@lesliechan721/agent-skill-eval 0.1.0-beta.4 → 0.1.0-beta.41

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 (117) hide show
  1. package/dist/bin.js +167 -39
  2. package/dist/bin.js.map +1 -1
  3. package/dist/codex-eval-app-server.d.ts +208 -0
  4. package/dist/codex-eval-app-server.js +1382 -0
  5. package/dist/codex-eval-app-server.js.map +1 -0
  6. package/dist/codex-eval-cases.d.ts +43 -0
  7. package/dist/codex-eval-cases.js +329 -0
  8. package/dist/codex-eval-cases.js.map +1 -0
  9. package/dist/codex-eval-isolation.d.ts +79 -0
  10. package/dist/codex-eval-isolation.js +420 -0
  11. package/dist/codex-eval-isolation.js.map +1 -0
  12. package/dist/codex-eval-process-tree.d.ts +42 -0
  13. package/dist/codex-eval-process-tree.js +860 -0
  14. package/dist/codex-eval-process-tree.js.map +1 -0
  15. package/dist/codex-eval-runtime-config.d.ts +43 -0
  16. package/dist/codex-eval-runtime-config.js +607 -0
  17. package/dist/codex-eval-runtime-config.js.map +1 -0
  18. package/dist/codex-trigger-eval.d.ts +84 -0
  19. package/dist/codex-trigger-eval.js +807 -0
  20. package/dist/codex-trigger-eval.js.map +1 -0
  21. package/dist/codex-workflow-evaluation.d.ts +91 -0
  22. package/dist/codex-workflow-evaluation.js +309 -0
  23. package/dist/codex-workflow-evaluation.js.map +1 -0
  24. package/dist/codex-workflow-fixtures.d.ts +37 -0
  25. package/dist/codex-workflow-fixtures.js +144 -0
  26. package/dist/codex-workflow-fixtures.js.map +1 -0
  27. package/dist/codex-workflow.d.ts +224 -0
  28. package/dist/codex-workflow.js +2198 -0
  29. package/dist/codex-workflow.js.map +1 -0
  30. package/dist/credentials.d.ts +5 -0
  31. package/dist/credentials.js +30 -0
  32. package/dist/credentials.js.map +1 -0
  33. package/dist/custom-process.d.ts +60 -0
  34. package/dist/custom-process.js +424 -0
  35. package/dist/custom-process.js.map +1 -0
  36. package/dist/eval-audit.d.ts +38 -0
  37. package/dist/eval-audit.js +71 -0
  38. package/dist/eval-audit.js.map +1 -0
  39. package/dist/eval-file.d.ts +9 -0
  40. package/dist/eval-file.js +15 -0
  41. package/dist/eval-file.js.map +1 -0
  42. package/dist/eval-process-management.d.ts +22 -0
  43. package/dist/eval-process-management.js +90 -0
  44. package/dist/eval-process-management.js.map +1 -0
  45. package/dist/eval-result.d.ts +79 -0
  46. package/dist/eval-result.js +144 -0
  47. package/dist/eval-result.js.map +1 -0
  48. package/dist/eval-run.d.ts +27 -0
  49. package/dist/eval-run.js +566 -0
  50. package/dist/eval-run.js.map +1 -0
  51. package/dist/eval-spec.d.ts +57 -0
  52. package/dist/eval-spec.js +120 -0
  53. package/dist/eval-spec.js.map +1 -0
  54. package/dist/eval-suite-binding.d.ts +4 -0
  55. package/dist/eval-suite-binding.js +19 -0
  56. package/dist/eval-suite-binding.js.map +1 -0
  57. package/dist/index.d.ts +44 -10
  58. package/dist/index.js +22 -6
  59. package/dist/index.js.map +1 -1
  60. package/dist/instruction-v2.d.ts +53 -0
  61. package/dist/instruction-v2.js +169 -0
  62. package/dist/instruction-v2.js.map +1 -0
  63. package/dist/json-schema.js +84 -5
  64. package/dist/json-schema.js.map +1 -1
  65. package/dist/output-path.js +2 -2
  66. package/dist/process-tree.d.ts +53 -1
  67. package/dist/process-tree.js +931 -160
  68. package/dist/process-tree.js.map +1 -1
  69. package/dist/process.d.ts +2 -2
  70. package/dist/process.js +99 -34
  71. package/dist/process.js.map +1 -1
  72. package/dist/provider-doctor.d.ts +46 -0
  73. package/dist/provider-doctor.js +120 -0
  74. package/dist/provider-doctor.js.map +1 -0
  75. package/dist/providers.d.ts +85 -0
  76. package/dist/providers.js +685 -0
  77. package/dist/providers.js.map +1 -0
  78. package/dist/run-paths.d.ts +8 -0
  79. package/dist/run-paths.js +71 -0
  80. package/dist/run-paths.js.map +1 -0
  81. package/dist/runtime-skill.d.ts +6 -0
  82. package/dist/runtime-skill.js +87 -3
  83. package/dist/runtime-skill.js.map +1 -1
  84. package/dist/schemas/digests.json +7 -3
  85. package/dist/schemas/eval-audit.schema.json +64 -0
  86. package/dist/schemas/eval-result.schema.json +232 -0
  87. package/dist/schemas/eval-spec.schema.json +109 -0
  88. package/dist/schemas/instruction-v2-receipt.schema.json +24 -0
  89. package/dist/schemas/instruction-v2-request.schema.json +51 -0
  90. package/dist/schemas/workflow-evaluation.schema.json +38 -0
  91. package/dist/schemas/workflow-fixture-action.schema.json +15 -0
  92. package/dist/schemas.d.ts +1 -3
  93. package/dist/schemas.js +22 -20
  94. package/dist/schemas.js.map +1 -1
  95. package/dist/selected-cases.d.ts +30 -0
  96. package/dist/selected-cases.js +388 -0
  97. package/dist/selected-cases.js.map +1 -0
  98. package/dist/types.d.ts +2 -34
  99. package/dist/windows-job.d.ts +26 -0
  100. package/dist/windows-job.js +97 -0
  101. package/dist/windows-job.js.map +1 -0
  102. package/package.json +17 -3
  103. package/dist/bundle.d.ts +0 -93
  104. package/dist/bundle.js +0 -216
  105. package/dist/bundle.js.map +0 -1
  106. package/dist/capabilities.d.ts +0 -2
  107. package/dist/capabilities.js +0 -38
  108. package/dist/capabilities.js.map +0 -1
  109. package/dist/plan.d.ts +0 -79
  110. package/dist/plan.js +0 -285
  111. package/dist/plan.js.map +0 -1
  112. package/dist/run.d.ts +0 -2
  113. package/dist/run.js +0 -592
  114. package/dist/run.js.map +0 -1
  115. package/dist/schemas/aw-skill-eval-execution-plan.schema.json +0 -220
  116. package/dist/schemas/eval-run-bundle.schema.json +0 -408
  117. package/dist/schemas/eval-runtime-capabilities.schema.json +0 -85
@@ -0,0 +1,84 @@
1
+ import { type CodexEvalProcessManagementEvidence } from './codex-eval-app-server.js';
2
+ import { type CodexEvalRuntimeConfig } from './codex-eval-runtime-config.js';
3
+ import { type AwSkillEvalTriggerCase, type AwSkillEvalTriggerSuiteCaseContract } from './codex-eval-cases.js';
4
+ import type { ProcessTreeDiagnostic } from './process-tree.js';
5
+ interface TriggerObservation {
6
+ loaded: boolean;
7
+ spawned: boolean;
8
+ loadEvidence: string | null;
9
+ spawnEvidence: string | null;
10
+ }
11
+ export interface TriggerRunEvidence {
12
+ loaded: boolean;
13
+ first_loaded_skill: string | null;
14
+ all_loaded_skills: string[];
15
+ load_evidence_by_skill: Record<string, string>;
16
+ spawned: boolean;
17
+ load_evidence: string | null;
18
+ spawn_evidence: string | null;
19
+ spawn_calls: Array<Record<string, unknown>>;
20
+ termination: string;
21
+ exit_code: number | null;
22
+ signal: NodeJS.Signals | null;
23
+ execution_error: boolean;
24
+ fixture_unchanged: boolean;
25
+ request_user_input_called: boolean;
26
+ waiting_on_user_input: boolean;
27
+ final_response_emitted: boolean;
28
+ stderr?: string;
29
+ final_message: string | null;
30
+ duration_ms: number;
31
+ usage: unknown;
32
+ event_trace: unknown[];
33
+ process_tree_snapshots?: ProcessTreeDiagnostic[];
34
+ diagnostics: string[];
35
+ }
36
+ interface TriggerCaseSummary {
37
+ id: string;
38
+ split: AwSkillEvalTriggerCase['split'];
39
+ query: string;
40
+ expected: {
41
+ load_skill?: boolean;
42
+ loaded_skills?: string[];
43
+ first_loaded_skill?: string;
44
+ spawn_agent?: boolean;
45
+ };
46
+ observed: {
47
+ load_rate: number;
48
+ skill_load_rates: Record<string, number>;
49
+ route_match_rate?: number;
50
+ first_route_match_rate?: number;
51
+ route_counts?: Record<string, number>;
52
+ multiple_skill_loads: number;
53
+ spawn_rate?: number;
54
+ execution_errors: number;
55
+ };
56
+ complete: boolean;
57
+ passed: boolean;
58
+ runs: TriggerRunEvidence[];
59
+ }
60
+ export declare function inspectTriggerAppServerFrame(frame: Record<string, unknown>, skillPath: string): Partial<TriggerObservation> & {
61
+ finalMessage?: string;
62
+ usage?: unknown;
63
+ turnStatus?: string;
64
+ spawnCall?: Record<string, unknown>;
65
+ requestUserInput?: boolean;
66
+ };
67
+ export declare function inspectTriggerAppServerConfigWarning(frame: Record<string, unknown>): string | null;
68
+ export declare function runCodexTriggerProviderCase(options: {
69
+ suite: AwSkillEvalTriggerSuiteCaseContract;
70
+ evalCase: AwSkillEvalTriggerCase;
71
+ subjectRoot: string;
72
+ fixtureRoot: string;
73
+ codexBin: string;
74
+ model?: string;
75
+ codexRuntime: CodexEvalRuntimeConfig;
76
+ sourceEnv?: NodeJS.ProcessEnv;
77
+ processManagement?: CodexEvalProcessManagementEvidence;
78
+ workRoot?: string;
79
+ manageTimeout?: boolean;
80
+ signal?: AbortSignal;
81
+ timeoutMs?: number;
82
+ }): Promise<TriggerRunEvidence>;
83
+ export declare function summarizeTriggerCase(suite: AwSkillEvalTriggerSuiteCaseContract, evalCase: AwSkillEvalTriggerCase, runs: TriggerRunEvidence[]): TriggerCaseSummary;
84
+ export {};