chivgent 0.14.0 → 0.16.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (64) hide show
  1. package/README.md +85 -5
  2. package/README.zh-CN.md +68 -5
  3. package/dist/agent.d.ts +4 -0
  4. package/dist/agent.d.ts.map +1 -1
  5. package/dist/agent.js +18 -1
  6. package/dist/agent.js.map +1 -1
  7. package/dist/cli-options.d.ts.map +1 -1
  8. package/dist/cli-options.js +6 -0
  9. package/dist/cli-options.js.map +1 -1
  10. package/dist/context/compaction.d.ts +11 -2
  11. package/dist/context/compaction.d.ts.map +1 -1
  12. package/dist/context/compaction.js +15 -3
  13. package/dist/context/compaction.js.map +1 -1
  14. package/dist/context/context-manager.d.ts +3 -0
  15. package/dist/context/context-manager.d.ts.map +1 -1
  16. package/dist/context/context-manager.js +2 -1
  17. package/dist/context/context-manager.js.map +1 -1
  18. package/dist/evals/cli.js +2 -0
  19. package/dist/evals/cli.js.map +1 -1
  20. package/dist/evals/graders.d.ts +2 -0
  21. package/dist/evals/graders.d.ts.map +1 -1
  22. package/dist/evals/graders.js +70 -0
  23. package/dist/evals/graders.js.map +1 -1
  24. package/dist/evals/parse-args.d.ts.map +1 -1
  25. package/dist/evals/parse-args.js +8 -1
  26. package/dist/evals/parse-args.js.map +1 -1
  27. package/dist/evals/report.d.ts.map +1 -1
  28. package/dist/evals/report.js +22 -3
  29. package/dist/evals/report.js.map +1 -1
  30. package/dist/evals/runner.d.ts +33 -1
  31. package/dist/evals/runner.d.ts.map +1 -1
  32. package/dist/evals/runner.js +48 -25
  33. package/dist/evals/runner.js.map +1 -1
  34. package/dist/evals/task.d.ts.map +1 -1
  35. package/dist/evals/task.js +35 -1
  36. package/dist/evals/task.js.map +1 -1
  37. package/dist/evals/tools.d.ts +5 -0
  38. package/dist/evals/tools.d.ts.map +1 -0
  39. package/dist/evals/tools.js +40 -0
  40. package/dist/evals/tools.js.map +1 -0
  41. package/dist/events.d.ts +6 -0
  42. package/dist/events.d.ts.map +1 -1
  43. package/dist/events.js.map +1 -1
  44. package/dist/llm.d.ts +22 -0
  45. package/dist/llm.d.ts.map +1 -1
  46. package/dist/llm.js.map +1 -1
  47. package/dist/providers/openai-compatible-chat.d.ts.map +1 -1
  48. package/dist/providers/openai-compatible-chat.js +13 -3
  49. package/dist/providers/openai-compatible-chat.js.map +1 -1
  50. package/dist/providers/openai.d.ts.map +1 -1
  51. package/dist/providers/openai.js +3 -0
  52. package/dist/providers/openai.js.map +1 -1
  53. package/dist/providers/usage.d.ts +22 -0
  54. package/dist/providers/usage.d.ts.map +1 -0
  55. package/dist/providers/usage.js +92 -0
  56. package/dist/providers/usage.js.map +1 -0
  57. package/dist/repl.d.ts.map +1 -1
  58. package/dist/repl.js +21 -0
  59. package/dist/repl.js.map +1 -1
  60. package/dist/session.d.ts +4 -0
  61. package/dist/session.d.ts.map +1 -1
  62. package/dist/session.js +10 -0
  63. package/dist/session.js.map +1 -1
  64. package/package.json +1 -1
@@ -1 +1 @@
1
- {"version":3,"file":"context-manager.js","sourceRoot":"","sources":["../../src/context/context-manager.ts"],"names":[],"mappings":"AACA,OAAO,EACL,SAAS,EACT,qBAAqB,GAEtB,MAAM,iBAAiB,CAAC;AACzB,OAAO,EACL,uBAAuB,GAExB,MAAM,sBAAsB,CAAC;AAE9B,MAAM,CAAC,MAAM,sBAAsB,GAAG,OAAO,CAAC;AAC9C,MAAM,CAAC,MAAM,sBAAsB,GAAG,MAAM,CAAC;AAC7C,MAAM,CAAC,MAAM,0BAA0B,GAAG,MAAM,CAAC;AA+CjD;;;;;GAKG;AACH,MAAM,OAAO,cAAc;IACR,aAAa,CAAS;IACtB,aAAa,CAAS;IACtB,gBAAgB,CAAS;IACzB,SAAS,CAAiB;IAC1B,SAAS,CAAa;IACtB,YAAY,CAAoC;IAEjE,YAAY,OAAO,GAA0B,EAAE;QAC7C,IAAI,CAAC,aAAa,GAAG,OAAO,CAAC,aAAa,IAAI,sBAAsB,CAAC;QACrE,yEAAyE;QACzE,+DAA+D;QAC/D,IAAI,CAAC,aAAa;YAChB,OAAO,CAAC,aAAa;gBACrB,IAAI,CAAC,GAAG,CAAC,sBAAsB,EAAE,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,aAAa,GAAG,CAAC,CAAC,CAAC,CAAC;QACvE,IAAI,CAAC,gBAAgB;YACnB,OAAO,CAAC,gBAAgB;gBACxB,IAAI,CAAC,GAAG,CAAC,0BAA0B,EAAE,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,aAAa,GAAG,CAAC,CAAC,CAAC,CAAC;QAC3E,IAAI,CAAC,SAAS,GAAG,OAAO,CAAC,SAAS,IAAI,IAAI,uBAAuB,EAAE,CAAC;QACpE,IAAI,OAAO,CAAC,SAAS,KAAK,SAAS,EAAE,CAAC;YACpC,IAAI,CAAC,SAAS,GAAG,OAAO,CAAC,SAAS,CAAC;QACrC,CAAC;QACD,IAAI,OAAO,CAAC,YAAY,KAAK,SAAS,EAAE,CAAC;YACvC,IAAI,CAAC,YAAY,GAAG,OAAO,CAAC,YAAY,CAAC;QAC3C,CAAC;QACD,IAAI,IAAI,CAAC,aAAa,IAAI,IAAI,CAAC,aAAa,EAAE,CAAC;YAC7C,MAAM,IAAI,SAAS,CAAC,mDAAmD,CAAC,CAAC;QAC3E,CAAC;IACH,CAAC;IAED,IAAI,MAAM;QACR,OAAO,IAAI,CAAC,aAAa,GAAG,IAAI,CAAC,aAAa,CAAC;IACjD,CAAC;IAED,KAAK,CAAC,KAAK,CACT,QAA4B,EAC5B,OAAO,GAAiB,EAAE;QAE1B,MAAM,QAAQ,GAAG,OAAO,CAAC,QAAQ,CAAC;QAClC,MAAM,SAAS,GACb,QAAQ,KAAK,SAAS;YACpB,CAAC,CAAC,QAAQ;YACV,CAAC,CAAC,CAAC,cAAc,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,GAAG,QAAQ,CAAC,KAAK,CAAC,QAAQ,CAAC,UAAU,CAAC,CAAC,CAAC;QAC/E,MAAM,eAAe,GAAG,IAAI,CAAC,SAAS,CAAC,gBAAgB,CAAC,SAAS,CAAC,CAAC;QAEnE,IAAI,IAAI,CAAC,SAAS,KAAK,SAAS,IAAI,eAAe,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC;YACnE,OAAO;gBACL,QAAQ,EAAE,SAAS;gBACnB,SAAS,EAAE,KAAK;gBAChB,eAAe;gBACf,GAAG,CAAC,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,QAAQ,EAAE,CAAC;aAC5D,CAAC;QACJ,CAAC;QAED,MAAM,GAAG,GAAG,cAAc,CAAC,SAAS,EAAE,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,gBAAgB,CAAC,CAAC;QAC7E,4EAA4E;QAC5E,oEAAoE;QACpE,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,KAAK,SAAS,IAAI,GAAG,GAAG,CAAC,CAAC,EAAE,CAAC;YACpD,OAAO;gBACL,QAAQ,EAAE,SAAS;gBACnB,SAAS,EAAE,KAAK;gBAChB,eAAe;gBACf,GAAG,CAAC,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,QAAQ,EAAE,CAAC;aAC5D,CAAC;QACJ,CAAC;QAED,MAAM,IAAI,GAAG,SAAS,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;QACrC,MAAM,IAAI,GAAG,SAAS,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QAClC,MAAM,KAAK,GAAG,MAAM,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,IAAI,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC;QACjE,MAAM,iBAAiB,GAAG,CAAC,cAAc,CAAC,KAAK,CAAC,EAAE,GAAG,IAAI,CAAC,CAAC;QAC3D,MAAM,WAAW,GAAG,IAAI,CAAC,SAAS,CAAC,gBAAgB,CAAC,iBAAiB,CAAC,CAAC;QAEvE,sEAAsE;QACtE,8DAA8D;QAC9D,MAAM,UAAU,GACd,QAAQ,KAAK,SAAS;YACpB,CAAC,CAAC,GAAG;YACL,CAAC,CAAC,QAAQ,CAAC,UAAU,GAAG,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC;QAEtC,IAAI,CAAC,YAAY,EAAE,CAAC;YAClB,eAAe,EAAE,IAAI,CAAC,MAAM;YAC5B,YAAY,EAAE,eAAe;YAC7B,WAAW;YACX,KAAK;SACN,CAAC,CAAC;QAEH,OAAO;YACL,QAAQ,EAAE,iBAAiB;YAC3B,SAAS,EAAE,IAAI;YACf,eAAe,EAAE,WAAW;YAC5B,UAAU,EAAE,EAAE,KAAK,EAAE,UAAU,EAAE;SAClC,CAAC;IACJ,CAAC;CACF;AAED,SAAS,cAAc,CAAC,KAAsB;IAC5C,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,qBAAqB,CAAC,KAAK,CAAC,EAAE,CAAC;AACjE,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,cAAc,CAC5B,QAA4B,EAC5B,SAAyB,EACzB,gBAAwB;IAExB,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,IAAI,KAAK,GAAG,QAAQ,CAAC,MAAM,CAAC;IAC5B,OAAO,KAAK,GAAG,CAAC,EAAE,CAAC;QACjB,MAAM,OAAO,GAAG,QAAQ,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC;QACpC,IAAI,OAAO,KAAK,SAAS,EAAE,CAAC;YAC1B,MAAM;QACR,CAAC;QACD,MAAM,IAAI,GAAG,SAAS,CAAC,eAAe,CAAC,OAAO,CAAC,CAAC;QAChD,IAAI,MAAM,GAAG,IAAI,GAAG,gBAAgB,EAAE,CAAC;YACrC,MAAM;QACR,CAAC;QACD,MAAM,IAAI,IAAI,CAAC;QACf,KAAK,IAAI,CAAC,CAAC;IACb,CAAC;IAED,OAAO,KAAK,GAAG,QAAQ,CAAC,MAAM,IAAI,QAAQ,CAAC,KAAK,CAAC,EAAE,IAAI,KAAK,MAAM,EAAE,CAAC;QACnE,KAAK,IAAI,CAAC,CAAC;IACb,CAAC;IAED,4DAA4D;IAC5D,IAAI,KAAK,IAAI,QAAQ,CAAC,MAAM,EAAE,CAAC;QAC7B,KAAK,GAAG,QAAQ,CAAC,MAAM,GAAG,CAAC,CAAC;QAC5B,OAAO,KAAK,GAAG,CAAC,IAAI,QAAQ,CAAC,KAAK,CAAC,EAAE,IAAI,KAAK,MAAM,EAAE,CAAC;YACrD,KAAK,IAAI,CAAC,CAAC;QACb,CAAC;IACH,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC"}
1
+ {"version":3,"file":"context-manager.js","sourceRoot":"","sources":["../../src/context/context-manager.ts"],"names":[],"mappings":"AAEA,OAAO,EACL,SAAS,EACT,qBAAqB,GAEtB,MAAM,iBAAiB,CAAC;AACzB,OAAO,EACL,uBAAuB,GAExB,MAAM,sBAAsB,CAAC;AAE9B,MAAM,CAAC,MAAM,sBAAsB,GAAG,OAAO,CAAC;AAC9C,MAAM,CAAC,MAAM,sBAAsB,GAAG,MAAM,CAAC;AAC7C,MAAM,CAAC,MAAM,0BAA0B,GAAG,MAAM,CAAC;AAiDjD;;;;;GAKG;AACH,MAAM,OAAO,cAAc;IACR,aAAa,CAAS;IACtB,aAAa,CAAS;IACtB,gBAAgB,CAAS;IACzB,SAAS,CAAiB;IAC1B,SAAS,CAAa;IACtB,YAAY,CAAoC;IAEjE,YAAY,OAAO,GAA0B,EAAE;QAC7C,IAAI,CAAC,aAAa,GAAG,OAAO,CAAC,aAAa,IAAI,sBAAsB,CAAC;QACrE,yEAAyE;QACzE,+DAA+D;QAC/D,IAAI,CAAC,aAAa;YAChB,OAAO,CAAC,aAAa;gBACrB,IAAI,CAAC,GAAG,CAAC,sBAAsB,EAAE,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,aAAa,GAAG,CAAC,CAAC,CAAC,CAAC;QACvE,IAAI,CAAC,gBAAgB;YACnB,OAAO,CAAC,gBAAgB;gBACxB,IAAI,CAAC,GAAG,CAAC,0BAA0B,EAAE,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,aAAa,GAAG,CAAC,CAAC,CAAC,CAAC;QAC3E,IAAI,CAAC,SAAS,GAAG,OAAO,CAAC,SAAS,IAAI,IAAI,uBAAuB,EAAE,CAAC;QACpE,IAAI,OAAO,CAAC,SAAS,KAAK,SAAS,EAAE,CAAC;YACpC,IAAI,CAAC,SAAS,GAAG,OAAO,CAAC,SAAS,CAAC;QACrC,CAAC;QACD,IAAI,OAAO,CAAC,YAAY,KAAK,SAAS,EAAE,CAAC;YACvC,IAAI,CAAC,YAAY,GAAG,OAAO,CAAC,YAAY,CAAC;QAC3C,CAAC;QACD,IAAI,IAAI,CAAC,aAAa,IAAI,IAAI,CAAC,aAAa,EAAE,CAAC;YAC7C,MAAM,IAAI,SAAS,CAAC,mDAAmD,CAAC,CAAC;QAC3E,CAAC;IACH,CAAC;IAED,IAAI,MAAM;QACR,OAAO,IAAI,CAAC,aAAa,GAAG,IAAI,CAAC,aAAa,CAAC;IACjD,CAAC;IAED,KAAK,CAAC,KAAK,CACT,QAA4B,EAC5B,OAAO,GAAiB,EAAE;QAE1B,MAAM,QAAQ,GAAG,OAAO,CAAC,QAAQ,CAAC;QAClC,MAAM,SAAS,GACb,QAAQ,KAAK,SAAS;YACpB,CAAC,CAAC,QAAQ;YACV,CAAC,CAAC,CAAC,cAAc,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,GAAG,QAAQ,CAAC,KAAK,CAAC,QAAQ,CAAC,UAAU,CAAC,CAAC,CAAC;QAC/E,MAAM,eAAe,GAAG,IAAI,CAAC,SAAS,CAAC,gBAAgB,CAAC,SAAS,CAAC,CAAC;QAEnE,IAAI,IAAI,CAAC,SAAS,KAAK,SAAS,IAAI,eAAe,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC;YACnE,OAAO;gBACL,QAAQ,EAAE,SAAS;gBACnB,SAAS,EAAE,KAAK;gBAChB,eAAe;gBACf,GAAG,CAAC,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,QAAQ,EAAE,CAAC;aAC5D,CAAC;QACJ,CAAC;QAED,MAAM,GAAG,GAAG,cAAc,CAAC,SAAS,EAAE,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,gBAAgB,CAAC,CAAC;QAC7E,4EAA4E;QAC5E,oEAAoE;QACpE,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,KAAK,SAAS,IAAI,GAAG,GAAG,CAAC,CAAC,EAAE,CAAC;YACpD,OAAO;gBACL,QAAQ,EAAE,SAAS;gBACnB,SAAS,EAAE,KAAK;gBAChB,eAAe;gBACf,GAAG,CAAC,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,QAAQ,EAAE,CAAC;aAC5D,CAAC;QACJ,CAAC;QAED,MAAM,IAAI,GAAG,SAAS,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;QACrC,MAAM,IAAI,GAAG,SAAS,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QAClC,MAAM,EAAE,KAAK,EAAE,KAAK,EAAE,GAAG,MAAM,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,IAAI,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC;QAC5E,MAAM,iBAAiB,GAAG,CAAC,cAAc,CAAC,KAAK,CAAC,EAAE,GAAG,IAAI,CAAC,CAAC;QAC3D,MAAM,WAAW,GAAG,IAAI,CAAC,SAAS,CAAC,gBAAgB,CAAC,iBAAiB,CAAC,CAAC;QAEvE,sEAAsE;QACtE,8DAA8D;QAC9D,MAAM,UAAU,GACd,QAAQ,KAAK,SAAS;YACpB,CAAC,CAAC,GAAG;YACL,CAAC,CAAC,QAAQ,CAAC,UAAU,GAAG,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC;QAEtC,IAAI,CAAC,YAAY,EAAE,CAAC;YAClB,eAAe,EAAE,IAAI,CAAC,MAAM;YAC5B,YAAY,EAAE,eAAe;YAC7B,WAAW;YACX,KAAK;SACN,CAAC,CAAC;QAEH,OAAO;YACL,QAAQ,EAAE,iBAAiB;YAC3B,SAAS,EAAE,IAAI;YACf,eAAe,EAAE,WAAW;YAC5B,UAAU,EAAE,EAAE,KAAK,EAAE,UAAU,EAAE;YACjC,GAAG,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,CAAC;SAC1C,CAAC;IACJ,CAAC;CACF;AAED,SAAS,cAAc,CAAC,KAAsB;IAC5C,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,qBAAqB,CAAC,KAAK,CAAC,EAAE,CAAC;AACjE,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,cAAc,CAC5B,QAA4B,EAC5B,SAAyB,EACzB,gBAAwB;IAExB,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,IAAI,KAAK,GAAG,QAAQ,CAAC,MAAM,CAAC;IAC5B,OAAO,KAAK,GAAG,CAAC,EAAE,CAAC;QACjB,MAAM,OAAO,GAAG,QAAQ,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC;QACpC,IAAI,OAAO,KAAK,SAAS,EAAE,CAAC;YAC1B,MAAM;QACR,CAAC;QACD,MAAM,IAAI,GAAG,SAAS,CAAC,eAAe,CAAC,OAAO,CAAC,CAAC;QAChD,IAAI,MAAM,GAAG,IAAI,GAAG,gBAAgB,EAAE,CAAC;YACrC,MAAM;QACR,CAAC;QACD,MAAM,IAAI,IAAI,CAAC;QACf,KAAK,IAAI,CAAC,CAAC;IACb,CAAC;IAED,OAAO,KAAK,GAAG,QAAQ,CAAC,MAAM,IAAI,QAAQ,CAAC,KAAK,CAAC,EAAE,IAAI,KAAK,MAAM,EAAE,CAAC;QACnE,KAAK,IAAI,CAAC,CAAC;IACb,CAAC;IAED,4DAA4D;IAC5D,IAAI,KAAK,IAAI,QAAQ,CAAC,MAAM,EAAE,CAAC;QAC7B,KAAK,GAAG,QAAQ,CAAC,MAAM,GAAG,CAAC,CAAC;QAC5B,OAAO,KAAK,GAAG,CAAC,IAAI,QAAQ,CAAC,KAAK,CAAC,EAAE,IAAI,KAAK,MAAM,EAAE,CAAC;YACrD,KAAK,IAAI,CAAC,CAAC;QACb,CAAC;IACH,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC"}
package/dist/evals/cli.js CHANGED
@@ -21,6 +21,8 @@ Options:
21
21
  --json PATH Also write a JSON report
22
22
  --provider NAME Provider to evaluate
23
23
  --model MODEL Model to evaluate
24
+ --api-key KEY Credential to evaluate with (else the environment or
25
+ the stored auth.json, as the main CLI resolves it)
24
26
  --allow-writes Grant tasks that need to change files
25
27
  --allow-shell Grant tasks that need a shell
26
28
  -h, --help Show this help
@@ -1 +1 @@
1
- {"version":3,"file":"cli.js","sourceRoot":"","sources":["../../src/evals/cli.ts"],"names":[],"mappings":";AAEA,OAAO,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAC;AAC7C,OAAO,IAAI,MAAM,WAAW,CAAC;AAC7B,OAAO,OAAO,MAAM,cAAc,CAAC;AACnC,OAAO,EAAE,YAAY,EAAE,OAAO,EAAE,MAAM,mBAAmB,CAAC;AAC1D,OAAO,EAAE,sBAAsB,EAAE,MAAM,wBAAwB,CAAC;AAChE,OAAO,EACL,mBAAmB,EACnB,aAAa,EACb,mBAAmB,GACpB,MAAM,eAAe,CAAC;AACvB,OAAO,EAAE,cAAc,EAAE,WAAW,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AACxE,OAAO,EAAE,sBAAsB,EAAE,OAAO,EAAmB,MAAM,aAAa,CAAC;AAC/E,OAAO,EAAE,SAAS,EAAE,SAAS,EAA8B,MAAM,WAAW,CAAC;AAC7E,OAAO,EAAE,aAAa,EAAoB,MAAM,iBAAiB,CAAC;AAElE,MAAM,IAAI,GAAG,kBAAkB,OAAO;;;;;;;;;;;;;;;;;;;CAmBrC,CAAC;AAGF,SAAS,iBAAiB,CAAC,YAAmC;IAC5D,MAAM,QAAQ,GAAG,CAAC,aAAa,CAAC,CAAC;IACjC,IAAI,YAAY,CAAC,QAAQ,CAAC,QAAQ,CAAC,EAAE,CAAC;QACpC,QAAQ,CAAC,IAAI,CAAC,mBAAmB,CAAC,CAAC;IACrC,CAAC;IACD,IAAI,YAAY,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;QACnC,QAAQ,CAAC,IAAI,CAAC,mBAAmB,CAAC,CAAC;IACrC,CAAC;IACD,OAAO,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC7B,CAAC;AAED,SAAS,WAAW,CAClB,GAAoB,EACpB,MAAyB;IAEzB,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACxB,OAAO,GAAG,CAAC;IACb,CAAC;IACD,MAAM,KAAK,GAAG,IAAI,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC;IACpD,KAAK,MAAM,IAAI,IAAI,MAAM,EAAE,CAAC;QAC1B,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC;YACrB,MAAM,IAAI,SAAS,CACjB,iBAAiB,IAAI,iBAAiB,CAAC,GAAG,KAAK,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAC/D,CAAC;QACJ,CAAC;IACH,CAAC;IACD,OAAO,GAAG,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC;AAC1D,CAAC;AAED,KAAK,UAAU,IAAI,CAAC,IAAuB;IACzC,IAAI,OAAoB,CAAC;IACzB,IAAI,CAAC;QACH,OAAO,GAAG,aAAa,CAAC,IAAI,CAAC,CAAC;IAChC,CAAC;IAAC,OAAO,KAAc,EAAE,CAAC;QACxB,OAAO,CAAC,MAAM,CAAC,KAAK,CAClB,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAC9D,CAAC;QACF,OAAO,CAAC,CAAC;IACX,CAAC;IACD,IAAI,OAAO,CAAC,IAAI,EAAE,CAAC;QACjB,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;QAC3B,OAAO,CAAC,CAAC;IACX,CAAC;IAED,IAAI,KAAsB,CAAC;IAC3B,IAAI,CAAC;QACH,KAAK,GAAG,WAAW,CACjB,MAAM,SAAS,CAAC,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,EAChD,OAAO,CAAC,KAAK,CACd,CAAC;IACJ,CAAC;IAAC,OAAO,KAAc,EAAE,CAAC;QACxB,OAAO,CAAC,MAAM,CAAC,KAAK,CAClB,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAC9D,CAAC;QACF,OAAO,CAAC,CAAC;IACX,CAAC;IACD,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACvB,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,qBAAqB,OAAO,CAAC,SAAS,KAAK,CAAC,CAAC;QAClE,OAAO,CAAC,CAAC;IACX,CAAC;IAED,MAAM,UAAU,GAAG,YAAY,CAAC,CAAC,GAAG,OAAO,CAAC,YAAY,CAAC,EAAE,OAAO,CAAC,GAAG,CAAC,CAAC;IACxE,MAAM,GAAG,GAAG,MAAM,sBAAsB,CAAC,UAAU,CAAC,CAAC;IACrD,IAAI,OAAO,GAAG,KAAK,QAAQ,EAAE,CAAC;QAC5B,0EAA0E;QAC1E,iDAAiD;QACjD,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,GAAG,IAAI,CAAC,CAAC;QACjC,OAAO,CAAC,CAAC;IACX,CAAC;IAED,MAAM,SAAS,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC3C,OAAO,CAAC,MAAM,CAAC,KAAK,CAClB,cAAc,UAAU,CAAC,QAAQ,IAAI,UAAU,CAAC,KAAK,IAAI,GAAG,OAAO,KAAK,CAAC,MAAM,eAAe,CAC/F,CAAC;IAEF,MAAM,OAAO,GAAiB,EAAE,CAAC;IACjC,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,IAAI,CAAC;YACH,OAAO,CAAC,IAAI,CACV,MAAM,OAAO,CAAC,IAAI,EAAE;gBAClB,YAAY,EAAE,GAAG,EAAE,CAAC,GAAG;gBACvB,YAAY,EAAE,iBAAiB,CAAC,IAAI,CAAC,YAAY,CAAC;gBAClD,YAAY,EAAE,OAAO,CAAC,YAAY;gBAClC,GAAG,CAAC,OAAO,CAAC,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,OAAO,CAAC,QAAQ,EAAE,CAAC;gBACzE,SAAS,EAAE,CAAC,MAAM,EAAE,EAAE;oBACpB,OAAO,CAAC,MAAM,CAAC,KAAK,CAClB,KAAK,IAAI,CAAC,IAAI,IAAI,MAAM,CAAC,OAAO,KAAK,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,IAAI,CACzE,CAAC;gBACJ,CAAC;aACF,CAAC,CACH,CAAC;QACJ,CAAC;QAAC,OAAO,KAAc,EAAE,CAAC;YACxB,IAAI,KAAK,YAAY,sBAAsB,EAAE,CAAC;gBAC5C,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,YAAY,KAAK,CAAC,OAAO,IAAI,CAAC,CAAC;gBACpD,SAAS;YACX,CAAC;YACD,MAAM,KAAK,CAAC;QACd,CAAC;IACH,CAAC;IAED,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,KAAK,WAAW,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;IAClD,MAAM,QAAQ,GAAG,cAAc,CAAC,OAAO,CAAC,CAAC;IACzC,IAAI,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACxB,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,KAAK,QAAQ,EAAE,CAAC,CAAC;IACxC,CAAC;IAED,IAAI,OAAO,CAAC,QAAQ,KAAK,SAAS,EAAE,CAAC;QACnC,MAAM,MAAM,GAAG,YAAY,CAAC,OAAO,EAAE;YACnC,QAAQ,EAAE,UAAU,CAAC,QAAQ;YAC7B,KAAK,EAAE,UAAU,CAAC,KAAK,IAAI,SAAS;YACpC,OAAO,EAAE,OAAO;YAChB,SAAS;SACV,CAAC,CAAC;QACH,MAAM,SAAS,CAAC,OAAO,CAAC,QAAQ,EAAE,GAAG,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC;QAC1E,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,WAAW,OAAO,CAAC,QAAQ,IAAI,CAAC,CAAC;IACxD,CAAC;IAED,yEAAyE;IACzE,OAAO,CAAC,CAAC;AACX,CAAC;AAED,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;KACxB,IAAI,CAAC,CAAC,QAAQ,EAAE,EAAE;IACjB,OAAO,CAAC,QAAQ,GAAG,QAAQ,CAAC;AAC9B,CAAC,CAAC;KACD,KAAK,CAAC,CAAC,KAAc,EAAE,EAAE;IACxB,MAAM,OAAO,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,eAAe,CAAC;IACzE,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,uBAAuB,OAAO,IAAI,CAAC,CAAC;IACzD,OAAO,CAAC,QAAQ,GAAG,CAAC,CAAC;AACvB,CAAC,CAAC,CAAC"}
1
+ {"version":3,"file":"cli.js","sourceRoot":"","sources":["../../src/evals/cli.ts"],"names":[],"mappings":";AAEA,OAAO,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAC;AAC7C,OAAO,IAAI,MAAM,WAAW,CAAC;AAC7B,OAAO,OAAO,MAAM,cAAc,CAAC;AACnC,OAAO,EAAE,YAAY,EAAE,OAAO,EAAE,MAAM,mBAAmB,CAAC;AAC1D,OAAO,EAAE,sBAAsB,EAAE,MAAM,wBAAwB,CAAC;AAChE,OAAO,EACL,mBAAmB,EACnB,aAAa,EACb,mBAAmB,GACpB,MAAM,eAAe,CAAC;AACvB,OAAO,EAAE,cAAc,EAAE,WAAW,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AACxE,OAAO,EAAE,sBAAsB,EAAE,OAAO,EAAmB,MAAM,aAAa,CAAC;AAC/E,OAAO,EAAE,SAAS,EAAE,SAAS,EAA8B,MAAM,WAAW,CAAC;AAC7E,OAAO,EAAE,aAAa,EAAoB,MAAM,iBAAiB,CAAC;AAElE,MAAM,IAAI,GAAG,kBAAkB,OAAO;;;;;;;;;;;;;;;;;;;;;CAqBrC,CAAC;AAGF,SAAS,iBAAiB,CAAC,YAAmC;IAC5D,MAAM,QAAQ,GAAG,CAAC,aAAa,CAAC,CAAC;IACjC,IAAI,YAAY,CAAC,QAAQ,CAAC,QAAQ,CAAC,EAAE,CAAC;QACpC,QAAQ,CAAC,IAAI,CAAC,mBAAmB,CAAC,CAAC;IACrC,CAAC;IACD,IAAI,YAAY,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;QACnC,QAAQ,CAAC,IAAI,CAAC,mBAAmB,CAAC,CAAC;IACrC,CAAC;IACD,OAAO,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC7B,CAAC;AAED,SAAS,WAAW,CAClB,GAAoB,EACpB,MAAyB;IAEzB,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACxB,OAAO,GAAG,CAAC;IACb,CAAC;IACD,MAAM,KAAK,GAAG,IAAI,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC;IACpD,KAAK,MAAM,IAAI,IAAI,MAAM,EAAE,CAAC;QAC1B,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC;YACrB,MAAM,IAAI,SAAS,CACjB,iBAAiB,IAAI,iBAAiB,CAAC,GAAG,KAAK,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAC/D,CAAC;QACJ,CAAC;IACH,CAAC;IACD,OAAO,GAAG,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC;AAC1D,CAAC;AAED,KAAK,UAAU,IAAI,CAAC,IAAuB;IACzC,IAAI,OAAoB,CAAC;IACzB,IAAI,CAAC;QACH,OAAO,GAAG,aAAa,CAAC,IAAI,CAAC,CAAC;IAChC,CAAC;IAAC,OAAO,KAAc,EAAE,CAAC;QACxB,OAAO,CAAC,MAAM,CAAC,KAAK,CAClB,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAC9D,CAAC;QACF,OAAO,CAAC,CAAC;IACX,CAAC;IACD,IAAI,OAAO,CAAC,IAAI,EAAE,CAAC;QACjB,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;QAC3B,OAAO,CAAC,CAAC;IACX,CAAC;IAED,IAAI,KAAsB,CAAC;IAC3B,IAAI,CAAC;QACH,KAAK,GAAG,WAAW,CACjB,MAAM,SAAS,CAAC,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,EAChD,OAAO,CAAC,KAAK,CACd,CAAC;IACJ,CAAC;IAAC,OAAO,KAAc,EAAE,CAAC;QACxB,OAAO,CAAC,MAAM,CAAC,KAAK,CAClB,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAC9D,CAAC;QACF,OAAO,CAAC,CAAC;IACX,CAAC;IACD,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACvB,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,qBAAqB,OAAO,CAAC,SAAS,KAAK,CAAC,CAAC;QAClE,OAAO,CAAC,CAAC;IACX,CAAC;IAED,MAAM,UAAU,GAAG,YAAY,CAAC,CAAC,GAAG,OAAO,CAAC,YAAY,CAAC,EAAE,OAAO,CAAC,GAAG,CAAC,CAAC;IACxE,MAAM,GAAG,GAAG,MAAM,sBAAsB,CAAC,UAAU,CAAC,CAAC;IACrD,IAAI,OAAO,GAAG,KAAK,QAAQ,EAAE,CAAC;QAC5B,0EAA0E;QAC1E,iDAAiD;QACjD,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,GAAG,IAAI,CAAC,CAAC;QACjC,OAAO,CAAC,CAAC;IACX,CAAC;IAED,MAAM,SAAS,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC3C,OAAO,CAAC,MAAM,CAAC,KAAK,CAClB,cAAc,UAAU,CAAC,QAAQ,IAAI,UAAU,CAAC,KAAK,IAAI,GAAG,OAAO,KAAK,CAAC,MAAM,eAAe,CAC/F,CAAC;IAEF,MAAM,OAAO,GAAiB,EAAE,CAAC;IACjC,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,IAAI,CAAC;YACH,OAAO,CAAC,IAAI,CACV,MAAM,OAAO,CAAC,IAAI,EAAE;gBAClB,YAAY,EAAE,GAAG,EAAE,CAAC,GAAG;gBACvB,YAAY,EAAE,iBAAiB,CAAC,IAAI,CAAC,YAAY,CAAC;gBAClD,YAAY,EAAE,OAAO,CAAC,YAAY;gBAClC,GAAG,CAAC,OAAO,CAAC,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,OAAO,CAAC,QAAQ,EAAE,CAAC;gBACzE,SAAS,EAAE,CAAC,MAAM,EAAE,EAAE;oBACpB,OAAO,CAAC,MAAM,CAAC,KAAK,CAClB,KAAK,IAAI,CAAC,IAAI,IAAI,MAAM,CAAC,OAAO,KAAK,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,IAAI,CACzE,CAAC;gBACJ,CAAC;aACF,CAAC,CACH,CAAC;QACJ,CAAC;QAAC,OAAO,KAAc,EAAE,CAAC;YACxB,IAAI,KAAK,YAAY,sBAAsB,EAAE,CAAC;gBAC5C,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,YAAY,KAAK,CAAC,OAAO,IAAI,CAAC,CAAC;gBACpD,SAAS;YACX,CAAC;YACD,MAAM,KAAK,CAAC;QACd,CAAC;IACH,CAAC;IAED,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,KAAK,WAAW,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;IAClD,MAAM,QAAQ,GAAG,cAAc,CAAC,OAAO,CAAC,CAAC;IACzC,IAAI,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACxB,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,KAAK,QAAQ,EAAE,CAAC,CAAC;IACxC,CAAC;IAED,IAAI,OAAO,CAAC,QAAQ,KAAK,SAAS,EAAE,CAAC;QACnC,MAAM,MAAM,GAAG,YAAY,CAAC,OAAO,EAAE;YACnC,QAAQ,EAAE,UAAU,CAAC,QAAQ;YAC7B,KAAK,EAAE,UAAU,CAAC,KAAK,IAAI,SAAS;YACpC,OAAO,EAAE,OAAO;YAChB,SAAS;SACV,CAAC,CAAC;QACH,MAAM,SAAS,CAAC,OAAO,CAAC,QAAQ,EAAE,GAAG,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC;QAC1E,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,WAAW,OAAO,CAAC,QAAQ,IAAI,CAAC,CAAC;IACxD,CAAC;IAED,yEAAyE;IACzE,OAAO,CAAC,CAAC;AACX,CAAC;AAED,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;KACxB,IAAI,CAAC,CAAC,QAAQ,EAAE,EAAE;IACjB,OAAO,CAAC,QAAQ,GAAG,QAAQ,CAAC;AAC9B,CAAC,CAAC;KACD,KAAK,CAAC,CAAC,KAAc,EAAE,EAAE;IACxB,MAAM,OAAO,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,eAAe,CAAC;IACzE,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,uBAAuB,OAAO,IAAI,CAAC,CAAC;IACzD,OAAO,CAAC,QAAQ,GAAG,CAAC,CAAC;AACvB,CAAC,CAAC,CAAC"}
@@ -3,6 +3,8 @@ import type { AgentEvent } from "../events.js";
3
3
  export interface AttemptFacts {
4
4
  /** The workspace as the attempt left it. */
5
5
  readonly workspace: string;
6
+ /** The pristine fixture, for graders that ask what the attempt changed. */
7
+ readonly fixtureDirectory: string;
6
8
  readonly finalAnswer: string;
7
9
  readonly turnCount: number;
8
10
  readonly status: "completed" | "max_turns" | "aborted" | "error";
@@ -1 +1 @@
1
- {"version":3,"file":"graders.d.ts","sourceRoot":"","sources":["../../src/evals/graders.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,WAAW,CAAC;AACvD,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,cAAc,CAAC;AAE/C,MAAM,WAAW,YAAY;IAC3B,4CAA4C;IAC5C,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,MAAM,EAAE,WAAW,GAAG,WAAW,GAAG,SAAS,GAAG,OAAO,CAAC;IACjE,QAAQ,CAAC,MAAM,EAAE,SAAS,UAAU,EAAE,CAAC;CACxC;AAED,MAAM,WAAW,YAAY;IAC3B,QAAQ,CAAC,MAAM,EAAE,OAAO,CAAC;IACzB,sEAAsE;IACtE,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;CACzB;AAED,MAAM,MAAM,MAAM,GAAG,CAAC,KAAK,EAAE,YAAY,KAAK,OAAO,CAAC,YAAY,CAAC,CAAC;AA4LpE,eAAO,MAAM,YAAY,EAAE,SAAS,MAAM,EAA2B,CAAC;AAEtE,wBAAgB,YAAY,CAAC,IAAI,EAAE,UAAU,GAAG,MAAM,CAQrD;AAED,wBAAgB,cAAc,CAAC,IAAI,EAAE,UAAU,GAAG,MAAM,CAMvD;AAED,YAAY,EAAE,SAAS,EAAE,CAAC"}
1
+ {"version":3,"file":"graders.d.ts","sourceRoot":"","sources":["../../src/evals/graders.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,WAAW,CAAC;AACvD,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,cAAc,CAAC;AAE/C,MAAM,WAAW,YAAY;IAC3B,4CAA4C;IAC5C,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,2EAA2E;IAC3E,QAAQ,CAAC,gBAAgB,EAAE,MAAM,CAAC;IAClC,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,MAAM,EAAE,WAAW,GAAG,WAAW,GAAG,SAAS,GAAG,OAAO,CAAC;IACjE,QAAQ,CAAC,MAAM,EAAE,SAAS,UAAU,EAAE,CAAC;CACxC;AAED,MAAM,WAAW,YAAY;IAC3B,QAAQ,CAAC,MAAM,EAAE,OAAO,CAAC;IACzB,sEAAsE;IACtE,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;CACzB;AAED,MAAM,MAAM,MAAM,GAAG,CAAC,KAAK,EAAE,YAAY,KAAK,OAAO,CAAC,YAAY,CAAC,CAAC;AAgRpE,eAAO,MAAM,YAAY,EAAE,SAAS,MAAM,EAA2B,CAAC;AAEtE,wBAAgB,YAAY,CAAC,IAAI,EAAE,UAAU,GAAG,MAAM,CAQrD;AAED,wBAAgB,cAAc,CAAC,IAAI,EAAE,UAAU,GAAG,MAAM,CAMvD;AAED,YAAY,EAAE,SAAS,EAAE,CAAC"}
@@ -25,12 +25,29 @@ function toolCalls(events) {
25
25
  .filter((event) => event.type === "tool_execution_end")
26
26
  .map((event) => event.toolName);
27
27
  }
28
+ function failedToolCalls(events) {
29
+ return events
30
+ .filter((event) => event.type === "tool_execution_end" &&
31
+ event.isError === true)
32
+ .map((event) => event.toolName);
33
+ }
28
34
  function successfulToolCalls(events) {
29
35
  return events
30
36
  .filter((event) => event.type === "tool_execution_end" &&
31
37
  event.isError === false)
32
38
  .map((event) => event.toolName);
33
39
  }
40
+ /** 1-based line where two texts first differ, for an actionable failure. */
41
+ function firstDifferingLine(before, after) {
42
+ const left = before.split("\n");
43
+ const right = after.split("\n");
44
+ for (let index = 0; index < Math.max(left.length, right.length); index += 1) {
45
+ if (left[index] !== right[index]) {
46
+ return index + 1;
47
+ }
48
+ }
49
+ return 1;
50
+ }
34
51
  const factories = {
35
52
  "file-contains": (spec) => {
36
53
  const file = requireString(spec, "path");
@@ -125,6 +142,59 @@ const factories = {
125
142
  : fail(`${name} never returned a successful result`);
126
143
  };
127
144
  },
145
+ "tool-never-failed": (spec) => {
146
+ // Sound only where a failure means the model was wrong. `bash` sets
147
+ // isError on any non-zero exit, so on a task that runs a failing test —
148
+ // fix-failing-test, wrong-test — the correct behaviour is a failed call
149
+ // and this grader would punish it.
150
+ const name = requireString(spec, "name");
151
+ return async ({ events }) => {
152
+ // A guessed path shows up here: reading a file that is not there is the
153
+ // failure, and the set of tool names a report keeps cannot see it.
154
+ const failures = failedToolCalls(events).filter((tool) => tool === name);
155
+ return failures.length === 0
156
+ ? pass
157
+ : fail(`${name} failed ${failures.length} time(s); this task expects it to be called only on paths the model established exist`);
158
+ };
159
+ },
160
+ "max-tool-calls": (spec) => {
161
+ const name = requireString(spec, "name");
162
+ const limit = spec.count;
163
+ if (!Number.isSafeInteger(limit) || limit < 1) {
164
+ throw new Error('max-tool-calls: "count" must be a positive integer.');
165
+ }
166
+ return async ({ events }) => {
167
+ // Turns cannot express this: a model may issue any number of calls in
168
+ // one turn, so "did it read the whole project" is a call count, not a
169
+ // turn count.
170
+ // Failed calls count. A budget that forgave them would let a model
171
+ // spend freely on guessed paths, which is the opposite of the intent.
172
+ const used = toolCalls(events).filter((tool) => tool === name).length;
173
+ return used <= limit
174
+ ? pass
175
+ : fail(`called ${name} ${used} times, over the budget of ${limit}`);
176
+ };
177
+ },
178
+ "file-unchanged": (spec) => {
179
+ const file = requireString(spec, "path");
180
+ return async ({ workspace, fixtureDirectory }) => {
181
+ // Compared against the fixture rather than against text copied into
182
+ // task.json, so there is no second copy to keep in sync.
183
+ const before = await readWorkspaceFile(fixtureDirectory, file);
184
+ if (before === undefined) {
185
+ throw new Error(`file-unchanged: ${file} is not in the fixture, so there is nothing to compare against.`);
186
+ }
187
+ const after = await readWorkspaceFile(workspace, file);
188
+ if (after === undefined) {
189
+ return fail(`${file} was deleted; this task expects it untouched`);
190
+ }
191
+ if (after === before) {
192
+ return pass;
193
+ }
194
+ const line = firstDifferingLine(before, after);
195
+ return fail(`${file} was modified at line ${line}; this task expects it untouched`);
196
+ };
197
+ },
128
198
  "answer-matches": (spec) => {
129
199
  const pattern = requireString(spec, "pattern");
130
200
  const flags = typeof spec.flags === "string" ? spec.flags : "i";
@@ -1 +1 @@
1
- {"version":3,"file":"graders.js","sourceRoot":"","sources":["../../src/evals/graders.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,kBAAkB,CAAC;AAClD,OAAO,IAAI,MAAM,WAAW,CAAC;AAqB7B,MAAM,IAAI,GAAiB,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,EAAE,EAAE,CAAC;AAExD,SAAS,IAAI,CAAC,MAAc;IAC1B,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC;AACnC,CAAC;AAED,SAAS,aAAa,CAAC,IAAgB,EAAE,KAAa;IACpD,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC;IAC1B,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACpD,MAAM,IAAI,KAAK,CAAC,GAAG,IAAI,CAAC,IAAI,MAAM,KAAK,+BAA+B,CAAC,CAAC;IAC1E,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED,KAAK,UAAU,iBAAiB,CAC9B,SAAiB,EACjB,YAAoB;IAEpB,IAAI,CAAC;QACH,OAAO,MAAM,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,YAAY,CAAC,EAAE,MAAM,CAAC,CAAC;IACpE,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,SAAS,CAAC;IACnB,CAAC;AACH,CAAC;AAED,yEAAyE;AACzE,SAAS,SAAS,CAAC,MAA6B;IAC9C,OAAO,MAAM;SACV,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,KAAK,oBAAoB,CAAC;SACtD,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAE,KAA8B,CAAC,QAAQ,CAAC,CAAC;AAC9D,CAAC;AAED,SAAS,mBAAmB,CAAC,MAA6B;IACxD,OAAO,MAAM;SACV,MAAM,CACL,CAAC,KAAK,EAAE,EAAE,CACR,KAAK,CAAC,IAAI,KAAK,oBAAoB;QAClC,KAA8B,CAAC,OAAO,KAAK,KAAK,CACpD;SACA,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAE,KAA8B,CAAC,QAAQ,CAAC,CAAC;AAC9D,CAAC;AAID,MAAM,SAAS,GAAkC;IAC/C,eAAe,EAAE,CAAC,IAAI,EAAE,EAAE;QACxB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,SAAS,EAAE,EAAE,EAAE;YAC7B,MAAM,QAAQ,GAAG,MAAM,iBAAiB,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC;YAC1D,IAAI,QAAQ,KAAK,SAAS,EAAE,CAAC;gBAC3B,OAAO,IAAI,CAAC,GAAG,IAAI,iBAAiB,CAAC,CAAC;YACxC,CAAC;YACD,OAAO,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC;gBAC5B,CAAC,CAAC,IAAI;gBACN,CAAC,CAAC,IAAI,CAAC,GAAG,IAAI,qBAAqB,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QAC/D,CAAC,CAAC;IACJ,CAAC;IAED,eAAe,EAAE,CAAC,IAAI,EAAE,EAAE;QACxB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,SAAS,EAAE,EAAE,EAAE;YAC7B,MAAM,QAAQ,GAAG,MAAM,iBAAiB,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC;YAC1D,IAAI,QAAQ,KAAK,SAAS,EAAE,CAAC;gBAC3B,OAAO,IAAI,CAAC,GAAG,IAAI,iBAAiB,CAAC,CAAC;YACxC,CAAC;YACD,IAAI,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC,EAAE,CAAC;gBAC7B,OAAO,IAAI,CAAC;YACd,CAAC;YACD,yEAAyE;YACzE,MAAM,IAAI,GACR,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,QAAQ,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC;YAC/D,OAAO,IAAI,CACT,GAAG,IAAI,mBAAmB,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,YAAY,IAAI,EAAE,CACjE,CAAC;QACJ,CAAC,CAAC;IACJ,CAAC;IAED,aAAa,EAAE,CAAC,IAAI,EAAE,EAAE;QACtB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,SAAS,EAAE,EAAE,EAAE;YAC7B,IAAI,CAAC;gBACH,MAAM,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC,CAAC;gBACvC,OAAO,IAAI,CAAC;YACd,CAAC;YAAC,MAAM,CAAC;gBACP,OAAO,IAAI,CAAC,GAAG,IAAI,kBAAkB,CAAC,CAAC;YACzC,CAAC;QACH,CAAC,CAAC;IACJ,CAAC;IAED,aAAa,EAAE,CAAC,IAAI,EAAE,EAAE;QACtB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,SAAS,EAAE,EAAE,EAAE;YAC7B,IAAI,CAAC;gBACH,MAAM,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC,CAAC;gBACvC,OAAO,IAAI,CAAC,GAAG,IAAI,eAAe,CAAC,CAAC;YACtC,CAAC;YAAC,MAAM,CAAC;gBACP,OAAO,IAAI,CAAC;YACd,CAAC;QACH,CAAC,CAAC;IACJ,CAAC;IAED,aAAa,EAAE,CAAC,IAAI,EAAE,EAAE;QACtB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,SAAS,EAAE,EAAE,EAAE;YAC7B,MAAM,QAAQ,GAAG,MAAM,iBAAiB,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC;YAC1D,IAAI,QAAQ,KAAK,SAAS,EAAE,CAAC;gBAC3B,OAAO,IAAI,CAAC,GAAG,IAAI,iBAAiB,CAAC,CAAC;YACxC,CAAC;YACD,OAAO,QAAQ,KAAK,IAAI;gBACtB,CAAC,CAAC,IAAI;gBACN,CAAC,CAAC,IAAI,CAAC,GAAG,IAAI,qDAAqD,CAAC,CAAC;QACzE,CAAC,CAAC;IACJ,CAAC;IAED,WAAW,EAAE,CAAC,IAAI,EAAE,EAAE;QACpB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE;YAC1B,MAAM,IAAI,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC;YAC/B,OAAO,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC;gBACxB,CAAC,CAAC,IAAI;gBACN,CAAC,CAAC,IAAI,CACF,gBAAgB,IAAI,YAAY,IAAI,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAChG,CAAC;QACR,CAAC,CAAC;IACJ,CAAC;IAED,eAAe,EAAE,CAAC,IAAI,EAAE,EAAE;QACxB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE;YAC1B,MAAM,IAAI,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC;YAC/B,MAAM,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,KAAK,IAAI,CAAC,CAAC,MAAM,CAAC;YAC1D,OAAO,KAAK,KAAK,CAAC;gBAChB,CAAC,CAAC,IAAI;gBACN,CAAC,CAAC,IAAI,CAAC,UAAU,IAAI,IAAI,KAAK,mCAAmC,CAAC,CAAC;QACvE,CAAC,CAAC;IACJ,CAAC;IAED,gBAAgB,EAAE,CAAC,IAAI,EAAE,EAAE;QACzB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE;YAC1B,OAAO,mBAAmB,CAAC,MAAM,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC;gBAC/C,CAAC,CAAC,IAAI;gBACN,CAAC,CAAC,IAAI,CAAC,GAAG,IAAI,qCAAqC,CAAC,CAAC;QACzD,CAAC,CAAC;IACJ,CAAC;IAED,gBAAgB,EAAE,CAAC,IAAI,EAAE,EAAE;QACzB,MAAM,OAAO,GAAG,aAAa,CAAC,IAAI,EAAE,SAAS,CAAC,CAAC;QAC/C,MAAM,KAAK,GAAG,OAAO,IAAI,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC;QAChE,MAAM,UAAU,GAAG,IAAI,MAAM,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC;QAC9C,OAAO,KAAK,EAAE,EAAE,WAAW,EAAE,EAAE,EAAE;YAC/B,OAAO,UAAU,CAAC,IAAI,CAAC,WAAW,CAAC;gBACjC,CAAC,CAAC,IAAI;gBACN,CAAC,CAAC,IAAI,CACF,6BAA6B,OAAO,IAAI,KAAK,KAAK,IAAI,CAAC,SAAS,CAAC,WAAW,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,EAAE,CAC9F,CAAC;QACR,CAAC,CAAC;IACJ,CAAC;IAED,iBAAiB,EAAE,CAAC,IAAI,EAAE,EAAE;QAC1B,MAAM,OAAO,GAAG,aAAa,CAAC,IAAI,EAAE,SAAS,CAAC,CAAC;QAC/C,MAAM,KAAK,GAAG,OAAO,IAAI,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC;QAChE,MAAM,UAAU,GAAG,IAAI,MAAM,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC;QAC9C,OAAO,KAAK,EAAE,EAAE,WAAW,EAAE,EAAE,EAAE;YAC/B,OAAO,UAAU,CAAC,IAAI,CAAC,WAAW,CAAC;gBACjC,CAAC,CAAC,IAAI,CAAC,uBAAuB,OAAO,IAAI,KAAK,2BAA2B,CAAC;gBAC1E,CAAC,CAAC,IAAI,CAAC;QACX,CAAC,CAAC;IACJ,CAAC;IAED,iBAAiB,EAAE,CAAC,IAAI,EAAE,EAAE;QAC1B,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC;QACzB,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,IAAK,KAAgB,GAAG,CAAC,EAAE,CAAC;YAC1D,MAAM,IAAI,KAAK,CAAC,wDAAwD,CAAC,CAAC;QAC5E,CAAC;QACD,OAAO,KAAK,EAAE,EAAE,SAAS,EAAE,EAAE,EAAE;YAC7B,OAAO,SAAS,IAAK,KAAgB;gBACnC,CAAC,CAAC,IAAI;gBACN,CAAC,CAAC,IAAI,CAAC,QAAQ,SAAS,6BAA6B,KAAe,EAAE,CAAC,CAAC;QAC5E,CAAC,CAAC;IACJ,CAAC;CACF,CAAC;AAEF,MAAM,CAAC,MAAM,YAAY,GAAsB,MAAM,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;AAEtE,MAAM,UAAU,YAAY,CAAC,IAAgB;IAC3C,MAAM,OAAO,GAAG,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACrC,IAAI,OAAO,KAAK,SAAS,EAAE,CAAC;QAC1B,MAAM,IAAI,KAAK,CACb,wBAAwB,IAAI,CAAC,IAAI,mBAAmB,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAC/E,CAAC;IACJ,CAAC;IACD,OAAO,OAAO,CAAC,IAAI,CAAC,CAAC;AACvB,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,IAAgB;IAC7C,MAAM,MAAM,GAAG,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC;SAChC,MAAM,CAAC,CAAC,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC,GAAG,KAAK,MAAM,CAAC;SACjC,GAAG,CAAC,CAAC,CAAC,GAAG,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,IAAI,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,EAAE,CAAC;SACxD,IAAI,CAAC,GAAG,CAAC,CAAC;IACb,OAAO,MAAM,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,IAAI,IAAI,MAAM,EAAE,CAAC;AACpE,CAAC"}
1
+ {"version":3,"file":"graders.js","sourceRoot":"","sources":["../../src/evals/graders.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,kBAAkB,CAAC;AAClD,OAAO,IAAI,MAAM,WAAW,CAAC;AAuB7B,MAAM,IAAI,GAAiB,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,EAAE,EAAE,CAAC;AAExD,SAAS,IAAI,CAAC,MAAc;IAC1B,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC;AACnC,CAAC;AAED,SAAS,aAAa,CAAC,IAAgB,EAAE,KAAa;IACpD,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC;IAC1B,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACpD,MAAM,IAAI,KAAK,CAAC,GAAG,IAAI,CAAC,IAAI,MAAM,KAAK,+BAA+B,CAAC,CAAC;IAC1E,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED,KAAK,UAAU,iBAAiB,CAC9B,SAAiB,EACjB,YAAoB;IAEpB,IAAI,CAAC;QACH,OAAO,MAAM,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,YAAY,CAAC,EAAE,MAAM,CAAC,CAAC;IACpE,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,SAAS,CAAC;IACnB,CAAC;AACH,CAAC;AAED,yEAAyE;AACzE,SAAS,SAAS,CAAC,MAA6B;IAC9C,OAAO,MAAM;SACV,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,KAAK,oBAAoB,CAAC;SACtD,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAE,KAA8B,CAAC,QAAQ,CAAC,CAAC;AAC9D,CAAC;AAED,SAAS,eAAe,CAAC,MAA6B;IACpD,OAAO,MAAM;SACV,MAAM,CACL,CAAC,KAAK,EAAE,EAAE,CACR,KAAK,CAAC,IAAI,KAAK,oBAAoB;QAClC,KAA8B,CAAC,OAAO,KAAK,IAAI,CACnD;SACA,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAE,KAA8B,CAAC,QAAQ,CAAC,CAAC;AAC9D,CAAC;AAED,SAAS,mBAAmB,CAAC,MAA6B;IACxD,OAAO,MAAM;SACV,MAAM,CACL,CAAC,KAAK,EAAE,EAAE,CACR,KAAK,CAAC,IAAI,KAAK,oBAAoB;QAClC,KAA8B,CAAC,OAAO,KAAK,KAAK,CACpD;SACA,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAE,KAA8B,CAAC,QAAQ,CAAC,CAAC;AAC9D,CAAC;AAED,4EAA4E;AAC5E,SAAS,kBAAkB,CAAC,MAAc,EAAE,KAAa;IACvD,MAAM,IAAI,GAAG,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAChC,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAChC,KAAK,IAAI,KAAK,GAAG,CAAC,EAAE,KAAK,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE,KAAK,IAAI,CAAC,EAAE,CAAC;QAC5E,IAAI,IAAI,CAAC,KAAK,CAAC,KAAK,KAAK,CAAC,KAAK,CAAC,EAAE,CAAC;YACjC,OAAO,KAAK,GAAG,CAAC,CAAC;QACnB,CAAC;IACH,CAAC;IACD,OAAO,CAAC,CAAC;AACX,CAAC;AAID,MAAM,SAAS,GAAkC;IAC/C,eAAe,EAAE,CAAC,IAAI,EAAE,EAAE;QACxB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,SAAS,EAAE,EAAE,EAAE;YAC7B,MAAM,QAAQ,GAAG,MAAM,iBAAiB,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC;YAC1D,IAAI,QAAQ,KAAK,SAAS,EAAE,CAAC;gBAC3B,OAAO,IAAI,CAAC,GAAG,IAAI,iBAAiB,CAAC,CAAC;YACxC,CAAC;YACD,OAAO,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC;gBAC5B,CAAC,CAAC,IAAI;gBACN,CAAC,CAAC,IAAI,CAAC,GAAG,IAAI,qBAAqB,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QAC/D,CAAC,CAAC;IACJ,CAAC;IAED,eAAe,EAAE,CAAC,IAAI,EAAE,EAAE;QACxB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,SAAS,EAAE,EAAE,EAAE;YAC7B,MAAM,QAAQ,GAAG,MAAM,iBAAiB,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC;YAC1D,IAAI,QAAQ,KAAK,SAAS,EAAE,CAAC;gBAC3B,OAAO,IAAI,CAAC,GAAG,IAAI,iBAAiB,CAAC,CAAC;YACxC,CAAC;YACD,IAAI,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC,EAAE,CAAC;gBAC7B,OAAO,IAAI,CAAC;YACd,CAAC;YACD,yEAAyE;YACzE,MAAM,IAAI,GACR,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,QAAQ,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC;YAC/D,OAAO,IAAI,CACT,GAAG,IAAI,mBAAmB,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,YAAY,IAAI,EAAE,CACjE,CAAC;QACJ,CAAC,CAAC;IACJ,CAAC;IAED,aAAa,EAAE,CAAC,IAAI,EAAE,EAAE;QACtB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,SAAS,EAAE,EAAE,EAAE;YAC7B,IAAI,CAAC;gBACH,MAAM,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC,CAAC;gBACvC,OAAO,IAAI,CAAC;YACd,CAAC;YAAC,MAAM,CAAC;gBACP,OAAO,IAAI,CAAC,GAAG,IAAI,kBAAkB,CAAC,CAAC;YACzC,CAAC;QACH,CAAC,CAAC;IACJ,CAAC;IAED,aAAa,EAAE,CAAC,IAAI,EAAE,EAAE;QACtB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,SAAS,EAAE,EAAE,EAAE;YAC7B,IAAI,CAAC;gBACH,MAAM,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC,CAAC;gBACvC,OAAO,IAAI,CAAC,GAAG,IAAI,eAAe,CAAC,CAAC;YACtC,CAAC;YAAC,MAAM,CAAC;gBACP,OAAO,IAAI,CAAC;YACd,CAAC;QACH,CAAC,CAAC;IACJ,CAAC;IAED,aAAa,EAAE,CAAC,IAAI,EAAE,EAAE;QACtB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,SAAS,EAAE,EAAE,EAAE;YAC7B,MAAM,QAAQ,GAAG,MAAM,iBAAiB,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC;YAC1D,IAAI,QAAQ,KAAK,SAAS,EAAE,CAAC;gBAC3B,OAAO,IAAI,CAAC,GAAG,IAAI,iBAAiB,CAAC,CAAC;YACxC,CAAC;YACD,OAAO,QAAQ,KAAK,IAAI;gBACtB,CAAC,CAAC,IAAI;gBACN,CAAC,CAAC,IAAI,CAAC,GAAG,IAAI,qDAAqD,CAAC,CAAC;QACzE,CAAC,CAAC;IACJ,CAAC;IAED,WAAW,EAAE,CAAC,IAAI,EAAE,EAAE;QACpB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE;YAC1B,MAAM,IAAI,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC;YAC/B,OAAO,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC;gBACxB,CAAC,CAAC,IAAI;gBACN,CAAC,CAAC,IAAI,CACF,gBAAgB,IAAI,YAAY,IAAI,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAChG,CAAC;QACR,CAAC,CAAC;IACJ,CAAC;IAED,eAAe,EAAE,CAAC,IAAI,EAAE,EAAE;QACxB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE;YAC1B,MAAM,IAAI,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC;YAC/B,MAAM,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,KAAK,IAAI,CAAC,CAAC,MAAM,CAAC;YAC1D,OAAO,KAAK,KAAK,CAAC;gBAChB,CAAC,CAAC,IAAI;gBACN,CAAC,CAAC,IAAI,CAAC,UAAU,IAAI,IAAI,KAAK,mCAAmC,CAAC,CAAC;QACvE,CAAC,CAAC;IACJ,CAAC;IAED,gBAAgB,EAAE,CAAC,IAAI,EAAE,EAAE;QACzB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE;YAC1B,OAAO,mBAAmB,CAAC,MAAM,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC;gBAC/C,CAAC,CAAC,IAAI;gBACN,CAAC,CAAC,IAAI,CAAC,GAAG,IAAI,qCAAqC,CAAC,CAAC;QACzD,CAAC,CAAC;IACJ,CAAC;IAED,mBAAmB,EAAE,CAAC,IAAI,EAAE,EAAE;QAC5B,oEAAoE;QACpE,wEAAwE;QACxE,wEAAwE;QACxE,mCAAmC;QACnC,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE;YAC1B,wEAAwE;YACxE,mEAAmE;YACnE,MAAM,QAAQ,GAAG,eAAe,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,KAAK,IAAI,CAAC,CAAC;YACzE,OAAO,QAAQ,CAAC,MAAM,KAAK,CAAC;gBAC1B,CAAC,CAAC,IAAI;gBACN,CAAC,CAAC,IAAI,CACF,GAAG,IAAI,WAAW,QAAQ,CAAC,MAAM,uFAAuF,CACzH,CAAC;QACR,CAAC,CAAC;IACJ,CAAC;IAED,gBAAgB,EAAE,CAAC,IAAI,EAAE,EAAE;QACzB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC;QACzB,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,IAAK,KAAgB,GAAG,CAAC,EAAE,CAAC;YAC1D,MAAM,IAAI,KAAK,CAAC,qDAAqD,CAAC,CAAC;QACzE,CAAC;QACD,OAAO,KAAK,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE;YAC1B,sEAAsE;YACtE,sEAAsE;YACtE,cAAc;YACd,mEAAmE;YACnE,sEAAsE;YACtE,MAAM,IAAI,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,KAAK,IAAI,CAAC,CAAC,MAAM,CAAC;YACtE,OAAO,IAAI,IAAK,KAAgB;gBAC9B,CAAC,CAAC,IAAI;gBACN,CAAC,CAAC,IAAI,CAAC,UAAU,IAAI,IAAI,IAAI,8BAA8B,KAAe,EAAE,CAAC,CAAC;QAClF,CAAC,CAAC;IACJ,CAAC;IAED,gBAAgB,EAAE,CAAC,IAAI,EAAE,EAAE;QACzB,MAAM,IAAI,GAAG,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QACzC,OAAO,KAAK,EAAE,EAAE,SAAS,EAAE,gBAAgB,EAAE,EAAE,EAAE;YAC/C,oEAAoE;YACpE,yDAAyD;YACzD,MAAM,MAAM,GAAG,MAAM,iBAAiB,CAAC,gBAAgB,EAAE,IAAI,CAAC,CAAC;YAC/D,IAAI,MAAM,KAAK,SAAS,EAAE,CAAC;gBACzB,MAAM,IAAI,KAAK,CACb,mBAAmB,IAAI,iEAAiE,CACzF,CAAC;YACJ,CAAC;YACD,MAAM,KAAK,GAAG,MAAM,iBAAiB,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC;YACvD,IAAI,KAAK,KAAK,SAAS,EAAE,CAAC;gBACxB,OAAO,IAAI,CAAC,GAAG,IAAI,8CAA8C,CAAC,CAAC;YACrE,CAAC;YACD,IAAI,KAAK,KAAK,MAAM,EAAE,CAAC;gBACrB,OAAO,IAAI,CAAC;YACd,CAAC;YACD,MAAM,IAAI,GAAG,kBAAkB,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC;YAC/C,OAAO,IAAI,CACT,GAAG,IAAI,yBAAyB,IAAI,kCAAkC,CACvE,CAAC;QACJ,CAAC,CAAC;IACJ,CAAC;IAED,gBAAgB,EAAE,CAAC,IAAI,EAAE,EAAE;QACzB,MAAM,OAAO,GAAG,aAAa,CAAC,IAAI,EAAE,SAAS,CAAC,CAAC;QAC/C,MAAM,KAAK,GAAG,OAAO,IAAI,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC;QAChE,MAAM,UAAU,GAAG,IAAI,MAAM,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC;QAC9C,OAAO,KAAK,EAAE,EAAE,WAAW,EAAE,EAAE,EAAE;YAC/B,OAAO,UAAU,CAAC,IAAI,CAAC,WAAW,CAAC;gBACjC,CAAC,CAAC,IAAI;gBACN,CAAC,CAAC,IAAI,CACF,6BAA6B,OAAO,IAAI,KAAK,KAAK,IAAI,CAAC,SAAS,CAAC,WAAW,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,EAAE,CAC9F,CAAC;QACR,CAAC,CAAC;IACJ,CAAC;IAED,iBAAiB,EAAE,CAAC,IAAI,EAAE,EAAE;QAC1B,MAAM,OAAO,GAAG,aAAa,CAAC,IAAI,EAAE,SAAS,CAAC,CAAC;QAC/C,MAAM,KAAK,GAAG,OAAO,IAAI,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC;QAChE,MAAM,UAAU,GAAG,IAAI,MAAM,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC;QAC9C,OAAO,KAAK,EAAE,EAAE,WAAW,EAAE,EAAE,EAAE;YAC/B,OAAO,UAAU,CAAC,IAAI,CAAC,WAAW,CAAC;gBACjC,CAAC,CAAC,IAAI,CAAC,uBAAuB,OAAO,IAAI,KAAK,2BAA2B,CAAC;gBAC1E,CAAC,CAAC,IAAI,CAAC;QACX,CAAC,CAAC;IACJ,CAAC;IAED,iBAAiB,EAAE,CAAC,IAAI,EAAE,EAAE;QAC1B,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC;QACzB,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,IAAK,KAAgB,GAAG,CAAC,EAAE,CAAC;YAC1D,MAAM,IAAI,KAAK,CAAC,wDAAwD,CAAC,CAAC;QAC5E,CAAC;QACD,OAAO,KAAK,EAAE,EAAE,SAAS,EAAE,EAAE,EAAE;YAC7B,OAAO,SAAS,IAAK,KAAgB;gBACnC,CAAC,CAAC,IAAI;gBACN,CAAC,CAAC,IAAI,CAAC,QAAQ,SAAS,6BAA6B,KAAe,EAAE,CAAC,CAAC;QAC5E,CAAC,CAAC;IACJ,CAAC;CACF,CAAC;AAEF,MAAM,CAAC,MAAM,YAAY,GAAsB,MAAM,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;AAEtE,MAAM,UAAU,YAAY,CAAC,IAAgB;IAC3C,MAAM,OAAO,GAAG,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACrC,IAAI,OAAO,KAAK,SAAS,EAAE,CAAC;QAC1B,MAAM,IAAI,KAAK,CACb,wBAAwB,IAAI,CAAC,IAAI,mBAAmB,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAC/E,CAAC;IACJ,CAAC;IACD,OAAO,OAAO,CAAC,IAAI,CAAC,CAAC;AACvB,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,IAAgB;IAC7C,MAAM,MAAM,GAAG,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC;SAChC,MAAM,CAAC,CAAC,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC,GAAG,KAAK,MAAM,CAAC;SACjC,GAAG,CAAC,CAAC,CAAC,GAAG,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,IAAI,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,EAAE,CAAC;SACxD,IAAI,CAAC,GAAG,CAAC,CAAC;IACb,OAAO,MAAM,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,IAAI,IAAI,MAAM,EAAE,CAAC;AACpE,CAAC"}
@@ -1 +1 @@
1
- {"version":3,"file":"parse-args.d.ts","sourceRoot":"","sources":["../../src/evals/parse-args.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,WAAW,CAAC;AAE5C;;;;;GAKG;AACH,MAAM,WAAW,WAAW;IAC1B,QAAQ,CAAC,KAAK,EAAE,SAAS,MAAM,EAAE,CAAC;IAClC,QAAQ,CAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,YAAY,EAAE,SAAS,UAAU,EAAE,CAAC;IAC7C,QAAQ,CAAC,YAAY,EAAE,SAAS,MAAM,EAAE,CAAC;IACzC,QAAQ,CAAC,IAAI,EAAE,OAAO,CAAC;CACxB;AAED,wBAAgB,aAAa,CAAC,IAAI,EAAE,SAAS,MAAM,EAAE,GAAG,WAAW,CA4DlE"}
1
+ {"version":3,"file":"parse-args.d.ts","sourceRoot":"","sources":["../../src/evals/parse-args.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,WAAW,CAAC;AAE5C;;;;;GAKG;AACH,MAAM,WAAW,WAAW;IAC1B,QAAQ,CAAC,KAAK,EAAE,SAAS,MAAM,EAAE,CAAC;IAClC,QAAQ,CAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,YAAY,EAAE,SAAS,UAAU,EAAE,CAAC;IAC7C,QAAQ,CAAC,YAAY,EAAE,SAAS,MAAM,EAAE,CAAC;IACzC,QAAQ,CAAC,IAAI,EAAE,OAAO,CAAC;CACxB;AAUD,wBAAgB,aAAa,CAAC,IAAI,EAAE,SAAS,MAAM,EAAE,GAAG,WAAW,CA4DlE"}
@@ -1,3 +1,10 @@
1
+ /**
2
+ * Options handed straight to the main CLI's parser.
3
+ *
4
+ * --api-key is here because the credential chain names it the highest-priority
5
+ * source: an ephemeral machine gets its key as a flag, not as stored state.
6
+ */
7
+ const PROVIDER_OPTIONS = new Set(["--provider", "--model", "--api-key"]);
1
8
  export function parseEvalArgs(argv) {
2
9
  const tasks = [];
3
10
  const capabilities = [];
@@ -44,7 +51,7 @@ export function parseEvalArgs(argv) {
44
51
  else if (argument === "--allow-shell") {
45
52
  capabilities.push("shell");
46
53
  }
47
- else if (argument === "--provider" || argument === "--model") {
54
+ else if (argument !== undefined && PROVIDER_OPTIONS.has(argument)) {
48
55
  // Passed through so the Provider registry resolves them exactly as the
49
56
  // real CLI does, including environment fallbacks.
50
57
  providerArgs.push(argument, value(index, argument));
@@ -1 +1 @@
1
- {"version":3,"file":"parse-args.js","sourceRoot":"","sources":["../../src/evals/parse-args.ts"],"names":[],"mappings":"AAkBA,MAAM,UAAU,aAAa,CAAC,IAAuB;IACnD,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,MAAM,YAAY,GAAiB,EAAE,CAAC;IACtC,MAAM,YAAY,GAAa,EAAE,CAAC;IAClC,IAAI,QAA4B,CAAC;IACjC,IAAI,SAAS,GAAG,OAAO,CAAC;IACxB,IAAI,QAA4B,CAAC;IACjC,IAAI,IAAI,GAAG,KAAK,CAAC;IAEjB,MAAM,KAAK,GAAG,CAAC,KAAa,EAAE,MAAc,EAAU,EAAE;QACtD,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC;QAC7B,IAAI,IAAI,KAAK,SAAS,IAAI,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,EAAE,CAAC;YAC/C,MAAM,IAAI,SAAS,CAAC,GAAG,MAAM,oBAAoB,CAAC,CAAC;QACrD,CAAC;QACD,OAAO,IAAI,CAAC;IACd,CAAC,CAAC;IAEF,KAAK,IAAI,KAAK,GAAG,CAAC,EAAE,KAAK,GAAG,IAAI,CAAC,MAAM,EAAE,KAAK,IAAI,CAAC,EAAE,CAAC;QACpD,MAAM,QAAQ,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC;QAC7B,IAAI,QAAQ,KAAK,QAAQ,IAAI,QAAQ,KAAK,IAAI,EAAE,CAAC;YAC/C,IAAI,GAAG,IAAI,CAAC;QACd,CAAC;aAAM,IAAI,QAAQ,KAAK,QAAQ,EAAE,CAAC;YACjC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,QAAQ,CAAC,CAAC,CAAC;YACnC,KAAK,IAAI,CAAC,CAAC;QACb,CAAC;aAAM,IAAI,QAAQ,KAAK,YAAY,EAAE,CAAC;YACrC,MAAM,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,KAAK,EAAE,YAAY,CAAC,CAAC,CAAC;YAClD,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,MAAM,GAAG,CAAC,IAAI,MAAM,GAAG,EAAE,EAAE,CAAC;gBAC/D,MAAM,IAAI,SAAS,CAAC,6CAA6C,CAAC,CAAC;YACrE,CAAC;YACD,QAAQ,GAAG,MAAM,CAAC;YAClB,KAAK,IAAI,CAAC,CAAC;QACb,CAAC;aAAM,IAAI,QAAQ,KAAK,OAAO,EAAE,CAAC;YAChC,SAAS,GAAG,KAAK,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC;YAClC,KAAK,IAAI,CAAC,CAAC;QACb,CAAC;aAAM,IAAI,QAAQ,KAAK,QAAQ,EAAE,CAAC;YACjC,QAAQ,GAAG,KAAK,CAAC,KAAK,EAAE,QAAQ,CAAC,CAAC;YAClC,KAAK,IAAI,CAAC,CAAC;QACb,CAAC;aAAM,IAAI,QAAQ,KAAK,gBAAgB,EAAE,CAAC;YACzC,YAAY,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QAC9B,CAAC;aAAM,IAAI,QAAQ,KAAK,eAAe,EAAE,CAAC;YACxC,YAAY,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QAC7B,CAAC;aAAM,IAAI,QAAQ,KAAK,YAAY,IAAI,QAAQ,KAAK,SAAS,EAAE,CAAC;YAC/D,uEAAuE;YACvE,kDAAkD;YAClD,YAAY,CAAC,IAAI,CAAC,QAAQ,EAAE,KAAK,CAAC,KAAK,EAAE,QAAQ,CAAC,CAAC,CAAC;YACpD,KAAK,IAAI,CAAC,CAAC;QACb,CAAC;aAAM,IAAI,QAAQ,KAAK,SAAS,EAAE,CAAC;YAClC,MAAM,IAAI,SAAS,CAAC,mBAAmB,QAAQ,EAAE,CAAC,CAAC;QACrD,CAAC;IACH,CAAC;IAED,OAAO;QACL,KAAK;QACL,GAAG,CAAC,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,CAAC;QAC/C,SAAS;QACT,GAAG,CAAC,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,CAAC;QAC/C,YAAY;QACZ,YAAY;QACZ,IAAI;KACL,CAAC;AACJ,CAAC"}
1
+ {"version":3,"file":"parse-args.js","sourceRoot":"","sources":["../../src/evals/parse-args.ts"],"names":[],"mappings":"AAkBA;;;;;GAKG;AACH,MAAM,gBAAgB,GAAG,IAAI,GAAG,CAAC,CAAC,YAAY,EAAE,SAAS,EAAE,WAAW,CAAC,CAAC,CAAC;AAEzE,MAAM,UAAU,aAAa,CAAC,IAAuB;IACnD,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,MAAM,YAAY,GAAiB,EAAE,CAAC;IACtC,MAAM,YAAY,GAAa,EAAE,CAAC;IAClC,IAAI,QAA4B,CAAC;IACjC,IAAI,SAAS,GAAG,OAAO,CAAC;IACxB,IAAI,QAA4B,CAAC;IACjC,IAAI,IAAI,GAAG,KAAK,CAAC;IAEjB,MAAM,KAAK,GAAG,CAAC,KAAa,EAAE,MAAc,EAAU,EAAE;QACtD,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC;QAC7B,IAAI,IAAI,KAAK,SAAS,IAAI,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,EAAE,CAAC;YAC/C,MAAM,IAAI,SAAS,CAAC,GAAG,MAAM,oBAAoB,CAAC,CAAC;QACrD,CAAC;QACD,OAAO,IAAI,CAAC;IACd,CAAC,CAAC;IAEF,KAAK,IAAI,KAAK,GAAG,CAAC,EAAE,KAAK,GAAG,IAAI,CAAC,MAAM,EAAE,KAAK,IAAI,CAAC,EAAE,CAAC;QACpD,MAAM,QAAQ,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC;QAC7B,IAAI,QAAQ,KAAK,QAAQ,IAAI,QAAQ,KAAK,IAAI,EAAE,CAAC;YAC/C,IAAI,GAAG,IAAI,CAAC;QACd,CAAC;aAAM,IAAI,QAAQ,KAAK,QAAQ,EAAE,CAAC;YACjC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,QAAQ,CAAC,CAAC,CAAC;YACnC,KAAK,IAAI,CAAC,CAAC;QACb,CAAC;aAAM,IAAI,QAAQ,KAAK,YAAY,EAAE,CAAC;YACrC,MAAM,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,KAAK,EAAE,YAAY,CAAC,CAAC,CAAC;YAClD,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,MAAM,GAAG,CAAC,IAAI,MAAM,GAAG,EAAE,EAAE,CAAC;gBAC/D,MAAM,IAAI,SAAS,CAAC,6CAA6C,CAAC,CAAC;YACrE,CAAC;YACD,QAAQ,GAAG,MAAM,CAAC;YAClB,KAAK,IAAI,CAAC,CAAC;QACb,CAAC;aAAM,IAAI,QAAQ,KAAK,OAAO,EAAE,CAAC;YAChC,SAAS,GAAG,KAAK,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC;YAClC,KAAK,IAAI,CAAC,CAAC;QACb,CAAC;aAAM,IAAI,QAAQ,KAAK,QAAQ,EAAE,CAAC;YACjC,QAAQ,GAAG,KAAK,CAAC,KAAK,EAAE,QAAQ,CAAC,CAAC;YAClC,KAAK,IAAI,CAAC,CAAC;QACb,CAAC;aAAM,IAAI,QAAQ,KAAK,gBAAgB,EAAE,CAAC;YACzC,YAAY,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QAC9B,CAAC;aAAM,IAAI,QAAQ,KAAK,eAAe,EAAE,CAAC;YACxC,YAAY,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QAC7B,CAAC;aAAM,IAAI,QAAQ,KAAK,SAAS,IAAI,gBAAgB,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC;YACpE,uEAAuE;YACvE,kDAAkD;YAClD,YAAY,CAAC,IAAI,CAAC,QAAQ,EAAE,KAAK,CAAC,KAAK,EAAE,QAAQ,CAAC,CAAC,CAAC;YACpD,KAAK,IAAI,CAAC,CAAC;QACb,CAAC;aAAM,IAAI,QAAQ,KAAK,SAAS,EAAE,CAAC;YAClC,MAAM,IAAI,SAAS,CAAC,mBAAmB,QAAQ,EAAE,CAAC,CAAC;QACrD,CAAC;IACH,CAAC;IAED,OAAO;QACL,KAAK;QACL,GAAG,CAAC,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,CAAC;QAC/C,SAAS;QACT,GAAG,CAAC,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,CAAC;QAC/C,YAAY;QACZ,YAAY;QACZ,IAAI;KACL,CAAC;AACJ,CAAC"}
@@ -1 +1 @@
1
- {"version":3,"file":"report.d.ts","sourceRoot":"","sources":["../../src/evals/report.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAE9C,MAAM,WAAW,UAAU;IACzB,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;CAC5B;AAwBD;;;;;GAKG;AACH,wBAAgB,WAAW,CAAC,OAAO,EAAE,SAAS,UAAU,EAAE,GAAG,MAAM,CAqClE;AAED,wEAAwE;AACxE,wBAAgB,cAAc,CAAC,OAAO,EAAE,SAAS,UAAU,EAAE,GAAG,MAAM,CAgBrE;AAED,wBAAgB,YAAY,CAC1B,OAAO,EAAE,SAAS,UAAU,EAAE,EAC9B,IAAI,EAAE,UAAU,GACf,MAAM,CA4BR"}
1
+ {"version":3,"file":"report.d.ts","sourceRoot":"","sources":["../../src/evals/report.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAG9C,MAAM,WAAW,UAAU;IACzB,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;CAC5B;AA4CD;;;;;GAKG;AACH,wBAAgB,WAAW,CAAC,OAAO,EAAE,SAAS,UAAU,EAAE,GAAG,MAAM,CA0ClE;AAED,wEAAwE;AACxE,wBAAgB,cAAc,CAAC,OAAO,EAAE,SAAS,UAAU,EAAE,GAAG,MAAM,CAgBrE;AAED,wBAAgB,YAAY,CAC1B,OAAO,EAAE,SAAS,UAAU,EAAE,EAC9B,IAAI,EAAE,UAAU,GACf,MAAM,CAmCR"}
@@ -1,3 +1,4 @@
1
+ import { formatTokens } from "../providers/usage.js";
1
2
  function median(values) {
2
3
  if (values.length === 0) {
3
4
  return 0;
@@ -14,6 +15,17 @@ function mean(values) {
14
15
  }
15
16
  return values.reduce((total, value) => total + value, 0) / values.length;
16
17
  }
18
+ /** Blank rather than zero when the Provider reported nothing. */
19
+ function medianTokens(result) {
20
+ const totals = result.attempts
21
+ .map((attempt) => attempt.usage?.usage.totalTokens)
22
+ .filter((value) => value !== undefined);
23
+ return totals.length === 0 ? "-" : formatTokens(median(totals));
24
+ }
25
+ function totalTokens(results) {
26
+ return results.reduce((sum, result) => sum +
27
+ result.attempts.reduce((inner, attempt) => inner + (attempt.usage?.usage.totalTokens ?? 0), 0), 0);
28
+ }
17
29
  function pad(value, width) {
18
30
  return value.length >= width ? value : value + " ".repeat(width - value.length);
19
31
  }
@@ -31,6 +43,7 @@ export function formatTable(results) {
31
43
  task: result.task,
32
44
  pass: `${result.passed}/${result.total}`,
33
45
  turns: mean(result.attempts.map((attempt) => attempt.turnCount)).toFixed(1),
46
+ tokens: medianTokens(result),
34
47
  tools: [
35
48
  ...new Set(result.attempts.flatMap((attempt) => attempt.toolsUsed)),
36
49
  ].join(","),
@@ -40,19 +53,21 @@ export function formatTable(results) {
40
53
  task: Math.max(4, ...rows.map((row) => row.task.length)),
41
54
  pass: Math.max(4, ...rows.map((row) => row.pass.length)),
42
55
  turns: Math.max(5, ...rows.map((row) => row.turns.length)),
56
+ tokens: Math.max(7, ...rows.map((row) => row.tokens.length)),
43
57
  tools: Math.max(5, ...rows.map((row) => row.tools.length)),
44
58
  };
45
59
  const lines = [
46
- `${pad("task", widths.task)} ${pad("pass", widths.pass)} ${pad("turns", widths.turns)} ${pad("tools", widths.tools)} p50`,
60
+ `${pad("task", widths.task)} ${pad("pass", widths.pass)} ${pad("turns", widths.turns)} ${pad("tok/att", widths.tokens)} ${pad("tools", widths.tools)} p50`,
47
61
  ];
48
62
  for (const row of rows) {
49
- lines.push(`${pad(row.task, widths.task)} ${pad(row.pass, widths.pass)} ${pad(row.turns, widths.turns)} ${pad(row.tools, widths.tools)} ${row.time}`);
63
+ lines.push(`${pad(row.task, widths.task)} ${pad(row.pass, widths.pass)} ${pad(row.turns, widths.turns)} ${pad(row.tokens, widths.tokens)} ${pad(row.tools, widths.tools)} ${row.time}`);
50
64
  }
51
65
  const passed = results.reduce((total, result) => total + result.passed, 0);
52
66
  const total = results.reduce((sum, result) => sum + result.total, 0);
53
67
  const percent = total === 0 ? 0 : Math.round((passed / total) * 100);
68
+ const spent = totalTokens(results);
54
69
  lines.push("");
55
- lines.push(`overall ${passed}/${total} (${percent}%)`);
70
+ lines.push(`overall ${passed}/${total} (${percent}%)${spent === 0 ? "" : ` ${formatTokens(spent)} tokens`}`);
56
71
  return `${lines.join("\n")}\n`;
57
72
  }
58
73
  /** Failures, grouped so the same reason across attempts is one line. */
@@ -83,6 +98,7 @@ export function toJsonReport(results, meta) {
83
98
  total: result.total,
84
99
  passRate: result.total === 0 ? 0 : result.passed / result.total,
85
100
  meanTurns: mean(result.attempts.map((attempt) => attempt.turnCount)),
101
+ totalTokens: result.attempts.reduce((sum, attempt) => sum + (attempt.usage?.usage.totalTokens ?? 0), 0),
86
102
  medianDurationMs: median(result.attempts.map((attempt) => attempt.durationMs)),
87
103
  attempts: result.attempts.map((attempt) => ({
88
104
  attempt: attempt.attempt,
@@ -91,12 +107,15 @@ export function toJsonReport(results, meta) {
91
107
  turnCount: attempt.turnCount,
92
108
  durationMs: attempt.durationMs,
93
109
  toolsUsed: attempt.toolsUsed,
110
+ toolCalls: attempt.toolCalls,
111
+ ...(attempt.usage === undefined ? {} : { usage: attempt.usage }),
94
112
  failures: attempt.failures,
95
113
  })),
96
114
  })),
97
115
  overall: {
98
116
  passed: results.reduce((total, result) => total + result.passed, 0),
99
117
  total: results.reduce((sum, result) => sum + result.total, 0),
118
+ totalTokens: totalTokens(results),
100
119
  },
101
120
  };
102
121
  }
@@ -1 +1 @@
1
- {"version":3,"file":"report.js","sourceRoot":"","sources":["../../src/evals/report.ts"],"names":[],"mappings":"AASA,SAAS,MAAM,CAAC,MAAyB;IACvC,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACxB,OAAO,CAAC,CAAC;IACX,CAAC;IACD,MAAM,MAAM,GAAG,CAAC,GAAG,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IACjD,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IAC7C,OAAO,MAAM,CAAC,MAAM,GAAG,CAAC,KAAK,CAAC;QAC5B,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;QACrE,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC;AAC5B,CAAC;AAED,SAAS,IAAI,CAAC,MAAyB;IACrC,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACxB,OAAO,CAAC,CAAC;IACX,CAAC;IACD,OAAO,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,KAAK,EAAE,EAAE,CAAC,KAAK,GAAG,KAAK,EAAE,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;AAC3E,CAAC;AAED,SAAS,GAAG,CAAC,KAAa,EAAE,KAAa;IACvC,OAAO,KAAK,CAAC,MAAM,IAAI,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,GAAG,GAAG,CAAC,MAAM,CAAC,KAAK,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC;AAClF,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,WAAW,CAAC,OAA8B;IACxD,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACzB,OAAO,iBAAiB,CAAC;IAC3B,CAAC;IAED,MAAM,IAAI,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;QACpC,IAAI,EAAE,MAAM,CAAC,IAAI;QACjB,IAAI,EAAE,GAAG,MAAM,CAAC,MAAM,IAAI,MAAM,CAAC,KAAK,EAAE;QACxC,KAAK,EAAE,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC;QAC3E,KAAK,EAAE;YACL,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC;SACpE,CAAC,IAAI,CAAC,GAAG,CAAC;QACX,IAAI,EAAE,GAAG,CAAC,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG;KAC7F,CAAC,CAAC,CAAC;IAEJ,MAAM,MAAM,GAAG;QACb,IAAI,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACxD,IAAI,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACxD,KAAK,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;QAC1D,KAAK,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;KAC3D,CAAC;IAEF,MAAM,KAAK,GAAG;QACZ,GAAG,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,OAAO,EAAE,MAAM,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,OAAO,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO;KAC9H,CAAC;IACF,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;QACvB,KAAK,CAAC,IAAI,CACR,GAAG,GAAG,CAAC,GAAG,CAAC,IAAI,EAAE,MAAM,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,GAAG,CAAC,IAAI,EAAE,MAAM,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,GAAG,CAAC,KAAK,EAAE,MAAM,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,GAAG,CAAC,KAAK,EAAE,MAAM,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,IAAI,EAAE,CAC9I,CAAC;IACJ,CAAC;IAED,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,CAAC,KAAK,GAAG,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;IAC3E,MAAM,KAAK,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,MAAM,EAAE,EAAE,CAAC,GAAG,GAAG,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC;IACrE,MAAM,OAAO,GAAG,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,MAAM,GAAG,KAAK,CAAC,GAAG,GAAG,CAAC,CAAC;IACrE,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACf,KAAK,CAAC,IAAI,CAAC,YAAY,MAAM,IAAI,KAAK,KAAK,OAAO,IAAI,CAAC,CAAC;IACxD,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC;AACjC,CAAC;AAED,wEAAwE;AACxE,MAAM,UAAU,cAAc,CAAC,OAA8B;IAC3D,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;QAC7B,MAAM,QAAQ,GAAG,IAAI,GAAG,EAAoB,CAAC;QAC7C,KAAK,MAAM,OAAO,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YACtC,KAAK,MAAM,OAAO,IAAI,OAAO,CAAC,QAAQ,EAAE,CAAC;gBACvC,MAAM,QAAQ,GAAG,QAAQ,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC;gBAC7C,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;gBAC/B,QAAQ,CAAC,GAAG,CAAC,OAAO,EAAE,QAAQ,CAAC,CAAC;YAClC,CAAC;QACH,CAAC;QACD,KAAK,MAAM,CAAC,MAAM,EAAE,QAAQ,CAAC,IAAI,QAAQ,EAAE,CAAC;YAC1C,KAAK,CAAC,IAAI,CAAC,GAAG,MAAM,CAAC,IAAI,YAAY,QAAQ,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,MAAM,EAAE,CAAC,CAAC;QACxE,CAAC;IACH,CAAC;IACD,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC;AAC3D,CAAC;AAED,MAAM,UAAU,YAAY,CAC1B,OAA8B,EAC9B,IAAgB;IAEhB,OAAO;QACL,aAAa,EAAE,CAAC;QAChB,GAAG,IAAI;QACP,KAAK,EAAE,OAAO,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;YAC9B,IAAI,EAAE,MAAM,CAAC,IAAI;YACjB,MAAM,EAAE,MAAM,CAAC,MAAM;YACrB,KAAK,EAAE,MAAM,CAAC,KAAK;YACnB,QAAQ,EAAE,MAAM,CAAC,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,GAAG,MAAM,CAAC,KAAK;YAC/D,SAAS,EAAE,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC;YACpE,gBAAgB,EAAE,MAAM,CACtB,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,UAAU,CAAC,CACrD;YACD,QAAQ,EAAE,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;gBAC1C,OAAO,EAAE,OAAO,CAAC,OAAO;gBACxB,MAAM,EAAE,OAAO,CAAC,MAAM;gBACtB,MAAM,EAAE,OAAO,CAAC,MAAM;gBACtB,SAAS,EAAE,OAAO,CAAC,SAAS;gBAC5B,UAAU,EAAE,OAAO,CAAC,UAAU;gBAC9B,SAAS,EAAE,OAAO,CAAC,SAAS;gBAC5B,QAAQ,EAAE,OAAO,CAAC,QAAQ;aAC3B,CAAC,CAAC;SACJ,CAAC,CAAC;QACH,OAAO,EAAE;YACP,MAAM,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,CAAC,KAAK,GAAG,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC;YACnE,KAAK,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,MAAM,EAAE,EAAE,CAAC,GAAG,GAAG,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC;SAC9D;KACF,CAAC;AACJ,CAAC"}
1
+ {"version":3,"file":"report.js","sourceRoot":"","sources":["../../src/evals/report.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,YAAY,EAAE,MAAM,uBAAuB,CAAC;AASrD,SAAS,MAAM,CAAC,MAAyB;IACvC,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACxB,OAAO,CAAC,CAAC;IACX,CAAC;IACD,MAAM,MAAM,GAAG,CAAC,GAAG,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IACjD,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IAC7C,OAAO,MAAM,CAAC,MAAM,GAAG,CAAC,KAAK,CAAC;QAC5B,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;QACrE,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC;AAC5B,CAAC;AAED,SAAS,IAAI,CAAC,MAAyB;IACrC,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACxB,OAAO,CAAC,CAAC;IACX,CAAC;IACD,OAAO,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,KAAK,EAAE,EAAE,CAAC,KAAK,GAAG,KAAK,EAAE,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;AAC3E,CAAC;AAED,iEAAiE;AACjE,SAAS,YAAY,CAAC,MAAkB;IACtC,MAAM,MAAM,GAAG,MAAM,CAAC,QAAQ;SAC3B,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,KAAK,EAAE,KAAK,CAAC,WAAW,CAAC;SAClD,MAAM,CAAC,CAAC,KAAK,EAAmB,EAAE,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC;IAC3D,OAAO,MAAM,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,YAAY,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC;AAClE,CAAC;AAED,SAAS,WAAW,CAAC,OAA8B;IACjD,OAAO,OAAO,CAAC,MAAM,CACnB,CAAC,GAAG,EAAE,MAAM,EAAE,EAAE,CACd,GAAG;QACH,MAAM,CAAC,QAAQ,CAAC,MAAM,CACpB,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE,CAAC,KAAK,GAAG,CAAC,OAAO,CAAC,KAAK,EAAE,KAAK,CAAC,WAAW,IAAI,CAAC,CAAC,EACnE,CAAC,CACF,EACH,CAAC,CACF,CAAC;AACJ,CAAC;AAED,SAAS,GAAG,CAAC,KAAa,EAAE,KAAa;IACvC,OAAO,KAAK,CAAC,MAAM,IAAI,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,GAAG,GAAG,CAAC,MAAM,CAAC,KAAK,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC;AAClF,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,WAAW,CAAC,OAA8B;IACxD,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACzB,OAAO,iBAAiB,CAAC;IAC3B,CAAC;IAED,MAAM,IAAI,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;QACpC,IAAI,EAAE,MAAM,CAAC,IAAI;QACjB,IAAI,EAAE,GAAG,MAAM,CAAC,MAAM,IAAI,MAAM,CAAC,KAAK,EAAE;QACxC,KAAK,EAAE,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC;QAC3E,MAAM,EAAE,YAAY,CAAC,MAAM,CAAC;QAC5B,KAAK,EAAE;YACL,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC;SACpE,CAAC,IAAI,CAAC,GAAG,CAAC;QACX,IAAI,EAAE,GAAG,CAAC,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG;KAC7F,CAAC,CAAC,CAAC;IAEJ,MAAM,MAAM,GAAG;QACb,IAAI,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACxD,IAAI,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACxD,KAAK,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;QAC1D,MAAM,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;QAC5D,KAAK,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;KAC3D,CAAC;IAEF,MAAM,KAAK,GAAG;QACZ,GAAG,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,OAAO,EAAE,MAAM,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,SAAS,EAAE,MAAM,CAAC,MAAM,CAAC,KAAK,GAAG,CAAC,OAAO,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO;KAChK,CAAC;IACF,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;QACvB,KAAK,CAAC,IAAI,CACR,GAAG,GAAG,CAAC,GAAG,CAAC,IAAI,EAAE,MAAM,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,GAAG,CAAC,IAAI,EAAE,MAAM,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,GAAG,CAAC,KAAK,EAAE,MAAM,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,CAAC,KAAK,GAAG,CAAC,GAAG,CAAC,KAAK,EAAE,MAAM,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,IAAI,EAAE,CACjL,CAAC;IACJ,CAAC;IAED,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,CAAC,KAAK,GAAG,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;IAC3E,MAAM,KAAK,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,MAAM,EAAE,EAAE,CAAC,GAAG,GAAG,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC;IACrE,MAAM,OAAO,GAAG,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,MAAM,GAAG,KAAK,CAAC,GAAG,GAAG,CAAC,CAAC;IACrE,MAAM,KAAK,GAAG,WAAW,CAAC,OAAO,CAAC,CAAC;IACnC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACf,KAAK,CAAC,IAAI,CACR,YAAY,MAAM,IAAI,KAAK,KAAK,OAAO,KAAK,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,KAAK,YAAY,CAAC,KAAK,CAAC,SAAS,EAAE,CACnG,CAAC;IACF,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC;AACjC,CAAC;AAED,wEAAwE;AACxE,MAAM,UAAU,cAAc,CAAC,OAA8B;IAC3D,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;QAC7B,MAAM,QAAQ,GAAG,IAAI,GAAG,EAAoB,CAAC;QAC7C,KAAK,MAAM,OAAO,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YACtC,KAAK,MAAM,OAAO,IAAI,OAAO,CAAC,QAAQ,EAAE,CAAC;gBACvC,MAAM,QAAQ,GAAG,QAAQ,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC;gBAC7C,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;gBAC/B,QAAQ,CAAC,GAAG,CAAC,OAAO,EAAE,QAAQ,CAAC,CAAC;YAClC,CAAC;QACH,CAAC;QACD,KAAK,MAAM,CAAC,MAAM,EAAE,QAAQ,CAAC,IAAI,QAAQ,EAAE,CAAC;YAC1C,KAAK,CAAC,IAAI,CAAC,GAAG,MAAM,CAAC,IAAI,YAAY,QAAQ,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,MAAM,EAAE,CAAC,CAAC;QACxE,CAAC;IACH,CAAC;IACD,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC;AAC3D,CAAC;AAED,MAAM,UAAU,YAAY,CAC1B,OAA8B,EAC9B,IAAgB;IAEhB,OAAO;QACL,aAAa,EAAE,CAAC;QAChB,GAAG,IAAI;QACP,KAAK,EAAE,OAAO,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;YAC9B,IAAI,EAAE,MAAM,CAAC,IAAI;YACjB,MAAM,EAAE,MAAM,CAAC,MAAM;YACrB,KAAK,EAAE,MAAM,CAAC,KAAK;YACnB,QAAQ,EAAE,MAAM,CAAC,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,GAAG,MAAM,CAAC,KAAK;YAC/D,SAAS,EAAE,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC;YACpE,WAAW,EAAE,MAAM,CAAC,QAAQ,CAAC,MAAM,CACjC,CAAC,GAAG,EAAE,OAAO,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,OAAO,CAAC,KAAK,EAAE,KAAK,CAAC,WAAW,IAAI,CAAC,CAAC,EAC/D,CAAC,CACF;YACD,gBAAgB,EAAE,MAAM,CACtB,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,UAAU,CAAC,CACrD;YACD,QAAQ,EAAE,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;gBAC1C,OAAO,EAAE,OAAO,CAAC,OAAO;gBACxB,MAAM,EAAE,OAAO,CAAC,MAAM;gBACtB,MAAM,EAAE,OAAO,CAAC,MAAM;gBACtB,SAAS,EAAE,OAAO,CAAC,SAAS;gBAC5B,UAAU,EAAE,OAAO,CAAC,UAAU;gBAC9B,SAAS,EAAE,OAAO,CAAC,SAAS;gBAC5B,SAAS,EAAE,OAAO,CAAC,SAAS;gBAC5B,GAAG,CAAC,OAAO,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,EAAE,CAAC;gBAChE,QAAQ,EAAE,OAAO,CAAC,QAAQ;aAC3B,CAAC,CAAC;SACJ,CAAC,CAAC;QACH,OAAO,EAAE;YACP,MAAM,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,CAAC,KAAK,GAAG,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC;YACnE,KAAK,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,MAAM,EAAE,EAAE,CAAC,GAAG,GAAG,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC;YAC7D,WAAW,EAAE,WAAW,CAAC,OAAO,CAAC;SAClC;KACF,CAAC;AACJ,CAAC"}
@@ -1,6 +1,23 @@
1
1
  import type { LLMClient } from "../llm.js";
2
+ import type { UsageTotal } from "../providers/usage.js";
2
3
  import { type AttemptFacts } from "./graders.js";
3
4
  import type { Capability, Task } from "./task.js";
5
+ /** One tool result, in call order. */
6
+ export interface ToolCallRecord {
7
+ readonly name: string;
8
+ readonly ok: boolean;
9
+ /**
10
+ * What the call was aimed at: the `path` argument, or a truncated `command`.
11
+ *
12
+ * Full arguments are not kept — an edit carries both halves of the change
13
+ * and would bury the report in file contents. But a name and a success flag
14
+ * alone cannot tell a model that read the four files on an import chain from
15
+ * one that read four unrelated files, and on a task built out of 41 files
16
+ * that distinction is the entire measurement. Omitted when the tool takes
17
+ * neither.
18
+ */
19
+ readonly target?: string;
20
+ }
4
21
  export interface AttemptResult {
5
22
  readonly attempt: number;
6
23
  readonly passed: boolean;
@@ -8,6 +25,16 @@ export interface AttemptResult {
8
25
  readonly turnCount: number;
9
26
  readonly durationMs: number;
10
27
  readonly toolsUsed: readonly string[];
28
+ /**
29
+ * Every call in order, with whether it succeeded.
30
+ *
31
+ * `toolsUsed` is a deduplicated set, so a model that guessed a path, got an
32
+ * error and recovered looks identical there to one that never guessed. This
33
+ * is what makes that question answerable from the report.
34
+ */
35
+ readonly toolCalls: readonly ToolCallRecord[];
36
+ /** What the attempt cost, when the Provider reported it. */
37
+ readonly usage?: UsageTotal;
11
38
  /** One entry per failed grader, in task order. */
12
39
  readonly failures: readonly string[];
13
40
  }
@@ -18,7 +45,12 @@ export interface TaskResult {
18
45
  readonly total: number;
19
46
  }
20
47
  export interface RunnerOptions {
21
- /** Built per attempt, so a test can hand out a fresh fake each time. */
48
+ /**
49
+ * Called once per attempt, so a test can hand out a fresh fake each time.
50
+ * The CLI returns the same client every time on purpose: it holds only
51
+ * readonly config and takes the history per request, so attempts cannot
52
+ * leak into one another through it.
53
+ */
22
54
  readonly createClient: () => LLMClient;
23
55
  readonly systemPrompt: string;
24
56
  readonly capabilities: readonly Capability[];
@@ -1 +1 @@
1
- {"version":3,"file":"runner.d.ts","sourceRoot":"","sources":["../../src/evals/runner.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,WAAW,CAAC;AAU3C,OAAO,EAAgC,KAAK,YAAY,EAAE,MAAM,cAAc,CAAC;AAC/E,OAAO,KAAK,EAAE,UAAU,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAElD,MAAM,WAAW,aAAa;IAC5B,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,MAAM,EAAE,OAAO,CAAC;IACzB,QAAQ,CAAC,MAAM,EAAE,YAAY,CAAC,QAAQ,CAAC,CAAC;IACxC,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B,QAAQ,CAAC,SAAS,EAAE,SAAS,MAAM,EAAE,CAAC;IACtC,kDAAkD;IAClD,QAAQ,CAAC,QAAQ,EAAE,SAAS,MAAM,EAAE,CAAC;CACtC;AAED,MAAM,WAAW,UAAU;IACzB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,QAAQ,EAAE,SAAS,aAAa,EAAE,CAAC;IAC5C,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;CACxB;AAED,MAAM,WAAW,aAAa;IAC5B,wEAAwE;IACxE,QAAQ,CAAC,YAAY,EAAE,MAAM,SAAS,CAAC;IACvC,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;IAC9B,QAAQ,CAAC,YAAY,EAAE,SAAS,UAAU,EAAE,CAAC;IAC7C,QAAQ,CAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,SAAS,CAAC,EAAE,CAAC,MAAM,EAAE,aAAa,EAAE,IAAI,EAAE,IAAI,KAAK,IAAI,CAAC;IACjE,QAAQ,CAAC,MAAM,CAAC,EAAE,WAAW,CAAC;CAC/B;AAED,qBAAa,sBAAuB,SAAQ,KAAK;IACnC,QAAQ,CAAC,IAAI,EAAE,MAAM;IAAE,QAAQ,CAAC,OAAO,EAAE,SAAS,UAAU,EAAE;IAA1E,YAAqB,IAAI,EAAE,MAAM,EAAW,OAAO,EAAE,SAAS,UAAU,EAAE,EAOzE;CACF;AAgHD;;;;;;GAMG;AACH,wBAAsB,OAAO,CAC3B,IAAI,EAAE,IAAI,EACV,OAAO,EAAE,aAAa,GACrB,OAAO,CAAC,UAAU,CAAC,CAoBrB;AAED,wBAAsB,QAAQ,CAC5B,KAAK,EAAE,SAAS,IAAI,EAAE,EACtB,OAAO,EAAE,aAAa,GACrB,OAAO,CAAC,SAAS,UAAU,EAAE,CAAC,CAMhC"}
1
+ {"version":3,"file":"runner.d.ts","sourceRoot":"","sources":["../../src/evals/runner.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,WAAW,CAAC;AAC3C,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,uBAAuB,CAAC;AAIxD,OAAO,EAAgC,KAAK,YAAY,EAAE,MAAM,cAAc,CAAC;AAC/E,OAAO,KAAK,EAAE,UAAU,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAElD,sCAAsC;AACtC,MAAM,WAAW,cAAc;IAC7B,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,EAAE,EAAE,OAAO,CAAC;IACrB;;;;;;;;;OASG;IACH,QAAQ,CAAC,MAAM,CAAC,EAAE,MAAM,CAAC;CAC1B;AAwBD,MAAM,WAAW,aAAa;IAC5B,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,MAAM,EAAE,OAAO,CAAC;IACzB,QAAQ,CAAC,MAAM,EAAE,YAAY,CAAC,QAAQ,CAAC,CAAC;IACxC,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B,QAAQ,CAAC,SAAS,EAAE,SAAS,MAAM,EAAE,CAAC;IACtC;;;;;;OAMG;IACH,QAAQ,CAAC,SAAS,EAAE,SAAS,cAAc,EAAE,CAAC;IAC9C,4DAA4D;IAC5D,QAAQ,CAAC,KAAK,CAAC,EAAE,UAAU,CAAC;IAC5B,kDAAkD;IAClD,QAAQ,CAAC,QAAQ,EAAE,SAAS,MAAM,EAAE,CAAC;CACtC;AAED,MAAM,WAAW,UAAU;IACzB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,QAAQ,EAAE,SAAS,aAAa,EAAE,CAAC;IAC5C,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC;CACxB;AAED,MAAM,WAAW,aAAa;IAC5B;;;;;OAKG;IACH,QAAQ,CAAC,YAAY,EAAE,MAAM,SAAS,CAAC;IACvC,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;IAC9B,QAAQ,CAAC,YAAY,EAAE,SAAS,UAAU,EAAE,CAAC;IAC7C,QAAQ,CAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,SAAS,CAAC,EAAE,CAAC,MAAM,EAAE,aAAa,EAAE,IAAI,EAAE,IAAI,KAAK,IAAI,CAAC;IACjE,QAAQ,CAAC,MAAM,CAAC,EAAE,WAAW,CAAC;CAC/B;AAED,qBAAa,sBAAuB,SAAQ,KAAK;IACnC,QAAQ,CAAC,IAAI,EAAE,MAAM;IAAE,QAAQ,CAAC,OAAO,EAAE,SAAS,UAAU,EAAE;IAA1E,YAAqB,IAAI,EAAE,MAAM,EAAW,OAAO,EAAE,SAAS,UAAU,EAAE,EAOzE;CACF;AAwHD;;;;;;GAMG;AACH,wBAAsB,OAAO,CAC3B,IAAI,EAAE,IAAI,EACV,OAAO,EAAE,aAAa,GACrB,OAAO,CAAC,UAAU,CAAC,CAoBrB;AAED,wBAAsB,QAAQ,CAC5B,KAAK,EAAE,SAAS,IAAI,EAAE,EACtB,OAAO,EAAE,aAAa,GACrB,OAAO,CAAC,SAAS,UAAU,EAAE,CAAC,CAMhC"}
@@ -1,13 +1,27 @@
1
1
  import { Agent as AgentClass } from "../agent.js";
2
2
  import { LocalWorkspace } from "../workspace.js";
3
- import { ListFilesTool } from "../tools/list-files.js";
4
- import { ReadFileTool } from "../tools/read-file.js";
5
- import { SearchTextTool } from "../tools/search-text.js";
6
- import { WriteFileTool } from "../tools/write-file.js";
7
- import { EditFileTool } from "../tools/edit-file.js";
8
- import { BashTool } from "../tools/bash.js";
3
+ import { toolsFor } from "./tools.js";
9
4
  import { createAttemptWorkspace } from "./fixture.js";
10
5
  import { createGrader, describeGrader } from "./graders.js";
6
+ const MAX_TARGET_LENGTH = 80;
7
+ /** The one argument worth keeping, short enough to sit in a report. */
8
+ function callTarget(argumentsValue) {
9
+ if (typeof argumentsValue !== "object" || argumentsValue === null) {
10
+ return undefined;
11
+ }
12
+ const record = argumentsValue;
13
+ const value = typeof record.path === "string"
14
+ ? record.path
15
+ : typeof record.command === "string"
16
+ ? record.command
17
+ : undefined;
18
+ if (value === undefined) {
19
+ return undefined;
20
+ }
21
+ return value.length > MAX_TARGET_LENGTH
22
+ ? `${value.slice(0, MAX_TARGET_LENGTH)}…`
23
+ : value;
24
+ }
11
25
  export class MissingCapabilityError extends Error {
12
26
  task;
13
27
  missing;
@@ -20,20 +34,6 @@ export class MissingCapabilityError extends Error {
20
34
  this.name = "MissingCapabilityError";
21
35
  }
22
36
  }
23
- function toolsFor(capabilities, cwd) {
24
- const tools = [
25
- new ListFilesTool(),
26
- new SearchTextTool(),
27
- new ReadFileTool(),
28
- ];
29
- if (capabilities.includes("writes")) {
30
- tools.push(new WriteFileTool(), new EditFileTool());
31
- }
32
- if (capabilities.includes("shell")) {
33
- tools.push(new BashTool({ cwd }));
34
- }
35
- return tools;
36
- }
37
37
  function missingCapabilities(task, granted) {
38
38
  return task.capabilities.filter((capability) => !granted.includes(capability));
39
39
  }
@@ -56,12 +56,14 @@ async function runAttempt(task, attempt, options) {
56
56
  let status = "error";
57
57
  let finalAnswer = "";
58
58
  let turnCount = 0;
59
+ let usage;
59
60
  try {
60
61
  const result = await agent.run(task.prompt, {
61
62
  ...(options.signal === undefined ? {} : { signal: options.signal }),
62
63
  });
63
64
  status = result.status;
64
65
  turnCount = result.turnCount;
66
+ usage = result.usage;
65
67
  finalAnswer =
66
68
  result.status === "completed" ? result.finalMessage.content : "";
67
69
  }
@@ -73,6 +75,7 @@ async function runAttempt(task, attempt, options) {
73
75
  }
74
76
  const facts = {
75
77
  workspace: workspace.path,
78
+ fixtureDirectory: task.fixtureDirectory,
76
79
  finalAnswer,
77
80
  turnCount,
78
81
  status,
@@ -90,17 +93,37 @@ async function runAttempt(task, attempt, options) {
90
93
  failures.push(`${describeGrader(spec)} — ${result.reason}`);
91
94
  }
92
95
  }
96
+ // The arguments live on the start event, the outcome on the end event;
97
+ // the call id is what joins them.
98
+ const targets = new Map();
99
+ for (const event of events) {
100
+ if (event.type === "tool_execution_start") {
101
+ const target = callTarget(event.arguments);
102
+ if (target !== undefined) {
103
+ targets.set(event.toolCallId, target);
104
+ }
105
+ }
106
+ }
107
+ const calls = events
108
+ .filter((event) => event.type === "tool_execution_end")
109
+ .map((event) => {
110
+ const end = event;
111
+ const target = targets.get(end.toolCallId);
112
+ return {
113
+ name: end.toolName,
114
+ ok: end.isError === false,
115
+ ...(target === undefined ? {} : { target }),
116
+ };
117
+ });
93
118
  return {
94
119
  attempt,
95
120
  passed: failures.length === 0,
96
121
  status,
97
122
  turnCount,
98
123
  durationMs: Date.now() - startedAt,
99
- toolsUsed: [
100
- ...new Set(events
101
- .filter((event) => event.type === "tool_execution_end")
102
- .map((event) => event.toolName)),
103
- ],
124
+ ...(usage === undefined ? {} : { usage }),
125
+ toolsUsed: [...new Set(calls.map((call) => call.name))],
126
+ toolCalls: calls,
104
127
  failures,
105
128
  };
106
129
  }
@@ -1 +1 @@
1
- {"version":3,"file":"runner.js","sourceRoot":"","sources":["../../src/evals/runner.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,KAAK,IAAI,UAAU,EAAE,MAAM,aAAa,CAAC;AAIlD,OAAO,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAC;AACjD,OAAO,EAAE,aAAa,EAAE,MAAM,wBAAwB,CAAC;AACvD,OAAO,EAAE,YAAY,EAAE,MAAM,uBAAuB,CAAC;AACrD,OAAO,EAAE,cAAc,EAAE,MAAM,yBAAyB,CAAC;AACzD,OAAO,EAAE,aAAa,EAAE,MAAM,wBAAwB,CAAC;AACvD,OAAO,EAAE,YAAY,EAAE,MAAM,uBAAuB,CAAC;AACrD,OAAO,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAC;AAC5C,OAAO,EAAE,sBAAsB,EAAE,MAAM,cAAc,CAAC;AACtD,OAAO,EAAE,YAAY,EAAE,cAAc,EAAqB,MAAM,cAAc,CAAC;AA+B/E,MAAM,OAAO,sBAAuB,SAAQ,KAAK;IAC1B,IAAI;IAAmB,OAAO;IAAnD,YAAqB,IAAY,EAAW,OAA8B;QACxE,KAAK,CACH,SAAS,IAAI,WAAW,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,gBAAgB,OAAO;aACjE,GAAG,CAAC,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC,UAAU,KAAK,QAAQ,CAAC,CAAC,CAAC,gBAAgB,CAAC,CAAC,CAAC,eAAe,CAAC,CAAC;aACnF,IAAI,CAAC,GAAG,CAAC,GAAG,CAChB,CAAC;oBALiB,IAAI;uBAAmB,OAAO;QAMjD,IAAI,CAAC,IAAI,GAAG,wBAAwB,CAAC;IACvC,CAAC;CACF;AAED,SAAS,QAAQ,CACf,YAAmC,EACnC,GAAW;IAEX,MAAM,KAAK,GAAW;QACpB,IAAI,aAAa,EAAE;QACnB,IAAI,cAAc,EAAE;QACpB,IAAI,YAAY,EAAE;KACnB,CAAC;IACF,IAAI,YAAY,CAAC,QAAQ,CAAC,QAAQ,CAAC,EAAE,CAAC;QACpC,KAAK,CAAC,IAAI,CAAC,IAAI,aAAa,EAAE,EAAE,IAAI,YAAY,EAAE,CAAC,CAAC;IACtD,CAAC;IACD,IAAI,YAAY,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;QACnC,KAAK,CAAC,IAAI,CAAC,IAAI,QAAQ,CAAC,EAAE,GAAG,EAAE,CAAC,CAAC,CAAC;IACpC,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED,SAAS,mBAAmB,CAC1B,IAAU,EACV,OAA8B;IAE9B,OAAO,IAAI,CAAC,YAAY,CAAC,MAAM,CAC7B,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAC,CAC9C,CAAC;AACJ,CAAC;AAED,KAAK,UAAU,UAAU,CACvB,IAAU,EACV,OAAe,EACf,OAAsB;IAEtB,MAAM,SAAS,GAAG,MAAM,sBAAsB,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC;IACtE,MAAM,MAAM,GAAiB,EAAE,CAAC;IAChC,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IAE7B,IAAI,CAAC;QACH,MAAM,KAAK,GAAU,IAAI,UAAU,CAAC;YAClC,YAAY,EAAE,OAAO,CAAC,YAAY;YAClC,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,GAAG,EAAE,OAAO,CAAC,YAAY,EAAE;YAC3B,KAAK,EAAE,QAAQ,CAAC,IAAI,CAAC,YAAY,EAAE,SAAS,CAAC,IAAI,CAAC;YAClD,SAAS,EAAE,IAAI,cAAc,CAAC,SAAS,CAAC,IAAI,EAAE;gBAC5C,WAAW,EAAE,IAAI,CAAC,YAAY,CAAC,QAAQ,CAAC,QAAQ,CAAC;aAClD,CAAC;YACF,SAAS,EAAE,KAAK;YAChB,OAAO,EAAE,CAAC,KAAK,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC;SACvC,CAAC,CAAC;QAEH,IAAI,MAAM,GAA2B,OAAO,CAAC;QAC7C,IAAI,WAAW,GAAG,EAAE,CAAC;QACrB,IAAI,SAAS,GAAG,CAAC,CAAC;QAClB,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,EAAE;gBAC1C,GAAG,CAAC,OAAO,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC;aACpE,CAAC,CAAC;YACH,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC;YACvB,SAAS,GAAG,MAAM,CAAC,SAAS,CAAC;YAC7B,WAAW;gBACT,MAAM,CAAC,MAAM,KAAK,WAAW,CAAC,CAAC,CAAC,MAAM,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC;QACrE,CAAC;QAAC,OAAO,KAAc,EAAE,CAAC;YACxB,yEAAyE;YACzE,mEAAmE;YACnE,MAAM,GAAG,OAAO,CAAC;YACjB,WAAW,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QACvE,CAAC;QAED,MAAM,KAAK,GAAiB;YAC1B,SAAS,EAAE,SAAS,CAAC,IAAI;YACzB,WAAW;YACX,SAAS;YACT,MAAM;YACN,MAAM;SACP,CAAC;QAEF,MAAM,QAAQ,GAAa,EAAE,CAAC;QAC9B,IAAI,MAAM,KAAK,WAAW,EAAE,CAAC;YAC3B,QAAQ,CAAC,IAAI,CACX,MAAM,KAAK,WAAW;gBACpB,CAAC,CAAC,WAAW,IAAI,CAAC,QAAQ,oCAAoC;gBAC9D,CAAC,CAAC,gBAAgB,MAAM,KAAK,WAAW,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,EAAE,CACrE,CAAC;QACJ,CAAC;QACD,KAAK,MAAM,IAAI,IAAI,IAAI,CAAC,OAAO,EAAE,CAAC;YAChC,MAAM,MAAM,GAAG,MAAM,YAAY,CAAC,IAAI,CAAC,CAAC,KAAK,CAAC,CAAC;YAC/C,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC;gBACnB,QAAQ,CAAC,IAAI,CAAC,GAAG,cAAc,CAAC,IAAI,CAAC,MAAM,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC;YAC9D,CAAC;QACH,CAAC;QAED,OAAO;YACL,OAAO;YACP,MAAM,EAAE,QAAQ,CAAC,MAAM,KAAK,CAAC;YAC7B,MAAM;YACN,SAAS;YACT,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;YAClC,SAAS,EAAE;gBACT,GAAG,IAAI,GAAG,CACR,MAAM;qBACH,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,KAAK,oBAAoB,CAAC;qBACtD,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAE,KAA8B,CAAC,QAAQ,CAAC,CAC5D;aACF;YACD,QAAQ;SACT,CAAC;IACJ,CAAC;YAAS,CAAC;QACT,MAAM,SAAS,CAAC,OAAO,EAAE,CAAC;IAC5B,CAAC;AACH,CAAC;AAED;;;;;;GAMG;AACH,MAAM,CAAC,KAAK,UAAU,OAAO,CAC3B,IAAU,EACV,OAAsB;IAEtB,MAAM,OAAO,GAAG,mBAAmB,CAAC,IAAI,EAAE,OAAO,CAAC,YAAY,CAAC,CAAC;IAChE,IAAI,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACvB,MAAM,IAAI,sBAAsB,CAAC,IAAI,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;IACvD,CAAC;IAED,MAAM,KAAK,GAAG,OAAO,CAAC,QAAQ,IAAI,IAAI,CAAC,QAAQ,CAAC;IAChD,MAAM,QAAQ,GAAoB,EAAE,CAAC;IACrC,KAAK,IAAI,KAAK,GAAG,CAAC,EAAE,KAAK,IAAI,KAAK,EAAE,KAAK,IAAI,CAAC,EAAE,CAAC;QAC/C,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,IAAI,EAAE,KAAK,EAAE,OAAO,CAAC,CAAC;QACtD,QAAQ,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACtB,OAAO,CAAC,SAAS,EAAE,CAAC,MAAM,EAAE,IAAI,CAAC,CAAC;IACpC,CAAC;IAED,OAAO;QACL,IAAI,EAAE,IAAI,CAAC,IAAI;QACf,QAAQ;QACR,MAAM,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,MAAM;QAC3D,KAAK;KACN,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,QAAQ,CAC5B,KAAsB,EACtB,OAAsB;IAEtB,MAAM,OAAO,GAAiB,EAAE,CAAC;IACjC,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,OAAO,CAAC,IAAI,CAAC,MAAM,OAAO,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC,CAAC;IAC7C,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC"}
1
+ {"version":3,"file":"runner.js","sourceRoot":"","sources":["../../src/evals/runner.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,KAAK,IAAI,UAAU,EAAE,MAAM,aAAa,CAAC;AAIlD,OAAO,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAC;AACjD,OAAO,EAAE,QAAQ,EAAE,MAAM,YAAY,CAAC;AACtC,OAAO,EAAE,sBAAsB,EAAE,MAAM,cAAc,CAAC;AACtD,OAAO,EAAE,YAAY,EAAE,cAAc,EAAqB,MAAM,cAAc,CAAC;AAoB/E,MAAM,iBAAiB,GAAG,EAAE,CAAC;AAE7B,uEAAuE;AACvE,SAAS,UAAU,CAAC,cAAuB;IACzC,IAAI,OAAO,cAAc,KAAK,QAAQ,IAAI,cAAc,KAAK,IAAI,EAAE,CAAC;QAClE,OAAO,SAAS,CAAC;IACnB,CAAC;IACD,MAAM,MAAM,GAAG,cAAyC,CAAC;IACzD,MAAM,KAAK,GACT,OAAO,MAAM,CAAC,IAAI,KAAK,QAAQ;QAC7B,CAAC,CAAC,MAAM,CAAC,IAAI;QACb,CAAC,CAAC,OAAO,MAAM,CAAC,OAAO,KAAK,QAAQ;YAClC,CAAC,CAAC,MAAM,CAAC,OAAO;YAChB,CAAC,CAAC,SAAS,CAAC;IAClB,IAAI,KAAK,KAAK,SAAS,EAAE,CAAC;QACxB,OAAO,SAAS,CAAC;IACnB,CAAC;IACD,OAAO,KAAK,CAAC,MAAM,GAAG,iBAAiB;QACrC,CAAC,CAAC,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,iBAAiB,CAAC,GAAG;QACzC,CAAC,CAAC,KAAK,CAAC;AACZ,CAAC;AA6CD,MAAM,OAAO,sBAAuB,SAAQ,KAAK;IAC1B,IAAI;IAAmB,OAAO;IAAnD,YAAqB,IAAY,EAAW,OAA8B;QACxE,KAAK,CACH,SAAS,IAAI,WAAW,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,gBAAgB,OAAO;aACjE,GAAG,CAAC,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC,UAAU,KAAK,QAAQ,CAAC,CAAC,CAAC,gBAAgB,CAAC,CAAC,CAAC,eAAe,CAAC,CAAC;aACnF,IAAI,CAAC,GAAG,CAAC,GAAG,CAChB,CAAC;oBALiB,IAAI;uBAAmB,OAAO;QAMjD,IAAI,CAAC,IAAI,GAAG,wBAAwB,CAAC;IACvC,CAAC;CACF;AAED,SAAS,mBAAmB,CAC1B,IAAU,EACV,OAA8B;IAE9B,OAAO,IAAI,CAAC,YAAY,CAAC,MAAM,CAC7B,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAC,CAC9C,CAAC;AACJ,CAAC;AAED,KAAK,UAAU,UAAU,CACvB,IAAU,EACV,OAAe,EACf,OAAsB;IAEtB,MAAM,SAAS,GAAG,MAAM,sBAAsB,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC;IACtE,MAAM,MAAM,GAAiB,EAAE,CAAC;IAChC,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IAE7B,IAAI,CAAC;QACH,MAAM,KAAK,GAAU,IAAI,UAAU,CAAC;YAClC,YAAY,EAAE,OAAO,CAAC,YAAY;YAClC,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,GAAG,EAAE,OAAO,CAAC,YAAY,EAAE;YAC3B,KAAK,EAAE,QAAQ,CAAC,IAAI,CAAC,YAAY,EAAE,SAAS,CAAC,IAAI,CAAC;YAClD,SAAS,EAAE,IAAI,cAAc,CAAC,SAAS,CAAC,IAAI,EAAE;gBAC5C,WAAW,EAAE,IAAI,CAAC,YAAY,CAAC,QAAQ,CAAC,QAAQ,CAAC;aAClD,CAAC;YACF,SAAS,EAAE,KAAK;YAChB,OAAO,EAAE,CAAC,KAAK,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC;SACvC,CAAC,CAAC;QAEH,IAAI,MAAM,GAA2B,OAAO,CAAC;QAC7C,IAAI,WAAW,GAAG,EAAE,CAAC;QACrB,IAAI,SAAS,GAAG,CAAC,CAAC;QAClB,IAAI,KAA6B,CAAC;QAClC,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,EAAE;gBAC1C,GAAG,CAAC,OAAO,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC;aACpE,CAAC,CAAC;YACH,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC;YACvB,SAAS,GAAG,MAAM,CAAC,SAAS,CAAC;YAC7B,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC;YACrB,WAAW;gBACT,MAAM,CAAC,MAAM,KAAK,WAAW,CAAC,CAAC,CAAC,MAAM,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC;QACrE,CAAC;QAAC,OAAO,KAAc,EAAE,CAAC;YACxB,yEAAyE;YACzE,mEAAmE;YACnE,MAAM,GAAG,OAAO,CAAC;YACjB,WAAW,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QACvE,CAAC;QAED,MAAM,KAAK,GAAiB;YAC1B,SAAS,EAAE,SAAS,CAAC,IAAI;YACzB,gBAAgB,EAAE,IAAI,CAAC,gBAAgB;YACvC,WAAW;YACX,SAAS;YACT,MAAM;YACN,MAAM;SACP,CAAC;QAEF,MAAM,QAAQ,GAAa,EAAE,CAAC;QAC9B,IAAI,MAAM,KAAK,WAAW,EAAE,CAAC;YAC3B,QAAQ,CAAC,IAAI,CACX,MAAM,KAAK,WAAW;gBACpB,CAAC,CAAC,WAAW,IAAI,CAAC,QAAQ,oCAAoC;gBAC9D,CAAC,CAAC,gBAAgB,MAAM,KAAK,WAAW,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,EAAE,CACrE,CAAC;QACJ,CAAC;QACD,KAAK,MAAM,IAAI,IAAI,IAAI,CAAC,OAAO,EAAE,CAAC;YAChC,MAAM,MAAM,GAAG,MAAM,YAAY,CAAC,IAAI,CAAC,CAAC,KAAK,CAAC,CAAC;YAC/C,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC;gBACnB,QAAQ,CAAC,IAAI,CAAC,GAAG,cAAc,CAAC,IAAI,CAAC,MAAM,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC;YAC9D,CAAC;QACH,CAAC;QAED,uEAAuE;QACvE,kCAAkC;QAClC,MAAM,OAAO,GAAG,IAAI,GAAG,EAAkB,CAAC;QAC1C,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;YAC3B,IAAI,KAAK,CAAC,IAAI,KAAK,sBAAsB,EAAE,CAAC;gBAC1C,MAAM,MAAM,GAAG,UAAU,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;gBAC3C,IAAI,MAAM,KAAK,SAAS,EAAE,CAAC;oBACzB,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,UAAU,EAAE,MAAM,CAAC,CAAC;gBACxC,CAAC;YACH,CAAC;QACH,CAAC;QACD,MAAM,KAAK,GAA8B,MAAM;aAC5C,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,KAAK,oBAAoB,CAAC;aACtD,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE;YACb,MAAM,GAAG,GAAG,KAIX,CAAC;YACF,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC;YAC3C,OAAO;gBACL,IAAI,EAAE,GAAG,CAAC,QAAQ;gBAClB,EAAE,EAAE,GAAG,CAAC,OAAO,KAAK,KAAK;gBACzB,GAAG,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC;aAC5C,CAAC;QACJ,CAAC,CAAC,CAAC;QAEL,OAAO;YACL,OAAO;YACP,MAAM,EAAE,QAAQ,CAAC,MAAM,KAAK,CAAC;YAC7B,MAAM;YACN,SAAS;YACT,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;YAClC,GAAG,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,CAAC;YACzC,SAAS,EAAE,CAAC,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC;YACvD,SAAS,EAAE,KAAK;YAChB,QAAQ;SACT,CAAC;IACJ,CAAC;YAAS,CAAC;QACT,MAAM,SAAS,CAAC,OAAO,EAAE,CAAC;IAC5B,CAAC;AACH,CAAC;AAED;;;;;;GAMG;AACH,MAAM,CAAC,KAAK,UAAU,OAAO,CAC3B,IAAU,EACV,OAAsB;IAEtB,MAAM,OAAO,GAAG,mBAAmB,CAAC,IAAI,EAAE,OAAO,CAAC,YAAY,CAAC,CAAC;IAChE,IAAI,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACvB,MAAM,IAAI,sBAAsB,CAAC,IAAI,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;IACvD,CAAC;IAED,MAAM,KAAK,GAAG,OAAO,CAAC,QAAQ,IAAI,IAAI,CAAC,QAAQ,CAAC;IAChD,MAAM,QAAQ,GAAoB,EAAE,CAAC;IACrC,KAAK,IAAI,KAAK,GAAG,CAAC,EAAE,KAAK,IAAI,KAAK,EAAE,KAAK,IAAI,CAAC,EAAE,CAAC;QAC/C,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,IAAI,EAAE,KAAK,EAAE,OAAO,CAAC,CAAC;QACtD,QAAQ,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACtB,OAAO,CAAC,SAAS,EAAE,CAAC,MAAM,EAAE,IAAI,CAAC,CAAC;IACpC,CAAC;IAED,OAAO;QACL,IAAI,EAAE,IAAI,CAAC,IAAI;QACf,QAAQ;QACR,MAAM,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,MAAM;QAC3D,KAAK;KACN,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,QAAQ,CAC5B,KAAsB,EACtB,OAAsB;IAEtB,MAAM,OAAO,GAAiB,EAAE,CAAC;IACjC,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,OAAO,CAAC,IAAI,CAAC,MAAM,OAAO,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC,CAAC;IAC7C,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC"}
@@ -1 +1 @@
1
- {"version":3,"file":"task.d.ts","sourceRoot":"","sources":["../../src/evals/task.ts"],"names":[],"mappings":"AAGA,qBAAa,SAAU,SAAQ,KAAK;IAClC,YAAY,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;QAAE,KAAK,CAAC,EAAE,OAAO,CAAA;KAAE,EAMzD;CACF;AAED,yEAAyE;AACzE,MAAM,MAAM,UAAU,GAAG,QAAQ,GAAG,OAAO,CAAC;AAE5C,MAAM,WAAW,UAAU;IACzB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,EAAE,GAAG,EAAE,MAAM,GAAG,OAAO,CAAC;CACjC;AAED,MAAM,WAAW,IAAI;IACnB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,YAAY,EAAE,SAAS,UAAU,EAAE,CAAC;IAC7C,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,OAAO,EAAE,SAAS,UAAU,EAAE,CAAC;IACxC,iEAAiE;IACjE,QAAQ,CAAC,gBAAgB,EAAE,MAAM,CAAC;IAClC,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;CAC5B;AA6CD,wBAAgB,SAAS,CACvB,KAAK,EAAE,OAAO,EACd,SAAS,EAAE,MAAM,EACjB,gBAAgB,EAAE,MAAM,GACvB,IAAI,CA2CN;AAED,wBAAsB,QAAQ,CAAC,SAAS,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAgB/D;AAED,6DAA6D;AAC7D,wBAAsB,SAAS,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC,SAAS,IAAI,EAAE,CAAC,CAwBtE"}
1
+ {"version":3,"file":"task.d.ts","sourceRoot":"","sources":["../../src/evals/task.ts"],"names":[],"mappings":"AAIA,qBAAa,SAAU,SAAQ,KAAK;IAClC,YAAY,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;QAAE,KAAK,CAAC,EAAE,OAAO,CAAA;KAAE,EAMzD;CACF;AAED,yEAAyE;AACzE,MAAM,MAAM,UAAU,GAAG,QAAQ,GAAG,OAAO,CAAC;AAE5C,MAAM,WAAW,UAAU;IACzB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,EAAE,GAAG,EAAE,MAAM,GAAG,OAAO,CAAC;CACjC;AAED,MAAM,WAAW,IAAI;IACnB,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,YAAY,EAAE,SAAS,UAAU,EAAE,CAAC;IAC7C,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,OAAO,EAAE,SAAS,UAAU,EAAE,CAAC;IACxC,iEAAiE;IACjE,QAAQ,CAAC,gBAAgB,EAAE,MAAM,CAAC;IAClC,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;CAC5B;AAoFD,wBAAgB,SAAS,CACvB,KAAK,EAAE,OAAO,EACd,SAAS,EAAE,MAAM,EACjB,gBAAgB,EAAE,MAAM,GACvB,IAAI,CA8CN;AAED,wBAAsB,QAAQ,CAAC,SAAS,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAgB/D;AAED,6DAA6D;AAC7D,wBAAsB,SAAS,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC,SAAS,IAAI,EAAE,CAAC,CAwBtE"}