJ,CAAC;AAED,SAAS,WAAW,CAAC,IAAY;IAC/B,OAAO,IAAI,KAAK,OAAO,IAAI,IAAI,KAAK,MAAM,CAAC;AAC7C,CAAC;AAED,SAAS,WAAW,CAAC,KAAc;IACjC,MAAM,IAAI,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,WAAW,CAAC,CAAC;IAC1C,OAAO,OAAO,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;AAChD,CAAC;AAED,SAAS,KAAK,CAAC,OAAuB;IACpC,OAAO,OAAO,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACrD,CAAC;AAED,SAAS,IAAI,CAAC,SAAoB,EAAE,QAAgB;IAClD,OAAO,EAAE,SAAS,EAAE,IAAI,EAAE,IAAI,EAAE,QAAQ,EAAE,CAAC;AAC7C,CAAC;AAED,SAAS,IAAI,CAAC,SAAoB,EAAE,QAAgB;IAClD,OAAO,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,QAAQ,EAAE,CAAC;AAC9C,CAAC"}
@@ -0,0 +1,4 @@
1
+ export { gradeResults } from "./results.js";
2
+ export { gradeRubric, type RubricResult, type DimensionResult } from "./rubric.js";
3
+ export { gradeAssertions, type AssertionResult } from "./deterministic.js";
4
+ //# sourceMappingURL=index.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/eval/grade/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,MAAM,cAAc,CAAC;AAC5C,OAAO,EAAE,WAAW,EAAE,KAAK,YAAY,EAAE,KAAK,eAAe,EAAE,MAAM,aAAa,CAAC;AACnF,OAAO,EAAE,eAAe,EAAE,KAAK,eAAe,EAAE,MAAM,oBAAoB,CAAC"}
@@ -0,0 +1,4 @@
1
+ export { gradeResults } from "./results.js";
2
+ export { gradeRubric } from "./rubric.js";
3
+ export { gradeAssertions } from "./deterministic.js";
4
+ //# sourceMappingURL=index.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/eval/grade/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,MAAM,cAAc,CAAC;AAC5C,OAAO,EAAE,WAAW,EAA2C,MAAM,aAAa,CAAC;AACnF,OAAO,EAAE,eAAe,EAAwB,MAAM,oBAAoB,CAAC"}
@@ -0,0 +1,5 @@
1
+ import type { Judge } from "../judge/index.js";
2
+ import type { CaseReport } from "../report.js";
3
+ import type { CaseResult } from "../runner.js";
4
+ export declare function gradeResults(results: readonly CaseResult[], judge: Judge | undefined): Promise<CaseReport[]>;
5
+ //# sourceMappingURL=results.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"results.d.ts","sourceRoot":"","sources":["../../../src/eval/grade/results.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,KAAK,EAAE,MAAM,mBAAmB,CAAC;AAC/C,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,cAAc,CAAC;AAC/C,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,cAAc,CAAC;AAI/C,wBAAsB,YAAY,CAChC,OAAO,EAAE,SAAS,UAAU,EAAE,EAC9B,KAAK,EAAE,KAAK,GAAG,SAAS,GACvB,OAAO,CAAC,UAAU,EAAE,CAAC,CAEvB"}
@@ -1,6 +1,6 @@
1
- import { gradeAssertions } from "./grade-deterministic.js";
2
- import { gradeRubric } from "./grade-judge.js";
3
- import { scoreCase, scoreSolving } from "./score.js";
1
+ import { gradeAssertions } from "./deterministic.js";
2
+ import { gradeRubric } from "./rubric.js";
3
+ import { scoreCase, scoreSolving } from "../score.js";
4
4
  export async function gradeResults(results, judge) {
5
5
  return Promise.all(results.map(async (result) => gradeOne(result, judge)));
6
6
  }
@@ -22,4 +22,4 @@ async function gradeRun(assertions, rubric, capture, judge) {
22
22
  const rubricResult = rubric && judge ? await gradeRubric(rubric, capture.outputText, judge) : null;
23
23
  return { assertions: graded, rubric: rubricResult };
24
24
  }
25
- //# sourceMappingURL=grade.js.map
25
+ //# sourceMappingURL=results.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"results.js","sourceRoot":"","sources":["../../../src/eval/grade/results.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,eAAe,EAAE,MAAM,oBAAoB,CAAC;AACrD,OAAO,EAAE,WAAW,EAAqB,MAAM,aAAa,CAAC;AAI7D,OAAO,EAAE,SAAS,EAAE,YAAY,EAAyB,MAAM,aAAa,CAAC;AAG7E,MAAM,CAAC,KAAK,UAAU,YAAY,CAChC,OAA8B,EAC9B,KAAwB;IAExB,OAAO,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,CAAC,QAAQ,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC;AAC7E,CAAC;AAED,KAAK,UAAU,QAAQ,CAAC,MAAkB,EAAE,KAAwB;IAClE,IAAI,MAAM,CAAC,IAAI,KAAK,SAAS,EAAE,CAAC;QAC9B,MAAM,EAAE,QAAQ,EAAE,IAAI,EAAE,GAAG,MAAM,CAAC;QAClC,OAAO,EAAE,QAAQ,EAAE,KAAK,EAAE,SAAS,CAAC,QAAQ,CAAC,MAAM,EAAE,IAAI,EAAE,QAAQ,CAAC,SAAS,CAAC,EAAE,CAAC;IACnF,CAAC;IAED,MAAM,EAAE,QAAQ,EAAE,QAAQ,EAAE,GAAG,MAAM,CAAC;IACtC,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,GAAG,CAC9B,QAAQ,CAAC,GAAG,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE,CAAC,QAAQ,CAAC,QAAQ,CAAC,MAAM,EAAE,QAAQ,CAAC,MAAM,EAAE,OAAO,EAAE,KAAK,CAAC,CAAC,CAC5F,CAAC;IACF,OAAO;QACL,QAAQ;QACR,KAAK,EAAE,YAAY,CAAC,MAAM,EAAE,QAAQ,CAAC,SAAS,CAAC;QAC/C,OAAO,EAAE,EAAE,MAAM,EAAE;KACpB,CAAC;AACJ,CAAC;AAED,KAAK,UAAU,QAAQ,CACrB,UAAgC,EAChC,MAA0B,EAC1B,OAAuB,EACvB,KAAwB;IAExB,MAAM,MAAM,GAAG,eAAe,CAAC,UAAU,EAAE,OAAO,CAAC,CAAC;IACpD,MAAM,YAAY,GAChB,MAAM,IAAI,KAAK,CAAC,CAAC,CAAC,MAAM,WAAW,CAAC,MAAM,EAAE,OAAO,CAAC,UAAU,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IAChF,OAAO,EAAE,UAAU,EAAE,MAAM,EAAE,MAAM,EAAE,YAAY,EAAE,CAAC;AACtD,CAAC"}
@@ -1,5 +1,5 @@
1
- import type { Judge, Verdict } from "./judge.js";
2
- import type { Rubric } from "./schema.js";
1
+ import type { Judge, Verdict } from "../judge/index.js";
2
+ import type { Rubric } from "../schema.js";
3
3
  export interface DimensionResult {
4
4
  readonly dimension: string;
5
5
  readonly verdict: Verdict;
@@ -9,4 +9,4 @@ export interface RubricResult {
9
9
  readonly pass: boolean;
10
10
  }
11
11
  export declare function gradeRubric(rubric: Rubric, output: string, judge: Judge): Promise<RubricResult>;
12
- //# sourceMappingURL=grade-judge.d.ts.map
12
+ //# sourceMappingURL=rubric.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"rubric.d.ts","sourceRoot":"","sources":["../../../src/eval/grade/rubric.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,KAAK,EAAE,OAAO,EAAE,MAAM,mBAAmB,CAAC;AACxD,OAAO,KAAK,EAAe,MAAM,EAAE,MAAM,cAAc,CAAC;AAExD,MAAM,WAAW,eAAe;IAC9B,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,OAAO,EAAE,OAAO,CAAC;CAC3B;AAED,MAAM,WAAW,YAAY;IAC3B,QAAQ,CAAC,UAAU,EAAE,SAAS,eAAe,EAAE,CAAC;IAChD,QAAQ,CAAC,IAAI,EAAE,OAAO,CAAC;CACxB;AAED,wBAAsB,WAAW,CAC/B,MAAM,EAAE,MAAM,EACd,MAAM,EAAE,MAAM,EACd,KAAK,EAAE,KAAK,GACX,OAAO,CAAC,YAAY,CAAC,CAQvB"}
@@ -11,4 +11,4 @@ function combine(rule, dimensions) {
11
11
  return passed === dimensions.length;
12
12
  return dimensions.length > 0 && passed / dimensions.length >= rule.threshold;
13
13
  }
14
- //# sourceMappingURL=grade-judge.js.map
14
+ //# sourceMappingURL=rubric.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"rubric.js","sourceRoot":"","sources":["../../../src/eval/grade/rubric.ts"],"names":[],"mappings":"AAaA,MAAM,CAAC,KAAK,UAAU,WAAW,CAC/B,MAAc,EACd,MAAc,EACd,KAAY;IAEZ,MAAM,UAAU,GAAG,MAAM,OAAO,CAAC,GAAG,CAClC,MAAM,CAAC,UAAU,CAAC,GAAG,CAAC,KAAK,EAAE,EAAE,SAAS,EAAE,SAAS,EAAE,EAAE,EAAE,CAAC,CAAC;QACzD,SAAS;QACT,OAAO,EAAE,MAAM,KAAK,CAAC,EAAE,SAAS,EAAE,SAAS,EAAE,MAAM,EAAE,CAAC;KACvD,CAAC,CAAC,CACJ,CAAC;IACF,OAAO,EAAE,UAAU,EAAE,IAAI,EAAE,OAAO,CAAC,MAAM,CAAC,OAAO,EAAE,UAAU,CAAC,EAAE,CAAC;AACnE,CAAC;AAED,SAAS,OAAO,CAAC,IAAiB,EAAE,UAAsC;IACxE,MAAM,MAAM,GAAG,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,KAAK,IAAI,CAAC,CAAC,MAAM,CAAC;IACxE,IAAI,IAAI,CAAC,OAAO,KAAK,KAAK;QAAE,OAAO,MAAM,KAAK,UAAU,CAAC,MAAM,CAAC;IAChE,OAAO,UAAU,CAAC,MAAM,GAAG,CAAC,IAAI,MAAM,GAAG,UAAU,CAAC,MAAM,IAAI,IAAI,CAAC,SAAS,CAAC;AAC/E,CAAC"}
@@ -1,6 +1,6 @@
1
1
  import { type Result } from "../result.js";
2
2
  import { type CaseLoadError, type LoadedCase } from "./cases.js";
3
- import { type Judge } from "./judge.js";
3
+ import { type Judge } from "./judge/index.js";
4
4
  import { type EvalReport } from "./report.js";
5
5
  import { type RunnerOptions } from "./runner.js";
6
6
  export interface EvalOptions {
@@ -1 +1 @@
1
- {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/eval/index.ts"],"names":[],"mappings":"AACA,OAAO,EAAW,KAAK,MAAM,EAAE,MAAM,cAAc,CAAC;AACpD,OAAO,EAA+B,KAAK,aAAa,EAAE,KAAK,UAAU,EAAE,MAAM,YAAY,CAAC;AAE9F,OAAO,EAAwB,KAAK,KAAK,EAAE,MAAM,YAAY,CAAC;AAC9D,OAAO,EAAe,KAAK,UAAU,EAAE,MAAM,aAAa,CAAC;AAC3D,OAAO,EAAY,KAAK,aAAa,EAAE,MAAM,aAAa,CAAC;AAI3D,MAAM,WAAW,WAAW;IAC1B,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,GAAG,EAAE,MAAM,CAAC;IACrB,QAAQ,CAAC,KAAK,CAAC,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,MAAM,CAAC,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,IAAI,CAAC,EAAE,UAAU,CAAC,MAAM,CAAC,CAAC;IACnC,QAAQ,CAAC,IAAI,CAAC,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,WAAW,CAAC,EAAE,MAAM,CAAC;IAC9B,QAAQ,CAAC,KAAK,CAAC,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,UAAU,CAAC,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,gBAAgB,CAAC,EAAE,MAAM,CAAC;IACnC,QAAQ,CAAC,KAAK,CAAC,EAAE,KAAK,CAAC;IACvB,QAAQ,CAAC,KAAK,CAAC,EAAE,aAAa,CAAC,OAAO,CAAC,CAAC;CACzC;AAED,wBAAsB,OAAO,CAAC,OAAO,EAAE,WAAW,GAAG,OAAO,CAAC,MAAM,CAAC,UAAU,EAAE,aAAa,EAAE,CAAC,CAAC,CA6BhG;AAkCD,YAAY,EAAE,UAAU,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAC1D,OAAO,EAAE,aAAa,EAAE,MAAM,EAAE,MAAM,aAAa,CAAC;AACpD,YAAY,EAAE,UAAU,EAAE,aAAa,EAAE,MAAM,YAAY,CAAC"}
1
+ {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/eval/index.ts"],"names":[],"mappings":"AACA,OAAO,EAAW,KAAK,MAAM,EAAE,MAAM,cAAc,CAAC;AACpD,OAAO,EAA+B,KAAK,aAAa,EAAE,KAAK,UAAU,EAAE,MAAM,YAAY,CAAC;AAE9F,OAAO,EAAoC,KAAK,KAAK,EAAE,MAAM,kBAAkB,CAAC;AAChF,OAAO,EAAe,KAAK,UAAU,EAAE,MAAM,aAAa,CAAC;AAC3D,OAAO,EAAY,KAAK,aAAa,EAAE,MAAM,aAAa,CAAC;AAE3D,MAAM,WAAW,WAAW;IAC1B,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,GAAG,EAAE,MAAM,CAAC;IACrB,QAAQ,CAAC,KAAK,CAAC,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,MAAM,CAAC,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,IAAI,CAAC,EAAE,UAAU,CAAC,MAAM,CAAC,CAAC;IACnC,QAAQ,CAAC,IAAI,CAAC,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,WAAW,CAAC,EAAE,MAAM,CAAC;IAC9B,QAAQ,CAAC,KAAK,CAAC,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,UAAU,CAAC,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,gBAAgB,CAAC,EAAE,MAAM,CAAC;IACnC,QAAQ,CAAC,KAAK,CAAC,EAAE,KAAK,CAAC;IACvB,QAAQ,CAAC,KAAK,CAAC,EAAE,aAAa,CAAC,OAAO,CAAC,CAAC;CACzC;AAED,wBAAsB,OAAO,CAAC,OAAO,EAAE,WAAW,GAAG,OAAO,CAAC,MAAM,CAAC,UAAU,EAAE,aAAa,EAAE,CAAC,CAAC,CA4BhG;AAuBD,YAAY,EAAE,UAAU,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAC1D,OAAO,EAAE,aAAa,EAAE,MAAM,EAAE,MAAM,aAAa,CAAC;AACpD,YAAY,EAAE,UAAU,EAAE,aAAa,EAAE,MAAM,YAAY,CAAC"}
@@ -1,11 +1,10 @@
1
1
  import { defaultSources, discoverInstalled, indexInstalled } from "../installed.js";
2
2
  import { err, ok } from "../result.js";
3
3
  import { loadCases, unresolvedSkills } from "./cases.js";
4
- import { gradeResults } from "./grade.js";
5
- import { createAnthropicJudge } from "./judge.js";
4
+ import { gradeResults } from "./grade/index.js";
5
+ import { DEFAULT_JUDGE_MODEL, selectJudge } from "./judge/index.js";
6
6
  import { buildReport } from "./report.js";
7
7
  import { runCases } from "./runner.js";
8
- const DEFAULT_JUDGE_MODEL = "claude-sonnet-4-5";
9
8
  export async function runEval(options) {
10
9
  const loaded = await loadCases(options.casesDir);
11
10
  if (!loaded.ok)
@@ -20,8 +19,6 @@ export async function runEval(options) {
20
19
  if (unresolved.length > 0)
21
20
  return err(unresolved);
22
21
  const judge = resolveJudge(selected, options);
23
- if (!judge.ok)
24
- return err(judge.error);
25
22
  const runnerOptions = {
26
23
  cwd: options.cwd,
27
24
  ...(options.runs !== undefined && { runs: options.runs }),
@@ -31,27 +28,19 @@ export async function runEval(options) {
31
28
  ...(options.onRun !== undefined && { onRun: options.onRun }),
32
29
  };
33
30
  const results = await runCases(selected, runnerOptions);
34
- const reports = await gradeResults(results, judge.value);
31
+ const reports = await gradeResults(results, judge);
35
32
  return ok(buildReport(reports));
36
33
  }
37
34
  function needsJudge(cases) {
38
35
  return cases.some((c) => c.tier === "solving" && c.rubric !== undefined);
39
36
  }
40
37
  function resolveJudge(cases, options) {
41
- if (options.judge)
42
- return ok(options.judge);
43
38
  if (!needsJudge(cases))
44
- return ok(undefined);
45
- const apiKey = process.env["ANTHROPIC_API_KEY"];
46
- if (!apiKey) {
47
- return err([
48
- {
49
- file: options.casesDir,
50
- message: "solving rubric grading requires ANTHROPIC_API_KEY for the judge",
51
- },
52
- ]);
53
- }
54
- return ok(createAnthropicJudge({ model: options.judgeModel ?? DEFAULT_JUDGE_MODEL, apiKey }));
39
+ return undefined;
40
+ return selectJudge({
41
+ model: options.judgeModel ?? DEFAULT_JUDGE_MODEL,
42
+ ...(options.judge !== undefined && { override: options.judge }),
43
+ });
55
44
  }
56
45
  function select(cases, options) {
57
46
  return cases.filter((evalCase) => {
@@ -1 +1 @@
1
- {"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/eval/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,iBAAiB,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAC;AACpF,OAAO,EAAE,GAAG,EAAE,EAAE,EAAe,MAAM,cAAc,CAAC;AACpD,OAAO,EAAE,SAAS,EAAE,gBAAgB,EAAuC,MAAM,YAAY,CAAC;AAC9F,OAAO,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAC1C,OAAO,EAAE,oBAAoB,EAAc,MAAM,YAAY,CAAC;AAC9D,OAAO,EAAE,WAAW,EAAmB,MAAM,aAAa,CAAC;AAC3D,OAAO,EAAE,QAAQ,EAAsB,MAAM,aAAa,CAAC;AAE3D,MAAM,mBAAmB,GAAG,mBAAmB,CAAC;AAiBhD,MAAM,CAAC,KAAK,UAAU,OAAO,CAAC,OAAoB;IAChD,MAAM,MAAM,GAAG,MAAM,SAAS,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;IACjD,IAAI,CAAC,MAAM,CAAC,EAAE;QAAE,OAAO,GAAG,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IAEzC,MAAM,QAAQ,GAAG,MAAM,CAAC,MAAM,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC;IAC/C,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC1B,OAAO,GAAG,CAAC,CAAC,EAAE,IAAI,EAAE,OAAO,CAAC,QAAQ,EAAE,OAAO,EAAE,oCAAoC,EAAE,CAAC,CAAC,CAAC;IAC1F,CAAC;IAED,MAAM,SAAS,GAAG,cAAc,CAAC,MAAM,iBAAiB,CAAC,cAAc,EAAE,CAAC,CAAC,CAAC;IAC5E,MAAM,YAAY,GAAG,IAAI,GAAG,CAAC,SAAS,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC;IACtD,MAAM,UAAU,GAAG,gBAAgB,CAAC,QAAQ,EAAE,YAAY,CAAC,CAAC;IAC5D,IAAI,UAAU,CAAC,MAAM,GAAG,CAAC;QAAE,OAAO,GAAG,CAAC,UAAU,CAAC,CAAC;IAElD,MAAM,KAAK,GAAG,YAAY,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;IAC9C,IAAI,CAAC,KAAK,CAAC,EAAE;QAAE,OAAO,GAAG,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;IAEvC,MAAM,aAAa,GAAkB;QACnC,GAAG,EAAE,OAAO,CAAC,GAAG;QAChB,GAAG,CAAC,OAAO,CAAC,IAAI,KAAK,SAAS,IAAI,EAAE,IAAI,EAAE,OAAO,CAAC,IAAI,EAAE,CAAC;QACzD,GAAG,CAAC,OAAO,CAAC,WAAW,KAAK,SAAS,IAAI,EAAE,WAAW,EAAE,OAAO,CAAC,WAAW,EAAE,CAAC;QAC9E,GAAG,CAAC,OAAO,CAAC,KAAK,KAAK,SAAS,IAAI,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,EAAE,CAAC;QAC5D,GAAG,CAAC,OAAO,CAAC,gBAAgB,KAAK,SAAS,IAAI,EAAE,gBAAgB,EAAE,OAAO,CAAC,gBAAgB,EAAE,CAAC;QAC7F,GAAG,CAAC,OAAO,CAAC,KAAK,KAAK,SAAS,IAAI,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,EAAE,CAAC;KAC7D,CAAC;IACF,MAAM,OAAO,GAAG,MAAM,QAAQ,CAAC,QAAQ,EAAE,aAAa,CAAC,CAAC;IACxD,MAAM,OAAO,GAAG,MAAM,YAAY,CAAC,OAAO,EAAE,KAAK,CAAC,KAAK,CAAC,CAAC;IAEzD,OAAO,EAAE,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC;AAClC,CAAC;AAED,SAAS,UAAU,CAAC,KAA4B;IAC9C,OAAO,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,SAAS,IAAI,CAAC,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC;AAC3E,CAAC;AAED,SAAS,YAAY,CACnB,KAA4B,EAC5B,OAAoB;IAEpB,IAAI,OAAO,CAAC,KAAK;QAAE,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;IAC5C,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC,SAAS,CAAC,CAAC;IAE7C,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,mBAAmB,CAAC,CAAC;IAChD,IAAI,CAAC,MAAM,EAAE,CAAC;QACZ,OAAO,GAAG,CAAC;YACT;gBACE,IAAI,EAAE,OAAO,CAAC,QAAQ;gBACtB,OAAO,EAAE,iEAAiE;aAC3E;SACF,CAAC,CAAC;IACL,CAAC;IACD,OAAO,EAAE,CAAC,oBAAoB,CAAC,EAAE,KAAK,EAAE,OAAO,CAAC,UAAU,IAAI,mBAAmB,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC;AAChG,CAAC;AAED,SAAS,MAAM,CAAC,KAA4B,EAAE,OAAoB;IAChE,OAAO,KAAK,CAAC,MAAM,CAAC,CAAC,QAAQ,EAAE,EAAE;QAC/B,IAAI,OAAO,CAAC,MAAM,IAAI,QAAQ,CAAC,EAAE,KAAK,OAAO,CAAC,MAAM;YAAE,OAAO,KAAK,CAAC;QACnE,IAAI,OAAO,CAAC,KAAK,IAAI,QAAQ,CAAC,KAAK,KAAK,OAAO,CAAC,KAAK;YAAE,OAAO,KAAK,CAAC;QACpE,IAAI,OAAO,CAAC,IAAI,IAAI,QAAQ,CAAC,IAAI,KAAK,OAAO,CAAC,IAAI;YAAE,OAAO,KAAK,CAAC;QACjE,OAAO,IAAI,CAAC;IACd,CAAC,CAAC,CAAC;AACL,CAAC;AAGD,OAAO,EAAE,aAAa,EAAE,MAAM,EAAE,MAAM,aAAa,CAAC"}
1
+ {"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/eval/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,iBAAiB,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAC;AACpF,OAAO,EAAE,GAAG,EAAE,EAAE,EAAe,MAAM,cAAc,CAAC;AACpD,OAAO,EAAE,SAAS,EAAE,gBAAgB,EAAuC,MAAM,YAAY,CAAC;AAC9F,OAAO,EAAE,YAAY,EAAE,MAAM,kBAAkB,CAAC;AAChD,OAAO,EAAE,mBAAmB,EAAE,WAAW,EAAc,MAAM,kBAAkB,CAAC;AAChF,OAAO,EAAE,WAAW,EAAmB,MAAM,aAAa,CAAC;AAC3D,OAAO,EAAE,QAAQ,EAAsB,MAAM,aAAa,CAAC;AAiB3D,MAAM,CAAC,KAAK,UAAU,OAAO,CAAC,OAAoB;IAChD,MAAM,MAAM,GAAG,MAAM,SAAS,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;IACjD,IAAI,CAAC,MAAM,CAAC,EAAE;QAAE,OAAO,GAAG,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IAEzC,MAAM,QAAQ,GAAG,MAAM,CAAC,MAAM,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC;IAC/C,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC1B,OAAO,GAAG,CAAC,CAAC,EAAE,IAAI,EAAE,OAAO,CAAC,QAAQ,EAAE,OAAO,EAAE,oCAAoC,EAAE,CAAC,CAAC,CAAC;IAC1F,CAAC;IAED,MAAM,SAAS,GAAG,cAAc,CAAC,MAAM,iBAAiB,CAAC,cAAc,EAAE,CAAC,CAAC,CAAC;IAC5E,MAAM,YAAY,GAAG,IAAI,GAAG,CAAC,SAAS,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC;IACtD,MAAM,UAAU,GAAG,gBAAgB,CAAC,QAAQ,EAAE,YAAY,CAAC,CAAC;IAC5D,IAAI,UAAU,CAAC,MAAM,GAAG,CAAC;QAAE,OAAO,GAAG,CAAC,UAAU,CAAC,CAAC;IAElD,MAAM,KAAK,GAAG,YAAY,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;IAE9C,MAAM,aAAa,GAAkB;QACnC,GAAG,EAAE,OAAO,CAAC,GAAG;QAChB,GAAG,CAAC,OAAO,CAAC,IAAI,KAAK,SAAS,IAAI,EAAE,IAAI,EAAE,OAAO,CAAC,IAAI,EAAE,CAAC;QACzD,GAAG,CAAC,OAAO,CAAC,WAAW,KAAK,SAAS,IAAI,EAAE,WAAW,EAAE,OAAO,CAAC,WAAW,EAAE,CAAC;QAC9E,GAAG,CAAC,OAAO,CAAC,KAAK,KAAK,SAAS,IAAI,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,EAAE,CAAC;QAC5D,GAAG,CAAC,OAAO,CAAC,gBAAgB,KAAK,SAAS,IAAI,EAAE,gBAAgB,EAAE,OAAO,CAAC,gBAAgB,EAAE,CAAC;QAC7F,GAAG,CAAC,OAAO,CAAC,KAAK,KAAK,SAAS,IAAI,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,EAAE,CAAC;KAC7D,CAAC;IACF,MAAM,OAAO,GAAG,MAAM,QAAQ,CAAC,QAAQ,EAAE,aAAa,CAAC,CAAC;IACxD,MAAM,OAAO,GAAG,MAAM,YAAY,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC;IAEnD,OAAO,EAAE,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC;AAClC,CAAC;AAED,SAAS,UAAU,CAAC,KAA4B;IAC9C,OAAO,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,SAAS,IAAI,CAAC,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC;AAC3E,CAAC;AAED,SAAS,YAAY,CAAC,KAA4B,EAAE,OAAoB;IACtE,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC;QAAE,OAAO,SAAS,CAAC;IACzC,OAAO,WAAW,CAAC;QACjB,KAAK,EAAE,OAAO,CAAC,UAAU,IAAI,mBAAmB;QAChD,GAAG,CAAC,OAAO,CAAC,KAAK,KAAK,SAAS,IAAI,EAAE,QAAQ,EAAE,OAAO,CAAC,KAAK,EAAE,CAAC;KAChE,CAAC,CAAC;AACL,CAAC;AAED,SAAS,MAAM,CAAC,KAA4B,EAAE,OAAoB;IAChE,OAAO,KAAK,CAAC,MAAM,CAAC,CAAC,QAAQ,EAAE,EAAE;QAC/B,IAAI,OAAO,CAAC,MAAM,IAAI,QAAQ,CAAC,EAAE,KAAK,OAAO,CAAC,MAAM;YAAE,OAAO,KAAK,CAAC;QACnE,IAAI,OAAO,CAAC,KAAK,IAAI,QAAQ,CAAC,KAAK,KAAK,OAAO,CAAC,KAAK;YAAE,OAAO,KAAK,CAAC;QACpE,IAAI,OAAO,CAAC,IAAI,IAAI,QAAQ,CAAC,IAAI,KAAK,OAAO,CAAC,IAAI;YAAE,OAAO,KAAK,CAAC;QACjE,OAAO,IAAI,CAAC;IACd,CAAC,CAAC,CAAC;AACL,CAAC;AAGD,OAAO,EAAE,aAAa,EAAE,MAAM,EAAE,MAAM,aAAa,CAAC"}
@@ -0,0 +1,9 @@
1
+ import type { SubagentRunner } from "./subagent.js";
2
+ export declare function judgeArgs(prompt: string, model: string, schema: string): string[];
3
+ export interface ClaudeCliOptions {
4
+ readonly bin?: string;
5
+ readonly schema: string;
6
+ readonly timeoutMs?: number;
7
+ }
8
+ export declare function createClaudeCliRunner(options: ClaudeCliOptions): SubagentRunner;
9
+ //# sourceMappingURL=claude-cli.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"claude-cli.d.ts","sourceRoot":"","sources":["../../../src/eval/judge/claude-cli.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,eAAe,CAAC;AAEpD,wBAAgB,SAAS,CAAC,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,GAAG,MAAM,EAAE,CAEjF;AAED,MAAM,WAAW,gBAAgB;IAC/B,QAAQ,CAAC,GAAG,CAAC,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,SAAS,CAAC,EAAE,MAAM,CAAC;CAC7B;AAID,wBAAgB,qBAAqB,CAAC,OAAO,EAAE,gBAAgB,GAAG,cAAc,CAK/E"}
@@ -0,0 +1,38 @@
1
+ import { spawn } from "node:child_process";
2
+ import { scrubEnv, SUBPROCESS_ENV_OVERRIDES } from "../claude-env.js";
3
+ export function judgeArgs(prompt, model, schema) {
4
+ return ["-p", prompt, "--output-format", "json", "--model", model, "--json-schema", schema];
5
+ }
6
+ const DEFAULT_TIMEOUT_MS = 120_000;
7
+ export function createClaudeCliRunner(options) {
8
+ return async ({ model, prompt }) => captureStdout(options.bin ?? "claude", judgeArgs(prompt, model, options.schema), {
9
+ timeoutMs: options.timeoutMs ?? DEFAULT_TIMEOUT_MS,
10
+ });
11
+ }
12
+ async function captureStdout(bin, args, opts) {
13
+ return new Promise((resolve, reject) => {
14
+ const child = spawn(bin, args, {
15
+ env: scrubEnv(process.env, SUBPROCESS_ENV_OVERRIDES),
16
+ stdio: ["ignore", "pipe", "ignore"],
17
+ });
18
+ const chunks = [];
19
+ child.stdout.on("data", (chunk) => chunks.push(chunk));
20
+ const timedOut = { reached: false };
21
+ const timer = setTimeout(() => {
22
+ timedOut.reached = true;
23
+ child.kill("SIGKILL");
24
+ }, opts.timeoutMs);
25
+ child.on("error", (cause) => {
26
+ clearTimeout(timer);
27
+ reject(new Error(`failed to spawn claude: ${cause.message}`));
28
+ });
29
+ child.on("close", () => {
30
+ clearTimeout(timer);
31
+ if (timedOut.reached)
32
+ reject(new Error(`claude judge timed out after ${opts.timeoutMs}ms`));
33
+ else
34
+ resolve(Buffer.concat(chunks).toString("utf8"));
35
+ });
36
+ });
37
+ }
38
+ //# sourceMappingURL=claude-cli.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"claude-cli.js","sourceRoot":"","sources":["../../../src/eval/judge/claude-cli.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,EAAE,MAAM,oBAAoB,CAAC;AAE3C,OAAO,EAAE,QAAQ,EAAE,wBAAwB,EAAE,MAAM,kBAAkB,CAAC;AAGtE,MAAM,UAAU,SAAS,CAAC,MAAc,EAAE,KAAa,EAAE,MAAc;IACrE,OAAO,CAAC,IAAI,EAAE,MAAM,EAAE,iBAAiB,EAAE,MAAM,EAAE,SAAS,EAAE,KAAK,EAAE,eAAe,EAAE,MAAM,CAAC,CAAC;AAC9F,CAAC;AAQD,MAAM,kBAAkB,GAAG,OAAO,CAAC;AAEnC,MAAM,UAAU,qBAAqB,CAAC,OAAyB;IAC7D,OAAO,KAAK,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,EAAE,EAAE,CACjC,aAAa,CAAC,OAAO,CAAC,GAAG,IAAI,QAAQ,EAAE,SAAS,CAAC,MAAM,EAAE,KAAK,EAAE,OAAO,CAAC,MAAM,CAAC,EAAE;QAC/E,SAAS,EAAE,OAAO,CAAC,SAAS,IAAI,kBAAkB;KACnD,CAAC,CAAC;AACP,CAAC;AAMD,KAAK,UAAU,aAAa,CAAC,GAAW,EAAE,IAAc,EAAE,IAAkB;IAC1E,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;QACrC,MAAM,KAAK,GAAG,KAAK,CAAC,GAAG,EAAE,IAAI,EAAE;YAC7B,GAAG,EAAE,QAAQ,CAAC,OAAO,CAAC,GAAG,EAAE,wBAAwB,CAAC;YACpD,KAAK,EAAE,CAAC,QAAQ,EAAE,MAAM,EAAE,QAAQ,CAAC;SACpC,CAAC,CAAC;QAEH,MAAM,MAAM,GAAa,EAAE,CAAC;QAC5B,KAAK,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,KAAa,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC;QAE/D,MAAM,QAAQ,GAAG,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;QACpC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE;YAC5B,QAAQ,CAAC,OAAO,GAAG,IAAI,CAAC;YACxB,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QACxB,CAAC,EAAE,IAAI,CAAC,SAAS,CAAC,CAAC;QACnB,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,KAAK,EAAE,EAAE;YAC1B,YAAY,CAAC,KAAK,CAAC,CAAC;YACpB,MAAM,CAAC,IAAI,KAAK,CAAC,2BAA2B,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;QAChE,CAAC,CAAC,CAAC;QACH,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,GAAG,EAAE;YACrB,YAAY,CAAC,KAAK,CAAC,CAAC;YACpB,IAAI,QAAQ,CAAC,OAAO;gBAAE,MAAM,CAAC,IAAI,KAAK,CAAC,gCAAgC,IAAI,CAAC,SAAS,IAAI,CAAC,CAAC,CAAC;;gBACvF,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC;QACvD,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;AACL,CAAC"}
@@ -0,0 +1,5 @@
1
+ export { Verdict, parseVerdict, type Judge, type JudgeRequest } from "./verdict.js";
2
+ export { createSubagentJudge, VERDICT_SCHEMA, type SubagentRunner, type SubagentJudgeOptions, } from "./subagent.js";
3
+ export { createClaudeCliRunner, type ClaudeCliOptions } from "./claude-cli.js";
4
+ export { selectJudge, DEFAULT_JUDGE_MODEL, type SelectJudgeOptions } from "./select.js";
5
+ //# sourceMappingURL=index.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/eval/judge/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,YAAY,EAAE,KAAK,KAAK,EAAE,KAAK,YAAY,EAAE,MAAM,cAAc,CAAC;AACpF,OAAO,EACL,mBAAmB,EACnB,cAAc,EACd,KAAK,cAAc,EACnB,KAAK,oBAAoB,GAC1B,MAAM,eAAe,CAAC;AACvB,OAAO,EAAE,qBAAqB,EAAE,KAAK,gBAAgB,EAAE,MAAM,iBAAiB,CAAC;AAC/E,OAAO,EAAE,WAAW,EAAE,mBAAmB,EAAE,KAAK,kBAAkB,EAAE,MAAM,aAAa,CAAC"}
@@ -0,0 +1,5 @@
1
+ export { Verdict, parseVerdict } from "./verdict.js";
2
+ export { createSubagentJudge, VERDICT_SCHEMA, } from "./subagent.js";
3
+ export { createClaudeCliRunner } from "./claude-cli.js";
4
+ export { selectJudge, DEFAULT_JUDGE_MODEL } from "./select.js";
5
+ //# sourceMappingURL=index.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/eval/judge/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,YAAY,EAAiC,MAAM,cAAc,CAAC;AACpF,OAAO,EACL,mBAAmB,EACnB,cAAc,GAGf,MAAM,eAAe,CAAC;AACvB,OAAO,EAAE,qBAAqB,EAAyB,MAAM,iBAAiB,CAAC;AAC/E,OAAO,EAAE,WAAW,EAAE,mBAAmB,EAA2B,MAAM,aAAa,CAAC"}
@@ -0,0 +1,9 @@
1
+ import type { Judge } from "./verdict.js";
2
+ export declare const DEFAULT_JUDGE_MODEL = "claude-haiku-4-5";
3
+ export interface SelectJudgeOptions {
4
+ readonly model: string;
5
+ readonly override?: Judge;
6
+ readonly claudeBin?: string;
7
+ }
8
+ export declare function selectJudge(options: SelectJudgeOptions): Judge;
9
+ //# sourceMappingURL=select.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"select.d.ts","sourceRoot":"","sources":["../../../src/eval/judge/select.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,KAAK,EAAE,MAAM,cAAc,CAAC;AAE1C,eAAO,MAAM,mBAAmB,qBAAqB,CAAC;AAEtD,MAAM,WAAW,kBAAkB;IACjC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,QAAQ,CAAC,EAAE,KAAK,CAAC;IAC1B,QAAQ,CAAC,SAAS,CAAC,EAAE,MAAM,CAAC;CAC7B;AAED,wBAAgB,WAAW,CAAC,OAAO,EAAE,kBAAkB,GAAG,KAAK,CAS9D"}
@@ -0,0 +1,15 @@
1
+ import { createClaudeCliRunner } from "./claude-cli.js";
2
+ import { createSubagentJudge, VERDICT_SCHEMA } from "./subagent.js";
3
+ export const DEFAULT_JUDGE_MODEL = "claude-haiku-4-5";
4
+ export function selectJudge(options) {
5
+ if (options.override)
6
+ return options.override;
7
+ return createSubagentJudge({
8
+ model: options.model,
9
+ run: createClaudeCliRunner({
10
+ schema: VERDICT_SCHEMA,
11
+ ...(options.claudeBin !== undefined && { bin: options.claudeBin }),
12
+ }),
13
+ });
14
+ }
15
+ //# sourceMappingURL=select.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"select.js","sourceRoot":"","sources":["../../../src/eval/judge/select.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,qBAAqB,EAAE,MAAM,iBAAiB,CAAC;AACxD,OAAO,EAAE,mBAAmB,EAAE,cAAc,EAAE,MAAM,eAAe,CAAC;AAGpE,MAAM,CAAC,MAAM,mBAAmB,GAAG,kBAAkB,CAAC;AAQtD,MAAM,UAAU,WAAW,CAAC,OAA2B;IACrD,IAAI,OAAO,CAAC,QAAQ;QAAE,OAAO,OAAO,CAAC,QAAQ,CAAC;IAC9C,OAAO,mBAAmB,CAAC;QACzB,KAAK,EAAE,OAAO,CAAC,KAAK;QACpB,GAAG,EAAE,qBAAqB,CAAC;YACzB,MAAM,EAAE,cAAc;YACtB,GAAG,CAAC,OAAO,CAAC,SAAS,KAAK,SAAS,IAAI,EAAE,GAAG,EAAE,OAAO,CAAC,SAAS,EAAE,CAAC;SACnE,CAAC;KACH,CAAC,CAAC;AACL,CAAC"}
@@ -0,0 +1,13 @@
1
+ import { type Judge } from "./verdict.js";
2
+ export declare const VERDICT_SCHEMA: string;
3
+ export interface SubagentRequest {
4
+ readonly model: string;
5
+ readonly prompt: string;
6
+ }
7
+ export type SubagentRunner = (request: SubagentRequest) => Promise<string>;
8
+ export interface SubagentJudgeOptions {
9
+ readonly model: string;
10
+ readonly run: SubagentRunner;
11
+ }
12
+ export declare function createSubagentJudge(options: SubagentJudgeOptions): Judge;
13
+ //# sourceMappingURL=subagent.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"subagent.d.ts","sourceRoot":"","sources":["../../../src/eval/judge/subagent.ts"],"names":[],"mappings":"AAEA,OAAO,EAAyB,KAAK,KAAK,EAAgB,MAAM,cAAc,CAAC;AAa/E,eAAO,MAAM,cAAc,QAQzB,CAAC;AAEH,MAAM,WAAW,eAAe;IAC9B,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;CACzB;AAED,MAAM,MAAM,cAAc,GAAG,CAAC,OAAO,EAAE,eAAe,KAAK,OAAO,CAAC,MAAM,CAAC,CAAC;AAE3E,MAAM,WAAW,oBAAoB;IACnC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,GAAG,EAAE,cAAc,CAAC;CAC9B;AAED,wBAAgB,mBAAmB,CAAC,OAAO,EAAE,oBAAoB,GAAG,KAAK,CAYxE"}
@@ -0,0 +1,65 @@
1
+ import { z } from "zod";
2
+ import { abstain, parseVerdict } from "./verdict.js";
3
+ const StructuredOutput = z.object({
4
+ verdict: z.enum(["pass", "fail", "unknown"]),
5
+ evidence: z.string(),
6
+ });
7
+ const CliResult = z.object({
8
+ is_error: z.boolean(),
9
+ result: z.string().optional(),
10
+ structured_output: z.unknown().optional(),
11
+ });
12
+ export const VERDICT_SCHEMA = JSON.stringify({
13
+ type: "object",
14
+ properties: {
15
+ verdict: { enum: ["pass", "fail", "unknown"] },
16
+ evidence: { type: "string" },
17
+ },
18
+ required: ["verdict", "evidence"],
19
+ additionalProperties: false,
20
+ });
21
+ export function createSubagentJudge(options) {
22
+ return async ({ dimension, criterion, output }) => {
23
+ try {
24
+ const stdout = await options.run({
25
+ model: options.model,
26
+ prompt: buildPrompt(dimension, criterion, output),
27
+ });
28
+ return verdictFrom(stdout);
29
+ }
30
+ catch (cause) {
31
+ return abstain(cause instanceof Error ? cause.message : "subagent judge failed");
32
+ }
33
+ };
34
+ }
35
+ function buildPrompt(dimension, criterion, output) {
36
+ return ("You are a strict grader for AI agent outputs. Judge the output against the single " +
37
+ "criterion in isolation. Set verdict to pass only if the criterion is clearly met, fail " +
38
+ "if clearly unmet, and unknown if you genuinely cannot tell. Always cite concrete evidence " +
39
+ `from the output.\n\nDimension: ${dimension}\nCriterion: ${criterion}\n\nOutput to grade:\n${output}`);
40
+ }
41
+ function verdictFrom(stdout) {
42
+ const cli = CliResult.safeParse(safeJson(stdout));
43
+ if (!cli.success)
44
+ return parseVerdict(null);
45
+ if (cli.data.is_error)
46
+ return abstain(cli.data.result ?? "subagent judge failed");
47
+ const structured = StructuredOutput.safeParse(cli.data.structured_output);
48
+ if (!structured.success)
49
+ return parseVerdict(null);
50
+ return toDomain(structured.data);
51
+ }
52
+ function safeJson(raw) {
53
+ try {
54
+ return JSON.parse(raw);
55
+ }
56
+ catch {
57
+ return null;
58
+ }
59
+ }
60
+ function toDomain({ verdict, evidence }) {
61
+ if (verdict === "unknown")
62
+ return { pass: "unknown", evidence };
63
+ return { pass: verdict === "pass", evidence };
64
+ }
65
+ //# sourceMappingURL=subagent.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"subagent.js","sourceRoot":"","sources":["../../../src/eval/judge/subagent.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,OAAO,EAAE,OAAO,EAAE,YAAY,EAA4B,MAAM,cAAc,CAAC;AAE/E,MAAM,gBAAgB,GAAG,CAAC,CAAC,MAAM,CAAC;IAChC,OAAO,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,MAAM,EAAE,SAAS,CAAC,CAAC;IAC5C,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE;CACrB,CAAC,CAAC;AAEH,MAAM,SAAS,GAAG,CAAC,CAAC,MAAM,CAAC;IACzB,QAAQ,EAAE,CAAC,CAAC,OAAO,EAAE;IACrB,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC7B,iBAAiB,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,EAAE;CAC1C,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,cAAc,GAAG,IAAI,CAAC,SAAS,CAAC;IAC3C,IAAI,EAAE,QAAQ;IACd,UAAU,EAAE;QACV,OAAO,EAAE,EAAE,IAAI,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,SAAS,CAAC,EAAE;QAC9C,QAAQ,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;KAC7B;IACD,QAAQ,EAAE,CAAC,SAAS,EAAE,UAAU,CAAC;IACjC,oBAAoB,EAAE,KAAK;CAC5B,CAAC,CAAC;AAcH,MAAM,UAAU,mBAAmB,CAAC,OAA6B;IAC/D,OAAO,KAAK,EAAE,EAAE,SAAS,EAAE,SAAS,EAAE,MAAM,EAAE,EAAE,EAAE;QAChD,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;gBAC/B,KAAK,EAAE,OAAO,CAAC,KAAK;gBACpB,MAAM,EAAE,WAAW,CAAC,SAAS,EAAE,SAAS,EAAE,MAAM,CAAC;aAClD,CAAC,CAAC;YACH,OAAO,WAAW,CAAC,MAAM,CAAC,CAAC;QAC7B,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,OAAO,OAAO,CAAC,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,uBAAuB,CAAC,CAAC;QACnF,CAAC;IACH,CAAC,CAAC;AACJ,CAAC;AAED,SAAS,WAAW,CAAC,SAAiB,EAAE,SAAiB,EAAE,MAAc;IACvE,OAAO,CACL,oFAAoF;QACpF,yFAAyF;QACzF,4FAA4F;QAC5F,kCAAkC,SAAS,gBAAgB,SAAS,yBAAyB,MAAM,EAAE,CACtG,CAAC;AACJ,CAAC;AAED,SAAS,WAAW,CAAC,MAAc;IACjC,MAAM,GAAG,GAAG,SAAS,CAAC,SAAS,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC;IAClD,IAAI,CAAC,GAAG,CAAC,OAAO;QAAE,OAAO,YAAY,CAAC,IAAI,CAAC,CAAC;IAC5C,IAAI,GAAG,CAAC,IAAI,CAAC,QAAQ;QAAE,OAAO,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,IAAI,uBAAuB,CAAC,CAAC;IAElF,MAAM,UAAU,GAAG,gBAAgB,CAAC,SAAS,CAAC,GAAG,CAAC,IAAI,CAAC,iBAAiB,CAAC,CAAC;IAC1E,IAAI,CAAC,UAAU,CAAC,OAAO;QAAE,OAAO,YAAY,CAAC,IAAI,CAAC,CAAC;IACnD,OAAO,QAAQ,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC;AACnC,CAAC;AAED,SAAS,QAAQ,CAAC,GAAW;IAC3B,IAAI,CAAC;QACH,OAAO,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;IACzB,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC;AAED,SAAS,QAAQ,CAAC,EAAE,OAAO,EAAE,QAAQ,EAAoC;IACvE,IAAI,OAAO,KAAK,SAAS;QAAE,OAAO,EAAE,IAAI,EAAE,SAAS,EAAE,QAAQ,EAAE,CAAC;IAChE,OAAO,EAAE,IAAI,EAAE,OAAO,KAAK,MAAM,EAAE,QAAQ,EAAE,CAAC;AAChD,CAAC"}
@@ -16,11 +16,6 @@ export interface JudgeRequest {
16
16
  readonly output: string;
17
17
  }
18
18
  export type Judge = (request: JudgeRequest) => Promise<Verdict>;
19
+ export declare function abstain(evidence: string): Verdict;
19
20
  export declare function parseVerdict(raw: unknown): Verdict;
20
- export interface AnthropicJudgeOptions {
21
- readonly model: string;
22
- readonly apiKey: string;
23
- readonly maxTokens?: number;
24
- }
25
- export declare function createAnthropicJudge(options: AnthropicJudgeOptions): Judge;
26
- //# sourceMappingURL=judge.d.ts.map
21
+ //# sourceMappingURL=verdict.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"verdict.d.ts","sourceRoot":"","sources":["../../../src/eval/judge/verdict.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,eAAO,MAAM,OAAO;;;;;;;;;EAGlB,CAAC;AAEH,MAAM,MAAM,OAAO,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,OAAO,CAAC,CAAC;AAE9C,MAAM,WAAW,YAAY;IAC3B,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;CACzB;AAED,MAAM,MAAM,KAAK,GAAG,CAAC,OAAO,EAAE,YAAY,KAAK,OAAO,CAAC,OAAO,CAAC,CAAC;AAEhE,wBAAgB,OAAO,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAEjD;AAED,wBAAgB,YAAY,CAAC,GAAG,EAAE,OAAO,GAAG,OAAO,CAIlD"}
@@ -0,0 +1,15 @@
1
+ import { z } from "zod";
2
+ export const Verdict = z.object({
3
+ pass: z.union([z.boolean(), z.literal("unknown")]),
4
+ evidence: z.string(),
5
+ });
6
+ export function abstain(evidence) {
7
+ return { pass: "unknown", evidence };
8
+ }
9
+ export function parseVerdict(raw) {
10
+ if (raw === undefined || raw === null)
11
+ return abstain("judge returned no verdict");
12
+ const result = Verdict.safeParse(raw);
13
+ return result.success ? result.data : abstain("judge returned an unparseable verdict");
14
+ }
15
+ //# sourceMappingURL=verdict.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"verdict.js","sourceRoot":"","sources":["../../../src/eval/judge/verdict.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,MAAM,CAAC,MAAM,OAAO,GAAG,CAAC,CAAC,MAAM,CAAC;IAC9B,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC;IAClD,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE;CACrB,CAAC,CAAC;AAYH,MAAM,UAAU,OAAO,CAAC,QAAgB;IACtC,OAAO,EAAE,IAAI,EAAE,SAAS,EAAE,QAAQ,EAAE,CAAC;AACvC,CAAC;AAED,MAAM,UAAU,YAAY,CAAC,GAAY;IACvC,IAAI,GAAG,KAAK,SAAS,IAAI,GAAG,KAAK,IAAI;QAAE,OAAO,OAAO,CAAC,2BAA2B,CAAC,CAAC;IACnF,MAAM,MAAM,GAAG,OAAO,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC;IACtC,OAAO,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,OAAO,CAAC,uCAAuC,CAAC,CAAC;AACzF,CAAC"}
@@ -1 +1 @@
1
- {"version":3,"file":"runner.d.ts","sourceRoot":"","sources":["../../src/eval/runner.ts"],"names":[],"mappings":"AAKA,OAAO,EAAgB,KAAK,cAAc,EAAoB,MAAM,cAAc,CAAC;AACnF,OAAO,EAAkB,KAAK,eAAe,EAAE,MAAM,aAAa,CAAC;AACnE,OAAO,KAAK,EAAE,UAAU,EAAE,iBAAiB,EAAE,iBAAiB,EAAE,MAAM,YAAY,CAAC;AAQnF,MAAM,WAAW,aAAa;IAC5B,QAAQ,CAAC,GAAG,EAAE,MAAM,CAAC;IACrB,QAAQ,CAAC,IAAI,CAAC,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,WAAW,CAAC,EAAE,MAAM,CAAC;IAC9B,QAAQ,CAAC,SAAS,CAAC,EAAE,MAAM,CAAC;IAC5B,QAAQ,CAAC,gBAAgB,CAAC,EAAE,MAAM,CAAC;IACnC,QAAQ,CAAC,KAAK,CAAC,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,SAAS,CAAC,EAAE,MAAM,CAAC;IAC5B,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,eAAe,KAAK,IAAI,CAAC;IACnE,QAAQ,CAAC,SAAS,CAAC,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,OAAO,EAAE,cAAc,KAAK,IAAI,CAAC;CACxE;AAED,MAAM,MAAM,UAAU,GAClB;IACE,QAAQ,CAAC,IAAI,EAAE,SAAS,CAAC;IACzB,QAAQ,CAAC,QAAQ,EAAE,iBAAiB,CAAC;IACrC,QAAQ,CAAC,IAAI,EAAE,SAAS,eAAe,EAAE,CAAC;CAC3C,GACD;IACE,QAAQ,CAAC,IAAI,EAAE,SAAS,CAAC;IACzB,QAAQ,CAAC,QAAQ,EAAE,iBAAiB,CAAC;IACrC,QAAQ,CAAC,QAAQ,EAAE,SAAS,cAAc,EAAE,CAAC;CAC9C,CAAC;AAEN,wBAAsB,QAAQ,CAC5B,KAAK,EAAE,SAAS,UAAU,EAAE,EAC5B,OAAO,EAAE,aAAa,GACrB,OAAO,CAAC,UAAU,EAAE,CAAC,CA2BvB"}
1
+ {"version":3,"file":"runner.d.ts","sourceRoot":"","sources":["../../src/eval/runner.ts"],"names":[],"mappings":"AAKA,OAAO,EAAgB,KAAK,cAAc,EAAoB,MAAM,cAAc,CAAC;AAEnF,OAAO,EAAkB,KAAK,eAAe,EAAE,MAAM,aAAa,CAAC;AACnE,OAAO,KAAK,EAAE,UAAU,EAAE,iBAAiB,EAAE,iBAAiB,EAAE,MAAM,YAAY,CAAC;AAQnF,MAAM,WAAW,aAAa;IAC5B,QAAQ,CAAC,GAAG,EAAE,MAAM,CAAC;IACrB,QAAQ,CAAC,IAAI,CAAC,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,WAAW,CAAC,EAAE,MAAM,CAAC;IAC9B,QAAQ,CAAC,SAAS,CAAC,EAAE,MAAM,CAAC;IAC5B,QAAQ,CAAC,gBAAgB,CAAC,EAAE,MAAM,CAAC;IACnC,QAAQ,CAAC,KAAK,CAAC,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,SAAS,CAAC,EAAE,MAAM,CAAC;IAC5B,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,eAAe,KAAK,IAAI,CAAC;IACnE,QAAQ,CAAC,SAAS,CAAC,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,OAAO,EAAE,cAAc,KAAK,IAAI,CAAC;CACxE;AAED,MAAM,MAAM,UAAU,GAClB;IACE,QAAQ,CAAC,IAAI,EAAE,SAAS,CAAC;IACzB,QAAQ,CAAC,QAAQ,EAAE,iBAAiB,CAAC;IACrC,QAAQ,CAAC,IAAI,EAAE,SAAS,eAAe,EAAE,CAAC;CAC3C,GACD;IACE,QAAQ,CAAC,IAAI,EAAE,SAAS,CAAC;IACzB,QAAQ,CAAC,QAAQ,EAAE,iBAAiB,CAAC;IACrC,QAAQ,CAAC,QAAQ,EAAE,SAAS,cAAc,EAAE,CAAC;CAC9C,CAAC;AAEN,wBAAsB,QAAQ,CAC5B,KAAK,EAAE,SAAS,UAAU,EAAE,EAC5B,OAAO,EAAE,aAAa,GACrB,OAAO,CAAC,UAAU,EAAE,CAAC,CA2BvB"}
@@ -3,6 +3,7 @@ import { readFile } from "node:fs/promises";
3
3
  import { isAbsolute, resolve } from "node:path";
4
4
  import { createInterface } from "node:readline";
5
5
  import { createCaptor } from "./capture.js";
6
+ import { scrubEnv, SUBPROCESS_ENV_OVERRIDES } from "./claude-env.js";
6
7
  import { createDetector } from "./detect.js";
7
8
  const DEFAULT_RUNS = 5;
8
9
  const DEFAULT_SOLVING_RUNS = 1;
@@ -68,7 +69,7 @@ async function runSession(evalCase, options, handlers) {
68
69
  const cwd = evalCase.cwd ? resolveCwd(options.cwd, evalCase.cwd) : options.cwd;
69
70
  const child = spawn(options.claudeBin ?? "claude", buildArgs(evalCase.prompt, options.model), {
70
71
  cwd,
71
- env: scrubbedEnv(),
72
+ env: scrubEnv(process.env, SUBPROCESS_ENV_OVERRIDES),
72
73
  stdio: ["ignore", "pipe", "ignore"],
73
74
  });
74
75
  const spawnFailure = new Promise((_, reject) => {
@@ -106,11 +107,6 @@ function buildArgs(prompt, model) {
106
107
  args.push("--model", model);
107
108
  return args;
108
109
  }
109
- function scrubbedEnv() {
110
- const env = { ...process.env };
111
- delete env["CLAUDECODE"];
112
- return env;
113
- }
114
110
  async function drain(stdout, onLine, done, killOnDone) {
115
111
  const lines = createInterface({ input: stdout });
116
112
  for await (const line of lines) {
@@ -1 +1 @@
1
- {"version":3,"file":"runner.js","sourceRoot":"","sources":["../../src/eval/runner.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,EAAE,MAAM,oBAAoB,CAAC;AAC3C,OAAO,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAC;AAC5C,OAAO,EAAE,UAAU,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAChD,OAAO,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AAEhD,OAAO,EAAE,YAAY,EAAyC,MAAM,cAAc,CAAC;AACnF,OAAO,EAAE,cAAc,EAAwB,MAAM,aAAa,CAAC;AAGnE,MAAM,YAAY,GAAG,CAAC,CAAC;AACvB,MAAM,oBAAoB,GAAG,CAAC,CAAC;AAC/B,MAAM,mBAAmB,GAAG,CAAC,CAAC;AAC9B,MAAM,kBAAkB,GAAG,MAAM,CAAC;AAClC,MAAM,0BAA0B,GAAG,OAAO,CAAC;AA0B3C,MAAM,CAAC,KAAK,UAAU,QAAQ,CAC5B,KAA4B,EAC5B,OAAsB;IAEtB,MAAM,OAAO,GAAG,IAAI,GAAG,EAAiC,CAAC;IACzD,MAAM,OAAO,GAAG,IAAI,GAAG,EAAgC,CAAC;IACxD,MAAM,IAAI,GAAG,KAAK,CAAC,OAAO,CAAC,CAAC,QAAQ,EAAE,EAAE;QACtC,IAAI,QAAQ,CAAC,IAAI,KAAK,SAAS;YAAE,OAAO,CAAC,GAAG,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC;;YACtD,OAAO,CAAC,GAAG,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC;QAC/B,OAAO,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,QAAQ,EAAE,OAAO,CAAC,EAAE,EAAE,GAAG,EAAE,CAAC,QAAQ,CAAC,CAAC;IAC5E,CAAC,CAAC,CAAC;IAEH,MAAM,WAAW,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,WAAW,IAAI,mBAAmB,CAAC,CAAC;IAC5E,MAAM,YAAY,CAAC,IAAI,EAAE,WAAW,EAAE,KAAK,EAAE,QAAQ,EAAE,EAAE;QACvD,IAAI,QAAQ,CAAC,IAAI,KAAK,SAAS,EAAE,CAAC;YAChC,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;YACnD,OAAO,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,IAAI,CAAC,MAAM,CAAC,CAAC;YACpC,OAAO,CAAC,KAAK,EAAE,CAAC,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC,CAAC;QACvC,CAAC;aAAM,CAAC;YACN,MAAM,OAAO,GAAG,MAAM,UAAU,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;YACpD,OAAO,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,IAAI,CAAC,OAAO,CAAC,CAAC;YACrC,OAAO,CAAC,SAAS,EAAE,CAAC,QAAQ,CAAC,EAAE,EAAE,OAAO,CAAC,CAAC;QAC5C,CAAC;IACH,CAAC,CAAC,CAAC;IAEH,OAAO,KAAK,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAC5B,QAAQ,CAAC,IAAI,KAAK,SAAS;QACzB,CAAC,CAAC,EAAE,IAAI,EAAE,SAAS,EAAE,QAAQ,EAAE,IAAI,EAAE,OAAO,CAAC,GAAG,CAAC,QAAQ,CAAC,IAAI,EAAE,EAAE;QAClE,CAAC,CAAC,EAAE,IAAI,EAAE,SAAS,EAAE,QAAQ,EAAE,QAAQ,EAAE,OAAO,CAAC,GAAG,CAAC,QAAQ,CAAC,IAAI,EAAE,EAAE,CACzE,CAAC;AACJ,CAAC;AAED,SAAS,OAAO,CAAC,QAAoB,EAAE,OAAsB;IAC3D,MAAM,QAAQ,GAAG,QAAQ,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,oBAAoB,CAAC,CAAC,CAAC,YAAY,CAAC;IACnF,OAAO,QAAQ,CAAC,IAAI,IAAI,OAAO,CAAC,IAAI,IAAI,QAAQ,CAAC;AACnD,CAAC;AAED,SAAS,eAAe,CAAC,QAA2B;IAClD,OAAO,MAAM,IAAI,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;AACrE,CAAC;AAED,KAAK,UAAU,UAAU,CACvB,QAA2B,EAC3B,OAAsB;IAEtB,MAAM,QAAQ,GAAG,cAAc,CAAC,eAAe,CAAC,QAAQ,CAAC,CAAC,CAAC;IAC3D,MAAM,OAAO,GAAG,MAAM,UAAU,CAAC,QAAQ,EAAE,OAAO,EAAE;QAClD,SAAS,EAAE,OAAO,CAAC,SAAS,IAAI,kBAAkB;QAClD,MAAM,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC;QACrC,IAAI,EAAE,GAAG,EAAE,CAAC,QAAQ,CAAC,IAAI;QACzB,UAAU,EAAE,IAAI;KACjB,CAAC,CAAC;IACH,OAAO,QAAQ,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC;AAC3D,CAAC;AAED,KAAK,UAAU,UAAU,CACvB,QAA2B,EAC3B,OAAsB;IAEtB,MAAM,MAAM,GAAG,YAAY,EAAE,CAAC;IAC9B,MAAM,OAAO,GAAG,MAAM,UAAU,CAAC,QAAQ,EAAE,OAAO,EAAE;QAClD,SAAS,EAAE,OAAO,CAAC,gBAAgB,IAAI,0BAA0B;QACjE,MAAM,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC;QACnC,IAAI,EAAE,GAAG,EAAE,CAAC,MAAM,CAAC,IAAI;QACvB,UAAU,EAAE,KAAK;KAClB,CAAC,CAAC;IACH,MAAM,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC;IAClE,OAAO,eAAe,CAAC,OAAO,EAAE,QAAQ,EAAE,OAAO,CAAC,GAAG,CAAC,CAAC;AACzD,CAAC;AASD,KAAK,UAAU,UAAU,CACvB,QAAoB,EACpB,OAAsB,EACtB,QAAyB;IAEzB,MAAM,GAAG,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,UAAU,CAAC,OAAO,CAAC,GAAG,EAAE,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC;IAC/E,MAAM,KAAK,GAAG,KAAK,CAAC,OAAO,CAAC,SAAS,IAAI,QAAQ,EAAE,SAAS,CAAC,QAAQ,CAAC,MAAM,EAAE,OAAO,CAAC,KAAK,CAAC,EAAE;QAC5F,GAAG;QACH,GAAG,EAAE,WAAW,EAAE;QAClB,KAAK,EAAE,CAAC,QAAQ,EAAE,MAAM,EAAE,QAAQ,CAAC;KACpC,CAAC,CAAC;IAEH,MAAM,YAAY,GAAG,IAAI,OAAO,CAAQ,CAAC,CAAC,EAAE,MAAM,EAAE,EAAE;QACpD,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,KAAK,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,2BAA2B,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,CAAC;IAC9F,CAAC,CAAC,CAAC;IAEH,MAAM,QAAQ,GAAG,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;IACpC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE;QAC5B,QAAQ,CAAC,OAAO,GAAG,IAAI,CAAC;QACxB,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;IACxB,CAAC,EAAE,QAAQ,CAAC,SAAS,CAAC,CAAC;IAEvB,MAAM,UAAU,GAAG,QAAQ,CAAC,UAAU,CAAC,CAAC,CAAC,GAAG,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACjF,IAAI,CAAC;QACH,MAAM,OAAO,CAAC,IAAI,CAAC;YACjB,KAAK,CAAC,KAAK,CAAC,MAAM,EAAE,QAAQ,CAAC,MAAM,EAAE,QAAQ,CAAC,IAAI,EAAE,UAAU,CAAC;YAC/D,YAAY;SACb,CAAC,CAAC;IACL,CAAC;YAAS,CAAC;QACT,YAAY,CAAC,KAAK,CAAC,CAAC;QACpB,IAAI,KAAK,CAAC,QAAQ,KAAK,IAAI;YAAE,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;IACrD,CAAC;IACD,OAAO,QAAQ,CAAC,OAAO,CAAC;AAC1B,CAAC;AAED,SAAS,SAAS,CAAC,MAAc,EAAE,KAAyB;IAC1D,MAAM,IAAI,GAAG;QACX,IAAI;QACJ,MAAM;QACN,iBAAiB;QACjB,aAAa;QACb,WAAW;QACX,4BAA4B;KAC7B,CAAC;IACF,IAAI,KAAK;QAAE,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,KAAK,CAAC,CAAC;IACvC,OAAO,IAAI,CAAC;AACd,CAAC;AAED,SAAS,WAAW;IAClB,MAAM,GAAG,GAAG,EAAE,GAAG,OAAO,CAAC,GAAG,EAAE,CAAC;IAC/B,OAAO,GAAG,CAAC,YAAY,CAAC,CAAC;IACzB,OAAO,GAAG,CAAC;AACb,CAAC;AAED,KAAK,UAAU,KAAK,CAClB,MAA6B,EAC7B,MAA8B,EAC9B,IAAmB,EACnB,UAAuB;IAEvB,MAAM,KAAK,GAAG,eAAe,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;IACjD,IAAI,KAAK,EAAE,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QAC/B,MAAM,CAAC,IAAI,CAAC,CAAC;QACb,IAAI,IAAI,EAAE,EAAE,CAAC;YACX,UAAU,EAAE,EAAE,CAAC;YACf,MAAM;QACR,CAAC;IACH,CAAC;AACH,CAAC;AAED,KAAK,UAAU,eAAe,CAC5B,OAAuB,EACvB,QAA2B,EAC3B,OAAe;IAEf,MAAM,QAAQ,GAAG,kBAAkB,CAAC,QAAQ,CAAC,CAAC;IAC9C,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,OAAO,CAAC;IAE1C,MAAM,GAAG,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,UAAU,CAAC,OAAO,EAAE,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC;IACvE,MAAM,MAAM,GAAG,IAAI,GAAG,CAAC,OAAO,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IAC/D,KAAK,MAAM,IAAI,IAAI,QAAQ,EAAE,CAAC;QAC5B,MAAM,MAAM,GAAG,MAAM,cAAc,CAAC,UAAU,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC,CAAC;QAC3D,IAAI,MAAM,KAAK,IAAI;YAAE,MAAM,CAAC,GAAG,CAAC,IAAI,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;IACnE,CAAC;IACD,OAAO,EAAE,GAAG,OAAO,EAAE,MAAM,EAAE,CAAC,GAAG,MAAM,CAAC,MAAM,EAAE,CAA2B,EAAE,CAAC;AAChF,CAAC;AAED,SAAS,kBAAkB,CAAC,QAA2B;IACrD,OAAO,QAAQ,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,KAAK,WAAW,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;AAClF,CAAC;AAED,KAAK,UAAU,cAAc,CAAC,IAAY;IACxC,IAAI,CAAC;QACH,OAAO,MAAM,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;IACtC,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC;AAED,SAAS,UAAU,CAAC,IAAY,EAAE,OAAe;IAC/C,OAAO,UAAU,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;AAChE,CAAC;AAED,KAAK,UAAU,YAAY,CACzB,KAAmB,EACnB,KAAa,EACb,MAAkC;IAElC,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,MAAM,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,IAAI,CAAC,GAAG,CAAC,KAAK,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,EAAE,KAAK,IAAI,EAAE;QAC/E,OAAO,MAAM,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC;YAC7B,MAAM,IAAI,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC;YAC3B,MAAM,IAAI,CAAC,CAAC;YACZ,IAAI,IAAI,KAAK,SAAS;gBAAE,MAAM,MAAM,CAAC,IAAI,CAAC,CAAC;QAC7C,CAAC;IACH,CAAC,CAAC,CAAC;IACH,MAAM,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;AAC7B,CAAC"}
1
+ {"version":3,"file":"runner.js","sourceRoot":"","sources":["../../src/eval/runner.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,EAAE,MAAM,oBAAoB,CAAC;AAC3C,OAAO,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAC;AAC5C,OAAO,EAAE,UAAU,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAChD,OAAO,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AAEhD,OAAO,EAAE,YAAY,EAAyC,MAAM,cAAc,CAAC;AACnF,OAAO,EAAE,QAAQ,EAAE,wBAAwB,EAAE,MAAM,iBAAiB,CAAC;AACrE,OAAO,EAAE,cAAc,EAAwB,MAAM,aAAa,CAAC;AAGnE,MAAM,YAAY,GAAG,CAAC,CAAC;AACvB,MAAM,oBAAoB,GAAG,CAAC,CAAC;AAC/B,MAAM,mBAAmB,GAAG,CAAC,CAAC;AAC9B,MAAM,kBAAkB,GAAG,MAAM,CAAC;AAClC,MAAM,0BAA0B,GAAG,OAAO,CAAC;AA0B3C,MAAM,CAAC,KAAK,UAAU,QAAQ,CAC5B,KAA4B,EAC5B,OAAsB;IAEtB,MAAM,OAAO,GAAG,IAAI,GAAG,EAAiC,CAAC;IACzD,MAAM,OAAO,GAAG,IAAI,GAAG,EAAgC,CAAC;IACxD,MAAM,IAAI,GAAG,KAAK,CAAC,OAAO,CAAC,CAAC,QAAQ,EAAE,EAAE;QACtC,IAAI,QAAQ,CAAC,IAAI,KAAK,SAAS;YAAE,OAAO,CAAC,GAAG,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC;;YACtD,OAAO,CAAC,GAAG,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC;QAC/B,OAAO,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,QAAQ,EAAE,OAAO,CAAC,EAAE,EAAE,GAAG,EAAE,CAAC,QAAQ,CAAC,CAAC;IAC5E,CAAC,CAAC,CAAC;IAEH,MAAM,WAAW,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,WAAW,IAAI,mBAAmB,CAAC,CAAC;IAC5E,MAAM,YAAY,CAAC,IAAI,EAAE,WAAW,EAAE,KAAK,EAAE,QAAQ,EAAE,EAAE;QACvD,IAAI,QAAQ,CAAC,IAAI,KAAK,SAAS,EAAE,CAAC;YAChC,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;YACnD,OAAO,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,IAAI,CAAC,MAAM,CAAC,CAAC;YACpC,OAAO,CAAC,KAAK,EAAE,CAAC,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC,CAAC;QACvC,CAAC;aAAM,CAAC;YACN,MAAM,OAAO,GAAG,MAAM,UAAU,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;YACpD,OAAO,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,IAAI,CAAC,OAAO,CAAC,CAAC;YACrC,OAAO,CAAC,SAAS,EAAE,CAAC,QAAQ,CAAC,EAAE,EAAE,OAAO,CAAC,CAAC;QAC5C,CAAC;IACH,CAAC,CAAC,CAAC;IAEH,OAAO,KAAK,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAC5B,QAAQ,CAAC,IAAI,KAAK,SAAS;QACzB,CAAC,CAAC,EAAE,IAAI,EAAE,SAAS,EAAE,QAAQ,EAAE,IAAI,EAAE,OAAO,CAAC,GAAG,CAAC,QAAQ,CAAC,IAAI,EAAE,EAAE;QAClE,CAAC,CAAC,EAAE,IAAI,EAAE,SAAS,EAAE,QAAQ,EAAE,QAAQ,EAAE,OAAO,CAAC,GAAG,CAAC,QAAQ,CAAC,IAAI,EAAE,EAAE,CACzE,CAAC;AACJ,CAAC;AAED,SAAS,OAAO,CAAC,QAAoB,EAAE,OAAsB;IAC3D,MAAM,QAAQ,GAAG,QAAQ,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,oBAAoB,CAAC,CAAC,CAAC,YAAY,CAAC;IACnF,OAAO,QAAQ,CAAC,IAAI,IAAI,OAAO,CAAC,IAAI,IAAI,QAAQ,CAAC;AACnD,CAAC;AAED,SAAS,eAAe,CAAC,QAA2B;IAClD,OAAO,MAAM,IAAI,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;AACrE,CAAC;AAED,KAAK,UAAU,UAAU,CACvB,QAA2B,EAC3B,OAAsB;IAEtB,MAAM,QAAQ,GAAG,cAAc,CAAC,eAAe,CAAC,QAAQ,CAAC,CAAC,CAAC;IAC3D,MAAM,OAAO,GAAG,MAAM,UAAU,CAAC,QAAQ,EAAE,OAAO,EAAE;QAClD,SAAS,EAAE,OAAO,CAAC,SAAS,IAAI,kBAAkB;QAClD,MAAM,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC;QACrC,IAAI,EAAE,GAAG,EAAE,CAAC,QAAQ,CAAC,IAAI;QACzB,UAAU,EAAE,IAAI;KACjB,CAAC,CAAC;IACH,OAAO,QAAQ,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC;AAC3D,CAAC;AAED,KAAK,UAAU,UAAU,CACvB,QAA2B,EAC3B,OAAsB;IAEtB,MAAM,MAAM,GAAG,YAAY,EAAE,CAAC;IAC9B,MAAM,OAAO,GAAG,MAAM,UAAU,CAAC,QAAQ,EAAE,OAAO,EAAE;QAClD,SAAS,EAAE,OAAO,CAAC,gBAAgB,IAAI,0BAA0B;QACjE,MAAM,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC;QACnC,IAAI,EAAE,GAAG,EAAE,CAAC,MAAM,CAAC,IAAI;QACvB,UAAU,EAAE,KAAK;KAClB,CAAC,CAAC;IACH,MAAM,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC;IAClE,OAAO,eAAe,CAAC,OAAO,EAAE,QAAQ,EAAE,OAAO,CAAC,GAAG,CAAC,CAAC;AACzD,CAAC;AASD,KAAK,UAAU,UAAU,CACvB,QAAoB,EACpB,OAAsB,EACtB,QAAyB;IAEzB,MAAM,GAAG,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,UAAU,CAAC,OAAO,CAAC,GAAG,EAAE,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC;IAC/E,MAAM,KAAK,GAAG,KAAK,CAAC,OAAO,CAAC,SAAS,IAAI,QAAQ,EAAE,SAAS,CAAC,QAAQ,CAAC,MAAM,EAAE,OAAO,CAAC,KAAK,CAAC,EAAE;QAC5F,GAAG;QACH,GAAG,EAAE,QAAQ,CAAC,OAAO,CAAC,GAAG,EAAE,wBAAwB,CAAC;QACpD,KAAK,EAAE,CAAC,QAAQ,EAAE,MAAM,EAAE,QAAQ,CAAC;KACpC,CAAC,CAAC;IAEH,MAAM,YAAY,GAAG,IAAI,OAAO,CAAQ,CAAC,CAAC,EAAE,MAAM,EAAE,EAAE;QACpD,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,KAAK,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,2BAA2B,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,CAAC;IAC9F,CAAC,CAAC,CAAC;IAEH,MAAM,QAAQ,GAAG,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;IACpC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE;QAC5B,QAAQ,CAAC,OAAO,GAAG,IAAI,CAAC;QACxB,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;IACxB,CAAC,EAAE,QAAQ,CAAC,SAAS,CAAC,CAAC;IAEvB,MAAM,UAAU,GAAG,QAAQ,CAAC,UAAU,CAAC,CAAC,CAAC,GAAG,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACjF,IAAI,CAAC;QACH,MAAM,OAAO,CAAC,IAAI,CAAC;YACjB,KAAK,CAAC,KAAK,CAAC,MAAM,EAAE,QAAQ,CAAC,MAAM,EAAE,QAAQ,CAAC,IAAI,EAAE,UAAU,CAAC;YAC/D,YAAY;SACb,CAAC,CAAC;IACL,CAAC;YAAS,CAAC;QACT,YAAY,CAAC,KAAK,CAAC,CAAC;QACpB,IAAI,KAAK,CAAC,QAAQ,KAAK,IAAI;YAAE,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;IACrD,CAAC;IACD,OAAO,QAAQ,CAAC,OAAO,CAAC;AAC1B,CAAC;AAED,SAAS,SAAS,CAAC,MAAc,EAAE,KAAyB;IAC1D,MAAM,IAAI,GAAG;QACX,IAAI;QACJ,MAAM;QACN,iBAAiB;QACjB,aAAa;QACb,WAAW;QACX,4BAA4B;KAC7B,CAAC;IACF,IAAI,KAAK;QAAE,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,KAAK,CAAC,CAAC;IACvC,OAAO,IAAI,CAAC;AACd,CAAC;AAED,KAAK,UAAU,KAAK,CAClB,MAA6B,EAC7B,MAA8B,EAC9B,IAAmB,EACnB,UAAuB;IAEvB,MAAM,KAAK,GAAG,eAAe,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;IACjD,IAAI,KAAK,EAAE,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QAC/B,MAAM,CAAC,IAAI,CAAC,CAAC;QACb,IAAI,IAAI,EAAE,EAAE,CAAC;YACX,UAAU,EAAE,EAAE,CAAC;YACf,MAAM;QACR,CAAC;IACH,CAAC;AACH,CAAC;AAED,KAAK,UAAU,eAAe,CAC5B,OAAuB,EACvB,QAA2B,EAC3B,OAAe;IAEf,MAAM,QAAQ,GAAG,kBAAkB,CAAC,QAAQ,CAAC,CAAC;IAC9C,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,OAAO,CAAC;IAE1C,MAAM,GAAG,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,UAAU,CAAC,OAAO,EAAE,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC;IACvE,MAAM,MAAM,GAAG,IAAI,GAAG,CAAC,OAAO,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IAC/D,KAAK,MAAM,IAAI,IAAI,QAAQ,EAAE,CAAC;QAC5B,MAAM,MAAM,GAAG,MAAM,cAAc,CAAC,UAAU,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC,CAAC;QAC3D,IAAI,MAAM,KAAK,IAAI;YAAE,MAAM,CAAC,GAAG,CAAC,IAAI,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;IACnE,CAAC;IACD,OAAO,EAAE,GAAG,OAAO,EAAE,MAAM,EAAE,CAAC,GAAG,MAAM,CAAC,MAAM,EAAE,CAA2B,EAAE,CAAC;AAChF,CAAC;AAED,SAAS,kBAAkB,CAAC,QAA2B;IACrD,OAAO,QAAQ,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,KAAK,WAAW,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;AAClF,CAAC;AAED,KAAK,UAAU,cAAc,CAAC,IAAY;IACxC,IAAI,CAAC;QACH,OAAO,MAAM,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;IACtC,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC;AAED,SAAS,UAAU,CAAC,IAAY,EAAE,OAAe;IAC/C,OAAO,UAAU,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;AAChE,CAAC;AAED,KAAK,UAAU,YAAY,CACzB,KAAmB,EACnB,KAAa,EACb,MAAkC;IAElC,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,MAAM,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,IAAI,CAAC,GAAG,CAAC,KAAK,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,EAAE,KAAK,IAAI,EAAE;QAC/E,OAAO,MAAM,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC;YAC7B,MAAM,IAAI,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC;YAC3B,MAAM,IAAI,CAAC,CAAC;YACZ,IAAI,IAAI,KAAK,SAAS;gBAAE,MAAM,MAAM,CAAC,IAAI,CAAC,CAAC;QAC7C,CAAC;IACH,CAAC,CAAC,CAAC;IACH,MAAM,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;AAC7B,CAAC"}
@@ -1,6 +1,5 @@
1
1
  import type { DetectionResult } from "./detect.js";
2
- import type { AssertionResult } from "./grade-deterministic.js";
3
- import type { RubricResult } from "./grade-judge.js";
2
+ import type { AssertionResult, RubricResult } from "./grade/index.js";
4
3
  import type { Expectation } from "./schema.js";
5
4
  export interface CaseScore {
6
5
  readonly matched: number;
@@ -1 +1 @@
1
- {"version":3,"file":"score.d.ts","sourceRoot":"","sources":["../../src/eval/score.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,aAAa,CAAC;AACnD,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,0BAA0B,CAAC;AAChE,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,kBAAkB,CAAC;AACrD,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAI/C,MAAM,WAAW,SAAS;IACxB,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,IAAI,EAAE,OAAO,CAAC;IACvB,QAAQ,CAAC,SAAS,EAAE,WAAW,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;CACjD;AAED,wBAAgB,kBAAkB,CAAC,WAAW,EAAE,WAAW,EAAE,GAAG,EAAE,eAAe,GAAG,OAAO,CAa1F;AAED,wBAAgB,SAAS,CACvB,WAAW,EAAE,WAAW,EACxB,IAAI,EAAE,SAAS,eAAe,EAAE,EAChC,SAAS,SAAoB,GAC5B,SAAS,CAWX;AAED,MAAM,WAAW,gBAAgB;IAC/B,QAAQ,CAAC,UAAU,EAAE,SAAS,eAAe,EAAE,CAAC;IAChD,QAAQ,CAAC,MAAM,EAAE,YAAY,GAAG,IAAI,CAAC;CACtC;AAED,wBAAgB,gBAAgB,CAAC,GAAG,EAAE,gBAAgB,GAAG,OAAO,CAI/D;AAED,wBAAgB,YAAY,CAC1B,MAAM,EAAE,SAAS,gBAAgB,EAAE,EACnC,SAAS,SAAoB,GAC5B,SAAS,CAWX"}
1
+ {"version":3,"file":"score.d.ts","sourceRoot":"","sources":["../../src/eval/score.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,aAAa,CAAC;AACnD,OAAO,KAAK,EAAE,eAAe,EAAE,YAAY,EAAE,MAAM,kBAAkB,CAAC;AACtE,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAI/C,MAAM,WAAW,SAAS;IACxB,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,IAAI,EAAE,OAAO,CAAC;IACvB,QAAQ,CAAC,SAAS,EAAE,WAAW,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;CACjD;AAED,wBAAgB,kBAAkB,CAAC,WAAW,EAAE,WAAW,EAAE,GAAG,EAAE,eAAe,GAAG,OAAO,CAa1F;AAED,wBAAgB,SAAS,CACvB,WAAW,EAAE,WAAW,EACxB,IAAI,EAAE,SAAS,eAAe,EAAE,EAChC,SAAS,SAAoB,GAC5B,SAAS,CAWX;AAED,MAAM,WAAW,gBAAgB;IAC/B,QAAQ,CAAC,UAAU,EAAE,SAAS,eAAe,EAAE,CAAC;IAChD,QAAQ,CAAC,MAAM,EAAE,YAAY,GAAG,IAAI,CAAC;CACtC;AAED,wBAAgB,gBAAgB,CAAC,GAAG,EAAE,gBAAgB,GAAG,OAAO,CAI/D;AAED,wBAAgB,YAAY,CAC1B,MAAM,EAAE,SAAS,gBAAgB,EAAE,EACnC,SAAS,SAAoB,GAC5B,SAAS,CAWX"}
@@ -1 +1 @@
1
- {"version":3,"file":"score.js","sourceRoot":"","sources":["../../src/eval/score.ts"],"names":[],"mappings":"AAKA,MAAM,iBAAiB,GAAG,GAAG,CAAC;AAW9B,MAAM,UAAU,kBAAkB,CAAC,WAAwB,EAAE,GAAoB;IAC/E,IAAI,WAAW,CAAC,WAAW,EAAE,GAAG,CAAC;QAAE,OAAO,KAAK,CAAC;IAEhD,IAAI,SAAS,IAAI,WAAW,EAAE,CAAC;QAC7B,OAAO,GAAG,CAAC,UAAU,KAAK,IAAI,CAAC;IACjC,CAAC;IACD,IAAI,OAAO,IAAI,WAAW,EAAE,CAAC;QAC3B,OAAO,GAAG,CAAC,UAAU,KAAK,WAAW,CAAC,KAAK,CAAC;IAC9C,CAAC;IACD,IAAI,OAAO,IAAI,WAAW,EAAE,CAAC;QAC3B,OAAO,GAAG,CAAC,UAAU,KAAK,IAAI,IAAI,WAAW,CAAC,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC;IAC/E,CAAC;IACD,OAAO,oBAAoB,CAAC,WAAW,CAAC,IAAI,EAAE,GAAG,CAAC,QAAQ,CAAC,CAAC;AAC9D,CAAC;AAED,MAAM,UAAU,SAAS,CACvB,WAAwB,EACxB,IAAgC,EAChC,SAAS,GAAG,iBAAiB;IAE7B,MAAM,OAAO,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,kBAAkB,CAAC,WAAW,EAAE,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC;IAClF,MAAM,WAAW,GAAG,IAAI,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,GAAG,IAAI,CAAC,MAAM,CAAC;IAClE,OAAO;QACL,OAAO;QACP,IAAI,EAAE,IAAI,CAAC,MAAM;QACjB,WAAW;QACX,SAAS;QACT,IAAI,EAAE,WAAW,IAAI,SAAS;QAC9B,SAAS,EAAE,WAAW,CAAC,IAAI,CAAC;KAC7B,CAAC;AACJ,CAAC;AAOD,MAAM,UAAU,gBAAgB,CAAC,GAAqB;IACpD,MAAM,cAAc,GAAG,GAAG,CAAC,UAAU,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;IAC3D,MAAM,UAAU,GAAG,GAAG,CAAC,MAAM,KAAK,IAAI,IAAI,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC;IAC1D,OAAO,cAAc,IAAI,UAAU,CAAC;AACtC,CAAC;AAED,MAAM,UAAU,YAAY,CAC1B,MAAmC,EACnC,SAAS,GAAG,iBAAiB;IAE7B,MAAM,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC,gBAAgB,CAAC,CAAC,MAAM,CAAC;IACvD,MAAM,WAAW,GAAG,MAAM,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC;IACtE,OAAO;QACL,OAAO;QACP,IAAI,EAAE,MAAM,CAAC,MAAM;QACnB,WAAW;QACX,SAAS;QACT,IAAI,EAAE,WAAW,IAAI,SAAS;QAC9B,SAAS,EAAE,IAAI,GAAG,EAAE;KACrB,CAAC;AACJ,CAAC;AAED,SAAS,WAAW,CAAC,WAAwB,EAAE,GAAoB;IACjE,MAAM,SAAS,GAAG,SAAS,IAAI,WAAW,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,WAAW,CAAC,GAAG,CAAC;IACzE,IAAI,CAAC,SAAS,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,KAAK,CAAC;IACvD,OAAO,GAAG,CAAC,UAAU,KAAK,IAAI,IAAI,SAAS,CAAC,QAAQ,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC;AACvE,CAAC;AAED,SAAS,oBAAoB,CAAC,MAAyB,EAAE,QAA2B;IAClF,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,KAAK,MAAM,EAAE,IAAI,QAAQ,EAAE,CAAC;QAC1B,IAAI,EAAE,KAAK,MAAM,CAAC,MAAM,CAAC;YAAE,MAAM,IAAI,CAAC,CAAC;QACvC,IAAI,MAAM,KAAK,MAAM,CAAC,MAAM;YAAE,OAAO,IAAI,CAAC;IAC5C,CAAC;IACD,OAAO,MAAM,KAAK,MAAM,CAAC,MAAM,CAAC;AAClC,CAAC;AAED,SAAS,WAAW,CAAC,IAAgC;IACnD,MAAM,MAAM,GAAG,IAAI,GAAG,EAAkB,CAAC;IACzC,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;QACvB,MAAM,GAAG,GAAG,GAAG,CAAC,UAAU,IAAI,YAAY,CAAC;QAC3C,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IAC9C,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC"}
1
+ {"version":3,"file":"score.js","sourceRoot":"","sources":["../../src/eval/score.ts"],"names":[],"mappings":"AAIA,MAAM,iBAAiB,GAAG,GAAG,CAAC;AAW9B,MAAM,UAAU,kBAAkB,CAAC,WAAwB,EAAE,GAAoB;IAC/E,IAAI,WAAW,CAAC,WAAW,EAAE,GAAG,CAAC;QAAE,OAAO,KAAK,CAAC;IAEhD,IAAI,SAAS,IAAI,WAAW,EAAE,CAAC;QAC7B,OAAO,GAAG,CAAC,UAAU,KAAK,IAAI,CAAC;IACjC,CAAC;IACD,IAAI,OAAO,IAAI,WAAW,EAAE,CAAC;QAC3B,OAAO,GAAG,CAAC,UAAU,KAAK,WAAW,CAAC,KAAK,CAAC;IAC9C,CAAC;IACD,IAAI,OAAO,IAAI,WAAW,EAAE,CAAC;QAC3B,OAAO,GAAG,CAAC,UAAU,KAAK,IAAI,IAAI,WAAW,CAAC,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC;IAC/E,CAAC;IACD,OAAO,oBAAoB,CAAC,WAAW,CAAC,IAAI,EAAE,GAAG,CAAC,QAAQ,CAAC,CAAC;AAC9D,CAAC;AAED,MAAM,UAAU,SAAS,CACvB,WAAwB,EACxB,IAAgC,EAChC,SAAS,GAAG,iBAAiB;IAE7B,MAAM,OAAO,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,kBAAkB,CAAC,WAAW,EAAE,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC;IAClF,MAAM,WAAW,GAAG,IAAI,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,GAAG,IAAI,CAAC,MAAM,CAAC;IAClE,OAAO;QACL,OAAO;QACP,IAAI,EAAE,IAAI,CAAC,MAAM;QACjB,WAAW;QACX,SAAS;QACT,IAAI,EAAE,WAAW,IAAI,SAAS;QAC9B,SAAS,EAAE,WAAW,CAAC,IAAI,CAAC;KAC7B,CAAC;AACJ,CAAC;AAOD,MAAM,UAAU,gBAAgB,CAAC,GAAqB;IACpD,MAAM,cAAc,GAAG,GAAG,CAAC,UAAU,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;IAC3D,MAAM,UAAU,GAAG,GAAG,CAAC,MAAM,KAAK,IAAI,IAAI,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC;IAC1D,OAAO,cAAc,IAAI,UAAU,CAAC;AACtC,CAAC;AAED,MAAM,UAAU,YAAY,CAC1B,MAAmC,EACnC,SAAS,GAAG,iBAAiB;IAE7B,MAAM,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC,gBAAgB,CAAC,CAAC,MAAM,CAAC;IACvD,MAAM,WAAW,GAAG,MAAM,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC;IACtE,OAAO;QACL,OAAO;QACP,IAAI,EAAE,MAAM,CAAC,MAAM;QACnB,WAAW;QACX,SAAS;QACT,IAAI,EAAE,WAAW,IAAI,SAAS;QAC9B,SAAS,EAAE,IAAI,GAAG,EAAE;KACrB,CAAC;AACJ,CAAC;AAED,SAAS,WAAW,CAAC,WAAwB,EAAE,GAAoB;IACjE,MAAM,SAAS,GAAG,SAAS,IAAI,WAAW,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,WAAW,CAAC,GAAG,CAAC;IACzE,IAAI,CAAC,SAAS,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,KAAK,CAAC;IACvD,OAAO,GAAG,CAAC,UAAU,KAAK,IAAI,IAAI,SAAS,CAAC,QAAQ,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC;AACvE,CAAC;AAED,SAAS,oBAAoB,CAAC,MAAyB,EAAE,QAA2B;IAClF,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,KAAK,MAAM,EAAE,IAAI,QAAQ,EAAE,CAAC;QAC1B,IAAI,EAAE,KAAK,MAAM,CAAC,MAAM,CAAC;YAAE,MAAM,IAAI,CAAC,CAAC;QACvC,IAAI,MAAM,KAAK,MAAM,CAAC,MAAM;YAAE,OAAO,IAAI,CAAC;IAC5C,CAAC;IACD,OAAO,MAAM,KAAK,MAAM,CAAC,MAAM,CAAC;AAClC,CAAC;AAED,SAAS,WAAW,CAAC,IAAgC;IACnD,MAAM,MAAM,GAAG,IAAI,GAAG,EAAkB,CAAC;IACzC,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;QACvB,MAAM,GAAG,GAAG,GAAG,CAAC,UAAU,IAAI,YAAY,CAAC;QAC3C,MAAM,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IAC9C,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC"}
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@jean.gnc/harness-kit",
3
- "version": "0.14.4",
3
+ "version": "0.14.5",
4
4
  "type": "module",
5
5
  "description": "Build your own multi-agent harness: typed toolkit for authoring plugins (skills, agents, commands, hooks) and shipping them to Claude Code and Codex from a single source tree.",
6
6
  "license": "MIT",
@@ -68,7 +68,6 @@
68
68
  ]
69
69
  },
70
70
  "dependencies": {
71
- "@anthropic-ai/sdk": "^0.100.1",
72
71
  "citty": "^0.2.2",
73
72
  "js-yaml": "^4.1.1",
74
73
  "markdownlint-cli2": "^0.18.1",
@@ -1 +0,0 @@
1
- {"version":3,"file":"grade-deterministic.d.ts","sourceRoot":"","sources":["../../src/eval/grade-deterministic.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,cAAc,CAAC;AAEnD,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,aAAa,CAAC;AAE7C,MAAM,WAAW,eAAe;IAC9B,QAAQ,CAAC,SAAS,EAAE,SAAS,CAAC;IAC9B,QAAQ,CAAC,IAAI,EAAE,OAAO,CAAC;IACvB,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;CAC3B;AAED,wBAAgB,eAAe,CAC7B,UAAU,EAAE,SAAS,SAAS,EAAE,EAChC,OAAO,EAAE,cAAc,GACtB,eAAe,EAAE,CAEnB"}
@@ -1 +0,0 @@
1
- {"version":3,"file":"grade-deterministic.js","sourceRoot":"","sources":["../../src/eval/grade-deterministic.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,GAAG,EAAE,EAAE,EAAe,MAAM,cAAc,CAAC;AAEpD,OAAO,EAAE,QAAQ,EAAE,MAAM,aAAa,CAAC;AASvC,MAAM,UAAU,eAAe,CAC7B,UAAgC,EAChC,OAAuB;IAEvB,OAAO,UAAU,CAAC,GAAG,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,KAAK,CAAC,SAAS,EAAE,OAAO,CAAC,CAAC,CAAC;AAClE,CAAC;AAED,SAAS,KAAK,CAAC,SAAoB,EAAE,OAAuB;IAC1D,QAAQ,SAAS,CAAC,IAAI,EAAE,CAAC;QACvB,KAAK,eAAe;YAClB,OAAO,WAAW,CAAC,SAAS,EAAE,OAAO,CAAC,UAAU,EAAE;gBAChD,WAAW,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC,SAAS,EAAE,YAAY,SAAS,CAAC,OAAO,GAAG,CAAC;gBACpE,UAAU,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC,SAAS,EAAE,iBAAiB,SAAS,CAAC,OAAO,GAAG,CAAC;aACzE,CAAC,CAAC;QAEL,KAAK,gBAAgB;YACnB,OAAO,WAAW,CAAC,SAAS,EAAE,OAAO,CAAC,UAAU,EAAE;gBAChD,WAAW,EAAE,CAAC,GAAG,EAAE,EAAE,CAAC,IAAI,CAAC,SAAS,EAAE,oBAAoB,GAAG,GAAG,CAAC;gBACjE,UAAU,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC,SAAS,EAAE,YAAY,SAAS,CAAC,OAAO,GAAG,CAAC;aACpE,CAAC,CAAC;QAEL,KAAK,UAAU;YACb,OAAO,KAAK,CAAC,OAAO,CAAC,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,CAAC;gBAC5C,CAAC,CAAC,IAAI,CAAC,SAAS,EAAE,QAAQ,SAAS,CAAC,IAAI,EAAE,CAAC;gBAC3C,CAAC,CAAC,IAAI,CAAC,SAAS,EAAE,gBAAgB,KAAK,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAEpE,KAAK,eAAe;YAClB,OAAO,KAAK,CAAC,OAAO,CAAC,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,CAAC;gBAC5C,CAAC,CAAC,IAAI,CAAC,SAAS,EAAE,qBAAqB,SAAS,CAAC,IAAI,EAAE,CAAC;gBACxD,CAAC,CAAC,IAAI,CAAC,SAAS,EAAE,eAAe,SAAS,CAAC,IAAI,EAAE,CAAC,CAAC;QAEvD,KAAK,WAAW;YACd,OAAO,cAAc,CAAC,SAAS,EAAE,OAAO,CAAC,CAAC;IAC9C,CAAC;AACH,CAAC;AAOD,SAAS,WAAW,CAClB,SAA2E,EAC3E,IAAY,EACZ,OAAqB;IAErB,MAAM,OAAO,GAAG,cAAc,CAAC,SAAS,CAAC,OAAO,EAAE,SAAS,CAAC,KAAK,CAAC,CAAC;IACnE,IAAI,CAAC,OAAO,CAAC,EAAE;QAAE,OAAO,IAAI,CAAC,SAAS,EAAE,OAAO,CAAC,KAAK,CAAC,CAAC;IACvD,MAAM,GAAG,GAAG,OAAO,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAChC,OAAO,GAAG,KAAK,IAAI,CAAC,CAAC,CAAC,OAAO,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC;AACxE,CAAC;AAED,SAAS,cAAc,CACrB,SAAoD,EACpD,OAAuB;IAEvB,IAAI,CAAC,OAAO,CAAC,OAAO,EAAE,SAAS,CAAC,IAAI,CAAC,EAAE,CAAC;QACtC,OAAO,IAAI,CACT,SAAS,EACT,GAAG,SAAS,CAAC,IAAI,yBAAyB,YAAY,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAC/E,CAAC;IACJ,CAAC;IACD,IAAI,SAAS,CAAC,cAAc,KAAK,SAAS,EAAE,CAAC;QAC3C,OAAO,IAAI,CAAC,SAAS,EAAE,SAAS,SAAS,CAAC,IAAI,EAAE,CAAC,CAAC;IACpD,CAAC;IACD,MAAM,OAAO,GAAG,OAAO,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,SAAS,CAAC,IAAI,CAAC,EAAE,OAAO,IAAI,EAAE,CAAC;IACrF,MAAM,OAAO,GAAG,cAAc,CAAC,SAAS,CAAC,cAAc,EAAE,SAAS,CAAC,KAAK,CAAC,CAAC;IAC1E,IAAI,CAAC,OAAO,CAAC,EAAE;QAAE,OAAO,IAAI,CAAC,SAAS,EAAE,OAAO,CAAC,KAAK,CAAC,CAAC;IACvD,OAAO,OAAO,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,IAAI;QACpC,CAAC,CAAC,IAAI,CAAC,SAAS,EAAE,GAAG,SAAS,CAAC,IAAI,kBAAkB,CAAC;QACtD,CAAC,CAAC,IAAI,CAAC,SAAS,EAAE,GAAG,SAAS,CAAC,IAAI,2BAA2B,SAAS,CAAC,cAAc,GAAG,CAAC,CAAC;AAC/F,CAAC;AAID,SAAS,cAAc,CAAC,OAAe,EAAE,KAAc;IACrD,IAAI,CAAC,KAAK,EAAE,CAAC;QACX,OAAO,EAAE,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC;IACjE,CAAC;IACD,IAAI,QAAgB,CAAC;IACrB,IAAI,CAAC;QACH,QAAQ,GAAG,IAAI,MAAM,CAAC,OAAO,CAAC,CAAC;IACjC,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,OAAO,GAAG,CAAC,kBAAkB,OAAO,MAAO,KAAe,CAAC,OAAO,EAAE,CAAC,CAAC;IACxE,CAAC;IACD,OAAO,EAAE,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC;AACxD,CAAC;AAED,SAAS,OAAO,CAAC,OAAuB,EAAE,IAAY;IACpD,OAAO,YAAY,CAAC,OAAO,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;AAC9C,CAAC;AAED,SAAS,YAAY,CAAC,OAAuB;IAC3C,OAAO,OAAO,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,EAAE,CACzC,WAAW,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,WAAW,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CACtD,CAAC;AACJ,CAAC;AAED,SAAS,WAAW,CAAC,IAAY;IAC/B,OAAO,IAAI,KAAK,OAAO,IAAI,IAAI,KAAK,MAAM,CAAC;AAC7C,CAAC;AAED,SAAS,WAAW,CAAC,KAAc;IACjC,MAAM,IAAI,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,WAAW,CAAC,CAAC;IAC1C,OAAO,OAAO,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;AAChD,CAAC;AAED,SAAS,KAAK,CAAC,OAAuB;IACpC,OAAO,OAAO,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACrD,CAAC;AAED,SAAS,IAAI,CAAC,SAAoB,EAAE,QAAgB;IAClD,OAAO,EAAE,SAAS,EAAE,IAAI,EAAE,IAAI,EAAE,QAAQ,EAAE,CAAC;AAC7C,CAAC;AAED,SAAS,IAAI,CAAC,SAAoB,EAAE,QAAgB;IAClD,OAAO,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,QAAQ,EAAE,CAAC;AAC9C,CAAC"}
@@ -1 +0,0 @@
1
- {"version":3,"file":"grade-judge.d.ts","sourceRoot":"","sources":["../../src/eval/grade-judge.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,KAAK,EAAE,OAAO,EAAE,MAAM,YAAY,CAAC;AACjD,OAAO,KAAK,EAAe,MAAM,EAAE,MAAM,aAAa,CAAC;AAEvD,MAAM,WAAW,eAAe;IAC9B,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,OAAO,EAAE,OAAO,CAAC;CAC3B;AAED,MAAM,WAAW,YAAY;IAC3B,QAAQ,CAAC,UAAU,EAAE,SAAS,eAAe,EAAE,CAAC;IAChD,QAAQ,CAAC,IAAI,EAAE,OAAO,CAAC;CACxB;AAED,wBAAsB,WAAW,CAC/B,MAAM,EAAE,MAAM,EACd,MAAM,EAAE,MAAM,EACd,KAAK,EAAE,KAAK,GACX,OAAO,CAAC,YAAY,CAAC,CAQvB"}
@@ -1 +0,0 @@
1
- {"version":3,"file":"grade-judge.js","sourceRoot":"","sources":["../../src/eval/grade-judge.ts"],"names":[],"mappings":"AAaA,MAAM,CAAC,KAAK,UAAU,WAAW,CAC/B,MAAc,EACd,MAAc,EACd,KAAY;IAEZ,MAAM,UAAU,GAAG,MAAM,OAAO,CAAC,GAAG,CAClC,MAAM,CAAC,UAAU,CAAC,GAAG,CAAC,KAAK,EAAE,EAAE,SAAS,EAAE,SAAS,EAAE,EAAE,EAAE,CAAC,CAAC;QACzD,SAAS;QACT,OAAO,EAAE,MAAM,KAAK,CAAC,EAAE,SAAS,EAAE,SAAS,EAAE,MAAM,EAAE,CAAC;KACvD,CAAC,CAAC,CACJ,CAAC;IACF,OAAO,EAAE,UAAU,EAAE,IAAI,EAAE,OAAO,CAAC,MAAM,CAAC,OAAO,EAAE,UAAU,CAAC,EAAE,CAAC;AACnE,CAAC;AAED,SAAS,OAAO,CAAC,IAAiB,EAAE,UAAsC;IACxE,MAAM,MAAM,GAAG,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,KAAK,IAAI,CAAC,CAAC,MAAM,CAAC;IACxE,IAAI,IAAI,CAAC,OAAO,KAAK,KAAK;QAAE,OAAO,MAAM,KAAK,UAAU,CAAC,MAAM,CAAC;IAChE,OAAO,UAAU,CAAC,MAAM,GAAG,CAAC,IAAI,MAAM,GAAG,UAAU,CAAC,MAAM,IAAI,IAAI,CAAC,SAAS,CAAC;AAC/E,CAAC"}
@@ -1,5 +0,0 @@
1
- import type { Judge } from "./judge.js";
2
- import type { CaseReport } from "./report.js";
3
- import type { CaseResult } from "./runner.js";
4
- export declare function gradeResults(results: readonly CaseResult[], judge: Judge | undefined): Promise<CaseReport[]>;
5
- //# sourceMappingURL=grade.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"grade.d.ts","sourceRoot":"","sources":["../../src/eval/grade.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,KAAK,EAAE,MAAM,YAAY,CAAC;AACxC,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAC9C,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAI9C,wBAAsB,YAAY,CAChC,OAAO,EAAE,SAAS,UAAU,EAAE,EAC9B,KAAK,EAAE,KAAK,GAAG,SAAS,GACvB,OAAO,CAAC,UAAU,EAAE,CAAC,CAEvB"}
@@ -1 +0,0 @@
1
- {"version":3,"file":"grade.js","sourceRoot":"","sources":["../../src/eval/grade.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,eAAe,EAAE,MAAM,0BAA0B,CAAC;AAC3D,OAAO,EAAE,WAAW,EAAqB,MAAM,kBAAkB,CAAC;AAIlE,OAAO,EAAE,SAAS,EAAE,YAAY,EAAyB,MAAM,YAAY,CAAC;AAG5E,MAAM,CAAC,KAAK,UAAU,YAAY,CAChC,OAA8B,EAC9B,KAAwB;IAExB,OAAO,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,CAAC,QAAQ,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC;AAC7E,CAAC;AAED,KAAK,UAAU,QAAQ,CAAC,MAAkB,EAAE,KAAwB;IAClE,IAAI,MAAM,CAAC,IAAI,KAAK,SAAS,EAAE,CAAC;QAC9B,MAAM,EAAE,QAAQ,EAAE,IAAI,EAAE,GAAG,MAAM,CAAC;QAClC,OAAO,EAAE,QAAQ,EAAE,KAAK,EAAE,SAAS,CAAC,QAAQ,CAAC,MAAM,EAAE,IAAI,EAAE,QAAQ,CAAC,SAAS,CAAC,EAAE,CAAC;IACnF,CAAC;IAED,MAAM,EAAE,QAAQ,EAAE,QAAQ,EAAE,GAAG,MAAM,CAAC;IACtC,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,GAAG,CAC9B,QAAQ,CAAC,GAAG,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE,CAAC,QAAQ,CAAC,QAAQ,CAAC,MAAM,EAAE,QAAQ,CAAC,MAAM,EAAE,OAAO,EAAE,KAAK,CAAC,CAAC,CAC5F,CAAC;IACF,OAAO;QACL,QAAQ;QACR,KAAK,EAAE,YAAY,CAAC,MAAM,EAAE,QAAQ,CAAC,SAAS,CAAC;QAC/C,OAAO,EAAE,EAAE,MAAM,EAAE;KACpB,CAAC;AACJ,CAAC;AAED,KAAK,UAAU,QAAQ,CACrB,UAAgC,EAChC,MAA0B,EAC1B,OAAuB,EACvB,KAAwB;IAExB,MAAM,MAAM,GAAG,eAAe,CAAC,UAAU,EAAE,OAAO,CAAC,CAAC;IACpD,MAAM,YAAY,GAChB,MAAM,IAAI,KAAK,CAAC,CAAC,CAAC,MAAM,WAAW,CAAC,MAAM,EAAE,OAAO,CAAC,UAAU,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IAChF,OAAO,EAAE,UAAU,EAAE,MAAM,EAAE,MAAM,EAAE,YAAY,EAAE,CAAC;AACtD,CAAC"}
@@ -1 +0,0 @@
1
- {"version":3,"file":"judge.d.ts","sourceRoot":"","sources":["../../src/eval/judge.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,eAAO,MAAM,OAAO;;;;;;;;;EAGlB,CAAC;AAEH,MAAM,MAAM,OAAO,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,OAAO,CAAC,CAAC;AAE9C,MAAM,WAAW,YAAY;IAC3B,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;CACzB;AAED,MAAM,MAAM,KAAK,GAAG,CAAC,OAAO,EAAE,YAAY,KAAK,OAAO,CAAC,OAAO,CAAC,CAAC;AAMhE,wBAAgB,YAAY,CAAC,GAAG,EAAE,OAAO,GAAG,OAAO,CAIlD;AAwBD,MAAM,WAAW,qBAAqB;IACpC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,SAAS,CAAC,EAAE,MAAM,CAAC;CAC7B;AAED,wBAAgB,oBAAoB,CAAC,OAAO,EAAE,qBAAqB,GAAG,KAAK,CAmB1E"}
@@ -1,55 +0,0 @@
1
- import Anthropic from "@anthropic-ai/sdk";
2
- import { z } from "zod";
3
- export const Verdict = z.object({
4
- pass: z.union([z.boolean(), z.literal("unknown")]),
5
- evidence: z.string(),
6
- });
7
- function abstain(evidence) {
8
- return { pass: "unknown", evidence };
9
- }
10
- export function parseVerdict(raw) {
11
- if (raw === undefined || raw === null)
12
- return abstain("judge returned no verdict");
13
- const result = Verdict.safeParse(raw);
14
- return result.success ? result.data : abstain("judge returned an unparseable verdict");
15
- }
16
- const SYSTEM_PREAMBLE = "You are a strict grader for AI agent outputs. Judge the output against the single criterion " +
17
- "you are given, in isolation. Record a verdict via the tool: pass=true only if the criterion is " +
18
- 'clearly met, pass=false if clearly unmet, and pass="unknown" if you genuinely cannot tell. ' +
19
- "Always cite concrete evidence from the output.";
20
- const VERDICT_TOOL = {
21
- name: "record_verdict",
22
- description: "Record the grading verdict for one rubric dimension.",
23
- input_schema: {
24
- type: "object",
25
- properties: {
26
- pass: {
27
- description: 'true if met, false if unmet, "unknown" if undeterminable',
28
- anyOf: [{ type: "boolean" }, { type: "string", enum: ["unknown"] }],
29
- },
30
- evidence: { type: "string", description: "concrete evidence from the output" },
31
- },
32
- required: ["pass", "evidence"],
33
- },
34
- };
35
- export function createAnthropicJudge(options) {
36
- const client = new Anthropic({ apiKey: options.apiKey });
37
- return async ({ dimension, criterion, output }) => {
38
- const response = await client.messages.create({
39
- model: options.model,
40
- max_tokens: options.maxTokens ?? 1024,
41
- tools: [VERDICT_TOOL],
42
- tool_choice: { type: "tool", name: VERDICT_TOOL.name },
43
- system: [{ type: "text", text: SYSTEM_PREAMBLE, cache_control: { type: "ephemeral" } }],
44
- messages: [
45
- {
46
- role: "user",
47
- content: `Dimension: ${dimension}\nCriterion: ${criterion}\n\nOutput to grade:\n${output}`,
48
- },
49
- ],
50
- });
51
- const block = response.content.find((b) => b.type === "tool_use");
52
- return parseVerdict(block?.input);
53
- };
54
- }
55
- //# sourceMappingURL=judge.js.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"judge.js","sourceRoot":"","sources":["../../src/eval/judge.ts"],"names":[],"mappings":"AAAA,OAAO,SAAS,MAAM,mBAAmB,CAAC;AAC1C,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,MAAM,CAAC,MAAM,OAAO,GAAG,CAAC,CAAC,MAAM,CAAC;IAC9B,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC;IAClD,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE;CACrB,CAAC,CAAC;AAYH,SAAS,OAAO,CAAC,QAAgB;IAC/B,OAAO,EAAE,IAAI,EAAE,SAAS,EAAE,QAAQ,EAAE,CAAC;AACvC,CAAC;AAED,MAAM,UAAU,YAAY,CAAC,GAAY;IACvC,IAAI,GAAG,KAAK,SAAS,IAAI,GAAG,KAAK,IAAI;QAAE,OAAO,OAAO,CAAC,2BAA2B,CAAC,CAAC;IACnF,MAAM,MAAM,GAAG,OAAO,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC;IACtC,OAAO,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,OAAO,CAAC,uCAAuC,CAAC,CAAC;AACzF,CAAC;AAED,MAAM,eAAe,GACnB,8FAA8F;IAC9F,iGAAiG;IACjG,6FAA6F;IAC7F,gDAAgD,CAAC;AAEnD,MAAM,YAAY,GAAmB;IACnC,IAAI,EAAE,gBAAgB;IACtB,WAAW,EAAE,sDAAsD;IACnE,YAAY,EAAE;QACZ,IAAI,EAAE,QAAQ;QACd,UAAU,EAAE;YACV,IAAI,EAAE;gBACJ,WAAW,EAAE,0DAA0D;gBACvE,KAAK,EAAE,CAAC,EAAE,IAAI,EAAE,SAAS,EAAE,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC,SAAS,CAAC,EAAE,CAAC;aACpE;YACD,QAAQ,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,mCAAmC,EAAE;SAC/E;QACD,QAAQ,EAAE,CAAC,MAAM,EAAE,UAAU,CAAC;KAC/B;CACF,CAAC;AAQF,MAAM,UAAU,oBAAoB,CAAC,OAA8B;IACjE,MAAM,MAAM,GAAG,IAAI,SAAS,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC;IACzD,OAAO,KAAK,EAAE,EAAE,SAAS,EAAE,SAAS,EAAE,MAAM,EAAE,EAAE,EAAE;QAChD,MAAM,QAAQ,GAAG,MAAM,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC;YAC5C,KAAK,EAAE,OAAO,CAAC,KAAK;YACpB,UAAU,EAAE,OAAO,CAAC,SAAS,IAAI,IAAI;YACrC,KAAK,EAAE,CAAC,YAAY,CAAC;YACrB,WAAW,EAAE,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,YAAY,CAAC,IAAI,EAAE;YACtD,MAAM,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,eAAe,EAAE,aAAa,EAAE,EAAE,IAAI,EAAE,WAAW,EAAE,EAAE,CAAC;YACvF,QAAQ,EAAE;gBACR;oBACE,IAAI,EAAE,MAAM;oBACZ,OAAO,EAAE,cAAc,SAAS,gBAAgB,SAAS,yBAAyB,MAAM,EAAE;iBAC3F;aACF;SACF,CAAC,CAAC;QACH,MAAM,KAAK,GAAG,QAAQ,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,UAAU,CAAC,CAAC;QAClE,OAAO,YAAY,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC;IACpC,CAAC,CAAC;AACJ,CAAC"}