oh-my-knowledge 0.0.1 → 0.19.0

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 (216) hide show
  1. package/LICENSE +21 -0
  2. package/README.md +600 -322
  3. package/README.zh.md +724 -0
  4. package/dist/src/analysis/coverage-analyzer.d.ts +57 -0
  5. package/dist/src/analysis/coverage-analyzer.d.ts.map +1 -0
  6. package/dist/src/analysis/coverage-analyzer.js +262 -0
  7. package/dist/src/analysis/coverage-analyzer.js.map +1 -0
  8. package/dist/src/analysis/gap-analyzer.d.ts +114 -0
  9. package/dist/src/analysis/gap-analyzer.d.ts.map +1 -0
  10. package/dist/src/analysis/gap-analyzer.js +439 -0
  11. package/dist/src/analysis/gap-analyzer.js.map +1 -0
  12. package/dist/src/analysis/hedging-classifier.d.ts +34 -0
  13. package/dist/src/analysis/hedging-classifier.d.ts.map +1 -0
  14. package/dist/src/analysis/hedging-classifier.js +144 -0
  15. package/dist/src/analysis/hedging-classifier.js.map +1 -0
  16. package/dist/src/analysis/report-diagnostics.d.ts +9 -0
  17. package/dist/src/analysis/report-diagnostics.d.ts.map +1 -0
  18. package/dist/src/analysis/report-diagnostics.js +631 -0
  19. package/dist/src/analysis/report-diagnostics.js.map +1 -0
  20. package/dist/src/authoring/evolver.d.ts +64 -0
  21. package/dist/src/authoring/evolver.d.ts.map +1 -0
  22. package/dist/src/authoring/evolver.js +275 -0
  23. package/dist/src/authoring/evolver.js.map +1 -0
  24. package/dist/src/authoring/generator.d.ts +13 -0
  25. package/dist/src/authoring/generator.d.ts.map +1 -0
  26. package/dist/src/authoring/generator.js +60 -0
  27. package/dist/src/authoring/generator.js.map +1 -0
  28. package/dist/src/cli.d.ts +3 -0
  29. package/dist/src/cli.d.ts.map +1 -0
  30. package/dist/src/cli.js +1012 -0
  31. package/dist/src/cli.js.map +1 -0
  32. package/dist/src/eval-core/cache.d.ts +11 -0
  33. package/dist/src/eval-core/cache.d.ts.map +1 -0
  34. package/dist/src/eval-core/cache.js +53 -0
  35. package/dist/src/eval-core/cache.js.map +1 -0
  36. package/dist/src/eval-core/ci-gates.d.ts +17 -0
  37. package/dist/src/eval-core/ci-gates.d.ts.map +1 -0
  38. package/dist/src/eval-core/ci-gates.js +42 -0
  39. package/dist/src/eval-core/ci-gates.js.map +1 -0
  40. package/dist/src/eval-core/dependency-checker.d.ts +37 -0
  41. package/dist/src/eval-core/dependency-checker.d.ts.map +1 -0
  42. package/dist/src/eval-core/dependency-checker.js +280 -0
  43. package/dist/src/eval-core/dependency-checker.js.map +1 -0
  44. package/dist/src/eval-core/evaluation-execution.d.ts +26 -0
  45. package/dist/src/eval-core/evaluation-execution.d.ts.map +1 -0
  46. package/dist/src/eval-core/evaluation-execution.js +196 -0
  47. package/dist/src/eval-core/evaluation-execution.js.map +1 -0
  48. package/dist/src/eval-core/evaluation-job.d.ts +50 -0
  49. package/dist/src/eval-core/evaluation-job.d.ts.map +1 -0
  50. package/dist/src/eval-core/evaluation-job.js +114 -0
  51. package/dist/src/eval-core/evaluation-job.js.map +1 -0
  52. package/dist/src/eval-core/evaluation-reporting.d.ts +28 -0
  53. package/dist/src/eval-core/evaluation-reporting.d.ts.map +1 -0
  54. package/dist/src/eval-core/evaluation-reporting.js +123 -0
  55. package/dist/src/eval-core/evaluation-reporting.js.map +1 -0
  56. package/dist/src/eval-core/execution-strategy.d.ts +11 -0
  57. package/dist/src/eval-core/execution-strategy.d.ts.map +1 -0
  58. package/dist/src/eval-core/execution-strategy.js +121 -0
  59. package/dist/src/eval-core/execution-strategy.js.map +1 -0
  60. package/dist/src/eval-core/fact-checker.d.ts +25 -0
  61. package/dist/src/eval-core/fact-checker.d.ts.map +1 -0
  62. package/dist/src/eval-core/fact-checker.js +66 -0
  63. package/dist/src/eval-core/fact-checker.js.map +1 -0
  64. package/dist/src/eval-core/schema.d.ts +30 -0
  65. package/dist/src/eval-core/schema.d.ts.map +1 -0
  66. package/dist/src/eval-core/schema.js +195 -0
  67. package/dist/src/eval-core/schema.js.map +1 -0
  68. package/dist/src/eval-core/statistics.d.ts +86 -0
  69. package/dist/src/eval-core/statistics.d.ts.map +1 -0
  70. package/dist/src/eval-core/statistics.js +210 -0
  71. package/dist/src/eval-core/statistics.js.map +1 -0
  72. package/dist/src/eval-core/task-planner.d.ts +4 -0
  73. package/dist/src/eval-core/task-planner.d.ts.map +1 -0
  74. package/dist/src/eval-core/task-planner.js +33 -0
  75. package/dist/src/eval-core/task-planner.js.map +1 -0
  76. package/dist/src/eval-workflows/each-evaluation-workflow.d.ts +137 -0
  77. package/dist/src/eval-workflows/each-evaluation-workflow.d.ts.map +1 -0
  78. package/dist/src/eval-workflows/each-evaluation-workflow.js +174 -0
  79. package/dist/src/eval-workflows/each-evaluation-workflow.js.map +1 -0
  80. package/dist/src/eval-workflows/evaluation-pipeline.d.ts +42 -0
  81. package/dist/src/eval-workflows/evaluation-pipeline.d.ts.map +1 -0
  82. package/dist/src/eval-workflows/evaluation-pipeline.js +218 -0
  83. package/dist/src/eval-workflows/evaluation-pipeline.js.map +1 -0
  84. package/dist/src/eval-workflows/evaluation-preparation.d.ts +64 -0
  85. package/dist/src/eval-workflows/evaluation-preparation.d.ts.map +1 -0
  86. package/dist/src/eval-workflows/evaluation-preparation.js +90 -0
  87. package/dist/src/eval-workflows/evaluation-preparation.js.map +1 -0
  88. package/dist/src/eval-workflows/run-evaluation.d.ts +111 -0
  89. package/dist/src/eval-workflows/run-evaluation.d.ts.map +1 -0
  90. package/dist/src/eval-workflows/run-evaluation.js +272 -0
  91. package/dist/src/eval-workflows/run-evaluation.js.map +1 -0
  92. package/dist/src/executors/anthropic-api.d.ts +3 -0
  93. package/dist/src/executors/anthropic-api.d.ts.map +1 -0
  94. package/dist/src/executors/anthropic-api.js +43 -0
  95. package/dist/src/executors/anthropic-api.js.map +1 -0
  96. package/dist/src/executors/claude-cli.d.ts +3 -0
  97. package/dist/src/executors/claude-cli.d.ts.map +1 -0
  98. package/dist/src/executors/claude-cli.js +103 -0
  99. package/dist/src/executors/claude-cli.js.map +1 -0
  100. package/dist/src/executors/claude-sdk-trace.d.ts +10 -0
  101. package/dist/src/executors/claude-sdk-trace.d.ts.map +1 -0
  102. package/dist/src/executors/claude-sdk-trace.js +100 -0
  103. package/dist/src/executors/claude-sdk-trace.js.map +1 -0
  104. package/dist/src/executors/claude-sdk.d.ts +3 -0
  105. package/dist/src/executors/claude-sdk.d.ts.map +1 -0
  106. package/dist/src/executors/claude-sdk.js +160 -0
  107. package/dist/src/executors/claude-sdk.js.map +1 -0
  108. package/dist/src/executors/gemini.d.ts +3 -0
  109. package/dist/src/executors/gemini.d.ts.map +1 -0
  110. package/dist/src/executors/gemini.js +78 -0
  111. package/dist/src/executors/gemini.js.map +1 -0
  112. package/dist/src/executors/index.d.ts +7 -0
  113. package/dist/src/executors/index.d.ts.map +1 -0
  114. package/dist/src/executors/index.js +22 -0
  115. package/dist/src/executors/index.js.map +1 -0
  116. package/dist/src/executors/openai-api.d.ts +3 -0
  117. package/dist/src/executors/openai-api.d.ts.map +1 -0
  118. package/dist/src/executors/openai-api.js +40 -0
  119. package/dist/src/executors/openai-api.js.map +1 -0
  120. package/dist/src/executors/openai-cli.d.ts +3 -0
  121. package/dist/src/executors/openai-cli.d.ts.map +1 -0
  122. package/dist/src/executors/openai-cli.js +60 -0
  123. package/dist/src/executors/openai-cli.js.map +1 -0
  124. package/dist/src/executors/script.d.ts +3 -0
  125. package/dist/src/executors/script.d.ts.map +1 -0
  126. package/dist/src/executors/script.js +63 -0
  127. package/dist/src/executors/script.js.map +1 -0
  128. package/dist/src/executors/shared.d.ts +117 -0
  129. package/dist/src/executors/shared.d.ts.map +1 -0
  130. package/dist/src/executors/shared.js +49 -0
  131. package/dist/src/executors/shared.js.map +1 -0
  132. package/dist/src/grading/assertions.d.ts +18 -0
  133. package/dist/src/grading/assertions.d.ts.map +1 -0
  134. package/dist/src/grading/assertions.js +239 -0
  135. package/dist/src/grading/assertions.js.map +1 -0
  136. package/dist/src/grading/index.d.ts +26 -0
  137. package/dist/src/grading/index.d.ts.map +1 -0
  138. package/dist/src/grading/index.js +98 -0
  139. package/dist/src/grading/index.js.map +1 -0
  140. package/dist/src/grading/judge.d.ts +13 -0
  141. package/dist/src/grading/judge.d.ts.map +1 -0
  142. package/dist/src/grading/judge.js +98 -0
  143. package/dist/src/grading/judge.js.map +1 -0
  144. package/dist/src/grading/layered-scores.d.ts +13 -0
  145. package/dist/src/grading/layered-scores.d.ts.map +1 -0
  146. package/dist/src/grading/layered-scores.js +62 -0
  147. package/dist/src/grading/layered-scores.js.map +1 -0
  148. package/dist/src/inputs/eval-config.d.ts +13 -0
  149. package/dist/src/inputs/eval-config.d.ts.map +1 -0
  150. package/dist/src/inputs/eval-config.js +136 -0
  151. package/dist/src/inputs/eval-config.js.map +1 -0
  152. package/dist/src/inputs/load-samples.d.ts +16 -0
  153. package/dist/src/inputs/load-samples.d.ts.map +1 -0
  154. package/dist/src/inputs/load-samples.js +51 -0
  155. package/dist/src/inputs/load-samples.js.map +1 -0
  156. package/dist/src/inputs/mcp-resolver.d.ts +50 -0
  157. package/dist/src/inputs/mcp-resolver.d.ts.map +1 -0
  158. package/dist/src/inputs/mcp-resolver.js +307 -0
  159. package/dist/src/inputs/mcp-resolver.js.map +1 -0
  160. package/dist/src/inputs/skill-loader.d.ts +18 -0
  161. package/dist/src/inputs/skill-loader.d.ts.map +1 -0
  162. package/dist/src/inputs/skill-loader.js +222 -0
  163. package/dist/src/inputs/skill-loader.js.map +1 -0
  164. package/dist/src/inputs/url-fetcher.d.ts +19 -0
  165. package/dist/src/inputs/url-fetcher.d.ts.map +1 -0
  166. package/dist/src/inputs/url-fetcher.js +241 -0
  167. package/dist/src/inputs/url-fetcher.js.map +1 -0
  168. package/dist/src/observability/skill-health-analyzer.d.ts +83 -0
  169. package/dist/src/observability/skill-health-analyzer.d.ts.map +1 -0
  170. package/dist/src/observability/skill-health-analyzer.js +238 -0
  171. package/dist/src/observability/skill-health-analyzer.js.map +1 -0
  172. package/dist/src/observability/trace-adapter.d.ts +75 -0
  173. package/dist/src/observability/trace-adapter.d.ts.map +1 -0
  174. package/dist/src/observability/trace-adapter.js +367 -0
  175. package/dist/src/observability/trace-adapter.js.map +1 -0
  176. package/dist/src/renderer/html-renderer.d.ts +9 -0
  177. package/dist/src/renderer/html-renderer.d.ts.map +1 -0
  178. package/dist/src/renderer/html-renderer.js +360 -0
  179. package/dist/src/renderer/html-renderer.js.map +1 -0
  180. package/dist/src/renderer/layout.d.ts +13 -0
  181. package/dist/src/renderer/layout.d.ts.map +1 -0
  182. package/dist/src/renderer/layout.js +582 -0
  183. package/dist/src/renderer/layout.js.map +1 -0
  184. package/dist/src/renderer/skill-health-renderer.d.ts +20 -0
  185. package/dist/src/renderer/skill-health-renderer.d.ts.map +1 -0
  186. package/dist/src/renderer/skill-health-renderer.js +253 -0
  187. package/dist/src/renderer/skill-health-renderer.js.map +1 -0
  188. package/dist/src/renderer/summary.d.ts +21 -0
  189. package/dist/src/renderer/summary.d.ts.map +1 -0
  190. package/dist/src/renderer/summary.js +1072 -0
  191. package/dist/src/renderer/summary.js.map +1 -0
  192. package/dist/src/renderer/table.d.ts +3 -0
  193. package/dist/src/renderer/table.d.ts.map +1 -0
  194. package/dist/src/renderer/table.js +134 -0
  195. package/dist/src/renderer/table.js.map +1 -0
  196. package/dist/src/renderer/trends.d.ts +6 -0
  197. package/dist/src/renderer/trends.d.ts.map +1 -0
  198. package/dist/src/renderer/trends.js +127 -0
  199. package/dist/src/renderer/trends.js.map +1 -0
  200. package/dist/src/server/job-store.d.ts +4 -0
  201. package/dist/src/server/job-store.d.ts.map +1 -0
  202. package/dist/src/server/job-store.js +68 -0
  203. package/dist/src/server/job-store.js.map +1 -0
  204. package/dist/src/server/report-server.d.ts +17 -0
  205. package/dist/src/server/report-server.d.ts.map +1 -0
  206. package/dist/src/server/report-server.js +591 -0
  207. package/dist/src/server/report-server.js.map +1 -0
  208. package/dist/src/server/report-store.d.ts +44 -0
  209. package/dist/src/server/report-store.d.ts.map +1 -0
  210. package/dist/src/server/report-store.js +190 -0
  211. package/dist/src/server/report-store.js.map +1 -0
  212. package/dist/src/types.d.ts +559 -0
  213. package/dist/src/types.d.ts.map +1 -0
  214. package/dist/src/types.js +2 -0
  215. package/dist/src/types.js.map +1 -0
  216. package/package.json +53 -5
@@ -0,0 +1,49 @@
1
+ import { execFile } from 'node:child_process';
2
+ import { existsSync } from 'node:fs';
3
+ import { delimiter, join } from 'node:path';
4
+ import { promisify } from 'node:util';
5
+ export const execFileAsync = promisify(execFile);
6
+ export const DEFAULT_MODEL = 'sonnet';
7
+ export const JUDGE_MODEL = 'haiku';
8
+ export const DEFAULT_TIMEOUT_MS = 120_000;
9
+ export const MAX_BUFFER = 10 * 1024 * 1024;
10
+ export function asErrorLike(err) {
11
+ return typeof err === 'object' && err !== null ? err : {};
12
+ }
13
+ export function errorMessage(err, fallback = 'unknown error') {
14
+ const details = asErrorLike(err);
15
+ return details.message || fallback;
16
+ }
17
+ export function parseJson(content) {
18
+ return JSON.parse(content);
19
+ }
20
+ export function buildExecEnv(skillDir) {
21
+ const proxyUrl = process.env.CCV_PROXY_URL || undefined;
22
+ const env = proxyUrl
23
+ ? { ...process.env, ANTHROPIC_BASE_URL: proxyUrl }
24
+ : { ...process.env };
25
+ if (skillDir) {
26
+ const nodeBin = join(skillDir, 'node_modules', '.bin');
27
+ if (existsSync(nodeBin)) {
28
+ env.PATH = `${nodeBin}${env.PATH ? delimiter + env.PATH : ''}`;
29
+ }
30
+ }
31
+ return env;
32
+ }
33
+ export function timeoutExecResult(timeoutMs, durationMs) {
34
+ return {
35
+ ok: false,
36
+ error: `execution timed out after ${timeoutMs / 1000}s`,
37
+ durationMs,
38
+ durationApiMs: 0,
39
+ inputTokens: 0,
40
+ outputTokens: 0,
41
+ cacheReadTokens: 0,
42
+ cacheCreationTokens: 0,
43
+ costUSD: 0,
44
+ output: null,
45
+ stopReason: 'timeout',
46
+ numTurns: 0,
47
+ };
48
+ }
49
+ //# sourceMappingURL=shared.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"shared.js","sourceRoot":"","sources":["../../../src/executors/shared.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,oBAAoB,CAAC;AAC9C,OAAO,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AACrC,OAAO,EAAE,SAAS,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAC5C,OAAO,EAAE,SAAS,EAAE,MAAM,WAAW,CAAC;AAGtC,MAAM,CAAC,MAAM,aAAa,GAAG,SAAS,CAAC,QAAQ,CAAC,CAAC;AAEjD,MAAM,CAAC,MAAM,aAAa,GAAG,QAAQ,CAAC;AACtC,MAAM,CAAC,MAAM,WAAW,GAAG,OAAO,CAAC;AACnC,MAAM,CAAC,MAAM,kBAAkB,GAAG,OAAO,CAAC;AAC1C,MAAM,CAAC,MAAM,UAAU,GAAG,EAAE,GAAG,IAAI,GAAG,IAAI,CAAC;AAmG3C,MAAM,UAAU,WAAW,CAAC,GAAY;IACtC,OAAO,OAAO,GAAG,KAAK,QAAQ,IAAI,GAAG,KAAK,IAAI,CAAC,CAAC,CAAC,GAAwB,CAAC,CAAC,CAAC,EAAE,CAAC;AACjF,CAAC;AAED,MAAM,UAAU,YAAY,CAAC,GAAY,EAAE,WAAmB,eAAe;IAC3E,MAAM,OAAO,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC;IACjC,OAAO,OAAO,CAAC,OAAO,IAAI,QAAQ,CAAC;AACrC,CAAC;AAED,MAAM,UAAU,SAAS,CAAI,OAAe;IAC1C,OAAO,IAAI,CAAC,KAAK,CAAC,OAAO,CAAM,CAAC;AAClC,CAAC;AAED,MAAM,UAAU,YAAY,CAAC,QAAwB;IACnD,MAAM,QAAQ,GAAG,OAAO,CAAC,GAAG,CAAC,aAAa,IAAI,SAAS,CAAC;IACxD,MAAM,GAAG,GAAsB,QAAQ;QACrC,CAAC,CAAC,EAAE,GAAG,OAAO,CAAC,GAAG,EAAE,kBAAkB,EAAE,QAAQ,EAAE;QAClD,CAAC,CAAC,EAAE,GAAG,OAAO,CAAC,GAAG,EAAE,CAAC;IAEvB,IAAI,QAAQ,EAAE,CAAC;QACb,MAAM,OAAO,GAAG,IAAI,CAAC,QAAQ,EAAE,cAAc,EAAE,MAAM,CAAC,CAAC;QACvD,IAAI,UAAU,CAAC,OAAO,CAAC,EAAE,CAAC;YACxB,GAAG,CAAC,IAAI,GAAG,GAAG,OAAO,GAAG,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,GAAG,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC;QACjE,CAAC;IACH,CAAC;IAED,OAAO,GAAG,CAAC;AACb,CAAC;AAED,MAAM,UAAU,iBAAiB,CAAC,SAAiB,EAAE,UAAkB;IACrE,OAAO;QACL,EAAE,EAAE,KAAK;QACT,KAAK,EAAE,6BAA6B,SAAS,GAAG,IAAI,GAAG;QACvD,UAAU;QACV,aAAa,EAAE,CAAC;QAChB,WAAW,EAAE,CAAC;QACd,YAAY,EAAE,CAAC;QACf,eAAe,EAAE,CAAC;QAClB,mBAAmB,EAAE,CAAC;QACtB,OAAO,EAAE,CAAC;QACV,MAAM,EAAE,IAAI;QACZ,UAAU,EAAE,SAAS;QACrB,QAAQ,EAAE,CAAC;KACZ,CAAC;AACJ,CAAC"}
@@ -0,0 +1,18 @@
1
+ import type { Assertion, AssertionResults, ExecutorFn, Sample, ToolCallInfo } from '../types.js';
2
+ export declare const ASYNC_ASSERTION_TYPES: Set<string>;
3
+ export interface AsyncAssertionContext {
4
+ executor: ExecutorFn;
5
+ judgeModel: string;
6
+ sample: Sample;
7
+ samplesDir: string;
8
+ }
9
+ export declare function ratioToScore(ratio: number): number;
10
+ export declare function validateJsonSchema(data: unknown, schema: Record<string, unknown>): boolean;
11
+ export declare function runAssertions(output: string, assertions: Assertion[], context?: {
12
+ costUSD?: number;
13
+ durationMs?: number;
14
+ numTurns?: number;
15
+ toolCalls?: ToolCallInfo[];
16
+ }): AssertionResults;
17
+ export declare function runAsyncAssertions(output: string, assertions: Assertion[], { executor, judgeModel, sample, samplesDir }: AsyncAssertionContext): Promise<AssertionResults>;
18
+ //# sourceMappingURL=assertions.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"assertions.d.ts","sourceRoot":"","sources":["../../../src/grading/assertions.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,SAAS,EAAmB,gBAAgB,EAAE,UAAU,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAMlH,eAAO,MAAM,qBAAqB,aAA6C,CAAC;AAEhF,MAAM,WAAW,qBAAqB;IACpC,QAAQ,EAAE,UAAU,CAAC;IACrB,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,EAAE,MAAM,CAAC;IACf,UAAU,EAAE,MAAM,CAAC;CACpB;AAwBD,wBAAgB,YAAY,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,CAElD;AAED,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,OAAO,CAQ1F;AAED,wBAAgB,aAAa,CAAC,MAAM,EAAE,MAAM,EAAE,UAAU,EAAE,SAAS,EAAE,EAAE,OAAO,GAAE;IAAE,OAAO,CAAC,EAAE,MAAM,CAAC;IAAC,UAAU,CAAC,EAAE,MAAM,CAAC;IAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAAC,SAAS,CAAC,EAAE,YAAY,EAAE,CAAA;CAAO,GAAG,gBAAgB,CAqI/L;AAED,wBAAsB,kBAAkB,CAAC,MAAM,EAAE,MAAM,EAAE,UAAU,EAAE,SAAS,EAAE,EAAE,EAAE,QAAQ,EAAE,UAAU,EAAE,MAAM,EAAE,UAAU,EAAE,EAAE,qBAAqB,GAAG,OAAO,CAAC,gBAAgB,CAAC,CAkFhL"}
@@ -0,0 +1,239 @@
1
+ import { resolve } from 'node:path';
2
+ import _Ajv from 'ajv';
3
+ const Ajv = _Ajv.default ?? _Ajv;
4
+ const ajv = new Ajv();
5
+ const CUSTOM_ASSERTION_TIMEOUT_MS = 30_000;
6
+ export const ASYNC_ASSERTION_TYPES = new Set(['semantic_similarity', 'custom']);
7
+ function getErrorMessage(err) {
8
+ return err instanceof Error ? err.message : String(err);
9
+ }
10
+ export function ratioToScore(ratio) {
11
+ return Number((1 + ratio * 4).toFixed(2));
12
+ }
13
+ export function validateJsonSchema(data, schema) {
14
+ if (!schema || typeof schema !== 'object')
15
+ return true;
16
+ try {
17
+ const validate = ajv.compile(schema);
18
+ return validate(data);
19
+ }
20
+ catch {
21
+ return false;
22
+ }
23
+ }
24
+ export function runAssertions(output, assertions, context = {}) {
25
+ const outputLower = output.toLowerCase();
26
+ const details = [];
27
+ const toolCalls = context.toolCalls || [];
28
+ const toolNames = toolCalls.map((tc) => tc.tool.toLowerCase());
29
+ for (const assertion of assertions) {
30
+ const weight = assertion.weight ?? 1;
31
+ let passed = false;
32
+ switch (assertion.type) {
33
+ case 'contains':
34
+ passed = outputLower.includes(String(assertion.value).toLowerCase());
35
+ break;
36
+ case 'not_contains':
37
+ passed = !outputLower.includes(String(assertion.value).toLowerCase());
38
+ break;
39
+ case 'regex': {
40
+ const flags = assertion.flags || 'i';
41
+ const re = new RegExp(assertion.pattern, flags);
42
+ passed = re.test(output);
43
+ break;
44
+ }
45
+ case 'min_length':
46
+ passed = output.length >= assertion.value;
47
+ break;
48
+ case 'max_length':
49
+ passed = output.length <= assertion.value;
50
+ break;
51
+ case 'json_valid':
52
+ try {
53
+ JSON.parse(output);
54
+ passed = true;
55
+ }
56
+ catch {
57
+ passed = false;
58
+ }
59
+ break;
60
+ case 'json_schema':
61
+ try {
62
+ const data = JSON.parse(output);
63
+ passed = validateJsonSchema(data, assertion.schema);
64
+ }
65
+ catch {
66
+ passed = false;
67
+ }
68
+ break;
69
+ case 'starts_with':
70
+ passed = outputLower.startsWith(String(assertion.value).toLowerCase());
71
+ break;
72
+ case 'ends_with':
73
+ passed = outputLower.endsWith(String(assertion.value).toLowerCase());
74
+ break;
75
+ case 'equals':
76
+ passed = output.trim() === String(assertion.value).trim();
77
+ break;
78
+ case 'not_equals':
79
+ passed = output.trim() !== String(assertion.value).trim();
80
+ break;
81
+ case 'word_count_min':
82
+ passed = output.split(/\s+/).filter(Boolean).length >= assertion.value;
83
+ break;
84
+ case 'word_count_max':
85
+ passed = output.split(/\s+/).filter(Boolean).length <= assertion.value;
86
+ break;
87
+ case 'contains_all':
88
+ passed = (assertion.values || []).every((v) => outputLower.includes(String(v).toLowerCase()));
89
+ break;
90
+ case 'contains_any':
91
+ passed = (assertion.values || []).some((v) => outputLower.includes(String(v).toLowerCase()));
92
+ break;
93
+ case 'cost_max':
94
+ passed = (context.costUSD ?? Infinity) <= assertion.value;
95
+ break;
96
+ case 'latency_max':
97
+ passed = (context.durationMs ?? Infinity) <= assertion.value;
98
+ break;
99
+ case 'turns_max':
100
+ passed = (context.numTurns ?? Infinity) <= assertion.value;
101
+ break;
102
+ case 'turns_min':
103
+ passed = (context.numTurns ?? 0) >= assertion.value;
104
+ break;
105
+ case 'tools_called':
106
+ passed = (assertion.values || []).every((v) => toolNames.includes(String(v).toLowerCase()));
107
+ break;
108
+ case 'tools_not_called':
109
+ passed = (assertion.values || []).every((v) => !toolNames.includes(String(v).toLowerCase()));
110
+ break;
111
+ case 'tools_count_max':
112
+ passed = toolCalls.length <= assertion.value;
113
+ break;
114
+ case 'tools_count_min':
115
+ passed = toolCalls.length >= assertion.value;
116
+ break;
117
+ case 'tool_output_contains': {
118
+ const sep = String(assertion.value).indexOf(':');
119
+ if (sep > 0) {
120
+ const targetTool = String(assertion.value).slice(0, sep).toLowerCase();
121
+ const expected = String(assertion.value).slice(sep + 1).toLowerCase();
122
+ passed = toolCalls.some((tc) => tc.tool.toLowerCase() === targetTool &&
123
+ String(tc.output || '').toLowerCase().includes(expected));
124
+ }
125
+ break;
126
+ }
127
+ case 'tool_input_contains': {
128
+ const sep = String(assertion.value).indexOf(':');
129
+ if (sep > 0) {
130
+ const targetTool = String(assertion.value).slice(0, sep).toLowerCase();
131
+ const expected = String(assertion.value).slice(sep + 1).toLowerCase();
132
+ passed = toolCalls.some((tc) => tc.tool.toLowerCase() === targetTool &&
133
+ JSON.stringify(tc.input || '').toLowerCase().includes(expected));
134
+ }
135
+ break;
136
+ }
137
+ default:
138
+ passed = false;
139
+ }
140
+ details.push({
141
+ type: assertion.type,
142
+ value: assertion.value ?? assertion.pattern ?? assertion.values?.join(', ') ?? '',
143
+ weight,
144
+ passed,
145
+ });
146
+ }
147
+ const totalWeight = details.reduce((s, d) => s + d.weight, 0);
148
+ const passedWeight = details.filter((d) => d.passed).reduce((s, d) => s + d.weight, 0);
149
+ const passedCount = details.filter((d) => d.passed).length;
150
+ const ratio = totalWeight > 0 ? passedWeight / totalWeight : 0;
151
+ return {
152
+ passed: passedCount,
153
+ total: details.length,
154
+ score: ratioToScore(ratio),
155
+ details,
156
+ };
157
+ }
158
+ export async function runAsyncAssertions(output, assertions, { executor, judgeModel, sample, samplesDir }) {
159
+ const details = [];
160
+ let asyncCostUSD = 0;
161
+ for (const assertion of assertions) {
162
+ const weight = assertion.weight ?? 1;
163
+ let passed = false;
164
+ let message = '';
165
+ if (assertion.type === 'semantic_similarity') {
166
+ const reference = assertion.reference || '';
167
+ const judgePrompt = [
168
+ '请判断以下两段文本的语义相似度。',
169
+ '',
170
+ '## 参考文本',
171
+ reference,
172
+ '',
173
+ '## 待评估文本',
174
+ output,
175
+ '',
176
+ '请返回 JSON(不要包含 markdown 代码块标记):',
177
+ '{"score": <1-5的整数>, "reason": "<简短理由>"}',
178
+ '',
179
+ '评分:1=完全无关, 2=略有关联, 3=部分相似, 4=大致相同, 5=高度一致',
180
+ ].join('\n');
181
+ const result = await executor({
182
+ model: judgeModel,
183
+ system: '你是语义相似度评审员。只返回 JSON,不要其他内容。',
184
+ prompt: judgePrompt,
185
+ });
186
+ asyncCostUSD += result.costUSD || 0;
187
+ if (result.ok) {
188
+ try {
189
+ const jsonMatch = result.output.trim().match(/\{[\s\S]*\}/);
190
+ if (jsonMatch) {
191
+ const parsed = JSON.parse(jsonMatch[0]);
192
+ const score = Number(parsed.score) || 0;
193
+ const threshold = assertion.threshold ?? 3;
194
+ passed = score >= threshold;
195
+ message = parsed.reason || '';
196
+ }
197
+ else {
198
+ process.stderr.write(`[omk] semantic_similarity judge returned non-JSON: ${result.output.slice(0, 100)}\n`);
199
+ }
200
+ }
201
+ catch (parseErr) {
202
+ process.stderr.write(`[omk] semantic_similarity judge parse error: ${getErrorMessage(parseErr)}\n`);
203
+ }
204
+ }
205
+ }
206
+ else if (assertion.type === 'custom') {
207
+ try {
208
+ const fnPath = resolve(samplesDir, assertion.fn);
209
+ const mod = await import(fnPath);
210
+ const fn = mod.default || mod.check;
211
+ if (!fn)
212
+ throw new Error('custom assertion module must export default or check');
213
+ const result = await Promise.race([
214
+ fn(output, { sample, assertion }),
215
+ new Promise((_, reject) => setTimeout(() => reject(new Error(`custom assertion timed out (${CUSTOM_ASSERTION_TIMEOUT_MS / 1000}s)`)), CUSTOM_ASSERTION_TIMEOUT_MS)),
216
+ ]);
217
+ passed = Boolean(result.pass);
218
+ message = result.message || '';
219
+ }
220
+ catch (err) {
221
+ passed = false;
222
+ message = `custom assertion error: ${getErrorMessage(err)}`;
223
+ }
224
+ }
225
+ details.push({
226
+ type: assertion.type,
227
+ value: assertion.reference || assertion.fn || '',
228
+ weight,
229
+ passed,
230
+ ...(message && { message }),
231
+ });
232
+ }
233
+ const totalWeight = details.reduce((s, d) => s + d.weight, 0);
234
+ const passedWeight = details.filter((d) => d.passed).reduce((s, d) => s + d.weight, 0);
235
+ const passedCount = details.filter((d) => d.passed).length;
236
+ const ratio = totalWeight > 0 ? passedWeight / totalWeight : 0;
237
+ return { passed: passedCount, total: details.length, score: ratioToScore(ratio), details, judgeCostUSD: asyncCostUSD };
238
+ }
239
+ //# sourceMappingURL=assertions.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"assertions.js","sourceRoot":"","sources":["../../../src/grading/assertions.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,IAAI,MAAM,KAAK,CAAC;AAGvB,MAAM,GAAG,GAAG,IAAI,CAAC,OAAO,IAAI,IAAI,CAAC;AACjC,MAAM,GAAG,GAAG,IAAI,GAAG,EAAE,CAAC;AACtB,MAAM,2BAA2B,GAAG,MAAM,CAAC;AAE3C,MAAM,CAAC,MAAM,qBAAqB,GAAG,IAAI,GAAG,CAAC,CAAC,qBAAqB,EAAE,QAAQ,CAAC,CAAC,CAAC;AA2BhF,SAAS,eAAe,CAAC,GAAY;IACnC,OAAO,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;AAC1D,CAAC;AAED,MAAM,UAAU,YAAY,CAAC,KAAa;IACxC,OAAO,MAAM,CAAC,CAAC,CAAC,GAAG,KAAK,GAAG,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC;AAC5C,CAAC;AAED,MAAM,UAAU,kBAAkB,CAAC,IAAa,EAAE,MAA+B;IAC/E,IAAI,CAAC,MAAM,IAAI,OAAO,MAAM,KAAK,QAAQ;QAAE,OAAO,IAAI,CAAC;IACvD,IAAI,CAAC;QACH,MAAM,QAAQ,GAAG,GAAG,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;QACrC,OAAO,QAAQ,CAAC,IAAI,CAAY,CAAC;IACnC,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,KAAK,CAAC;IACf,CAAC;AACH,CAAC;AAED,MAAM,UAAU,aAAa,CAAC,MAAc,EAAE,UAAuB,EAAE,UAAoG,EAAE;IAC3K,MAAM,WAAW,GAAG,MAAM,CAAC,WAAW,EAAE,CAAC;IACzC,MAAM,OAAO,GAAsB,EAAE,CAAC;IACtC,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,IAAI,EAAE,CAAC;IAC1C,MAAM,SAAS,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,EAAE,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC,CAAC;IAE/D,KAAK,MAAM,SAAS,IAAI,UAAU,EAAE,CAAC;QACnC,MAAM,MAAM,GAAG,SAAS,CAAC,MAAM,IAAI,CAAC,CAAC;QACrC,IAAI,MAAM,GAAG,KAAK,CAAC;QAEnB,QAAQ,SAAS,CAAC,IAAI,EAAE,CAAC;YACvB,KAAK,UAAU;gBACb,MAAM,GAAG,WAAW,CAAC,QAAQ,CAAC,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC;gBACrE,MAAM;YACR,KAAK,cAAc;gBACjB,MAAM,GAAG,CAAC,WAAW,CAAC,QAAQ,CAAC,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC;gBACtE,MAAM;YACR,KAAK,OAAO,CAAC,CAAC,CAAC;gBACb,MAAM,KAAK,GAAG,SAAS,CAAC,KAAK,IAAI,GAAG,CAAC;gBACrC,MAAM,EAAE,GAAG,IAAI,MAAM,CAAC,SAAS,CAAC,OAAQ,EAAE,KAAK,CAAC,CAAC;gBACjD,MAAM,GAAG,EAAE,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;gBACzB,MAAM;YACR,CAAC;YACD,KAAK,YAAY;gBACf,MAAM,GAAG,MAAM,CAAC,MAAM,IAAK,SAAS,CAAC,KAAgB,CAAC;gBACtD,MAAM;YACR,KAAK,YAAY;gBACf,MAAM,GAAG,MAAM,CAAC,MAAM,IAAK,SAAS,CAAC,KAAgB,CAAC;gBACtD,MAAM;YACR,KAAK,YAAY;gBACf,IAAI,CAAC;oBAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;oBAAC,MAAM,GAAG,IAAI,CAAC;gBAAC,CAAC;gBAAC,MAAM,CAAC;oBAAC,MAAM,GAAG,KAAK,CAAC;gBAAC,CAAC;gBACpE,MAAM;YACR,KAAK,aAAa;gBAChB,IAAI,CAAC;oBACH,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;oBAChC,MAAM,GAAG,kBAAkB,CAAC,IAAI,EAAE,SAAS,CAAC,MAAO,CAAC,CAAC;gBACvD,CAAC;gBAAC,MAAM,CAAC;oBAAC,MAAM,GAAG,KAAK,CAAC;gBAAC,CAAC;gBAC3B,MAAM;YACR,KAAK,aAAa;gBAChB,MAAM,GAAG,WAAW,CAAC,UAAU,CAAC,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC;gBACvE,MAAM;YACR,KAAK,WAAW;gBACd,MAAM,GAAG,WAAW,CAAC,QAAQ,CAAC,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC;gBACrE,MAAM;YACR,KAAK,QAAQ;gBACX,MAAM,GAAG,MAAM,CAAC,IAAI,EAAE,KAAK,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,IAAI,EAAE,CAAC;gBAC1D,MAAM;YACR,KAAK,YAAY;gBACf,MAAM,GAAG,MAAM,CAAC,IAAI,EAAE,KAAK,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,IAAI,EAAE,CAAC;gBAC1D,MAAM;YACR,KAAK,gBAAgB;gBACnB,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,MAAM,IAAK,SAAS,CAAC,KAAgB,CAAC;gBACnF,MAAM;YACR,KAAK,gBAAgB;gBACnB,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,MAAM,IAAK,SAAS,CAAC,KAAgB,CAAC;gBACnF,MAAM;YACR,KAAK,cAAc;gBACjB,MAAM,GAAG,CAAC,SAAS,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,WAAW,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC;gBAC9F,MAAM;YACR,KAAK,cAAc;gBACjB,MAAM,GAAG,CAAC,SAAS,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,WAAW,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC;gBAC7F,MAAM;YACR,KAAK,UAAU;gBACb,MAAM,GAAG,CAAC,OAAO,CAAC,OAAO,IAAI,QAAQ,CAAC,IAAK,SAAS,CAAC,KAAgB,CAAC;gBACtE,MAAM;YACR,KAAK,aAAa;gBAChB,MAAM,GAAG,CAAC,OAAO,CAAC,UAAU,IAAI,QAAQ,CAAC,IAAK,SAAS,CAAC,KAAgB,CAAC;gBACzE,MAAM;YACR,KAAK,WAAW;gBACd,MAAM,GAAG,CAAC,OAAO,CAAC,QAAQ,IAAI,QAAQ,CAAC,IAAK,SAAS,CAAC,KAAgB,CAAC;gBACvE,MAAM;YACR,KAAK,WAAW;gBACd,MAAM,GAAG,CAAC,OAAO,CAAC,QAAQ,IAAI,CAAC,CAAC,IAAK,SAAS,CAAC,KAAgB,CAAC;gBAChE,MAAM;YACR,KAAK,cAAc;gBACjB,MAAM,GAAG,CAAC,SAAS,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,SAAS,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC;gBAC5F,MAAM;YACR,KAAK,kBAAkB;gBACrB,MAAM,GAAG,CAAC,SAAS,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,SAAS,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC;gBAC7F,MAAM;YACR,KAAK,iBAAiB;gBACpB,MAAM,GAAG,SAAS,CAAC,MAAM,IAAK,SAAS,CAAC,KAAgB,CAAC;gBACzD,MAAM;YACR,KAAK,iBAAiB;gBACpB,MAAM,GAAG,SAAS,CAAC,MAAM,IAAK,SAAS,CAAC,KAAgB,CAAC;gBACzD,MAAM;YACR,KAAK,sBAAsB,CAAC,CAAC,CAAC;gBAC5B,MAAM,GAAG,GAAG,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;gBACjD,IAAI,GAAG,GAAG,CAAC,EAAE,CAAC;oBACZ,MAAM,UAAU,GAAG,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,WAAW,EAAE,CAAC;oBACvE,MAAM,QAAQ,GAAG,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC;oBACtE,MAAM,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,EAAE,EAAE,EAAE,CAC7B,EAAE,CAAC,IAAI,CAAC,WAAW,EAAE,KAAK,UAAU;wBACpC,MAAM,CAAC,EAAE,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,QAAQ,CAAC,CACzD,CAAC;gBACJ,CAAC;gBACD,MAAM;YACR,CAAC;YACD,KAAK,qBAAqB,CAAC,CAAC,CAAC;gBAC3B,MAAM,GAAG,GAAG,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;gBACjD,IAAI,GAAG,GAAG,CAAC,EAAE,CAAC;oBACZ,MAAM,UAAU,GAAG,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,WAAW,EAAE,CAAC;oBACvE,MAAM,QAAQ,GAAG,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC;oBACtE,MAAM,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,EAAE,EAAE,EAAE,CAC7B,EAAE,CAAC,IAAI,CAAC,WAAW,EAAE,KAAK,UAAU;wBACpC,IAAI,CAAC,SAAS,CAAC,EAAE,CAAC,KAAK,IAAI,EAAE,CAAC,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAChE,CAAC;gBACJ,CAAC;gBACD,MAAM;YACR,CAAC;YACD;gBACE,MAAM,GAAG,KAAK,CAAC;QACnB,CAAC;QAED,OAAO,CAAC,IAAI,CAAC;YACX,IAAI,EAAE,SAAS,CAAC,IAAI;YACpB,KAAK,EAAE,SAAS,CAAC,KAAK,IAAI,SAAS,CAAC,OAAO,IAAI,SAAS,CAAC,MAAM,EAAE,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE;YACjF,MAAM;YACN,MAAM;SACP,CAAC,CAAC;IACL,CAAC;IAED,MAAM,WAAW,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;IAC9D,MAAM,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;IACvF,MAAM,WAAW,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC;IAC3D,MAAM,KAAK,GAAG,WAAW,GAAG,CAAC,CAAC,CAAC,CAAC,YAAY,GAAG,WAAW,CAAC,CAAC,CAAC,CAAC,CAAC;IAE/D,OAAO;QACL,MAAM,EAAE,WAAW;QACnB,KAAK,EAAE,OAAO,CAAC,MAAM;QACrB,KAAK,EAAE,YAAY,CAAC,KAAK,CAAC;QAC1B,OAAO;KACR,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,kBAAkB,CAAC,MAAc,EAAE,UAAuB,EAAE,EAAE,QAAQ,EAAE,UAAU,EAAE,MAAM,EAAE,UAAU,EAAyB;IACnJ,MAAM,OAAO,GAAsB,EAAE,CAAC;IACtC,IAAI,YAAY,GAAG,CAAC,CAAC;IAErB,KAAK,MAAM,SAAS,IAAI,UAAU,EAAE,CAAC;QACnC,MAAM,MAAM,GAAG,SAAS,CAAC,MAAM,IAAI,CAAC,CAAC;QACrC,IAAI,MAAM,GAAG,KAAK,CAAC;QACnB,IAAI,OAAO,GAAG,EAAE,CAAC;QAEjB,IAAI,SAAS,CAAC,IAAI,KAAK,qBAAqB,EAAE,CAAC;YAC7C,MAAM,SAAS,GAAG,SAAS,CAAC,SAAS,IAAI,EAAE,CAAC;YAC5C,MAAM,WAAW,GAAG;gBAClB,kBAAkB;gBAClB,EAAE;gBACF,SAAS;gBACT,SAAS;gBACT,EAAE;gBACF,UAAU;gBACV,MAAM;gBACN,EAAE;gBACF,gCAAgC;gBAChC,yCAAyC;gBACzC,EAAE;gBACF,2CAA2C;aAC5C,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YAEb,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC;gBAC5B,KAAK,EAAE,UAAU;gBACjB,MAAM,EAAE,6BAA6B;gBACrC,MAAM,EAAE,WAAW;aACpB,CAAC,CAAC;YAEH,YAAY,IAAI,MAAM,CAAC,OAAO,IAAI,CAAC,CAAC;YACpC,IAAI,MAAM,CAAC,EAAE,EAAE,CAAC;gBACd,IAAI,CAAC;oBACH,MAAM,SAAS,GAAG,MAAM,CAAC,MAAO,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC;oBAC7D,IAAI,SAAS,EAAE,CAAC;wBACd,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAAkB,CAAC;wBACzD,MAAM,KAAK,GAAG,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;wBACxC,MAAM,SAAS,GAAG,SAAS,CAAC,SAAS,IAAI,CAAC,CAAC;wBAC3C,MAAM,GAAG,KAAK,IAAI,SAAS,CAAC;wBAC5B,OAAO,GAAG,MAAM,CAAC,MAAM,IAAI,EAAE,CAAC;oBAChC,CAAC;yBAAM,CAAC;wBACN,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,sDAAsD,MAAM,CAAC,MAAO,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,IAAI,CAAC,CAAC;oBAC/G,CAAC;gBACH,CAAC;gBAAC,OAAO,QAAiB,EAAE,CAAC;oBAC3B,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,gDAAgD,eAAe,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;gBACtG,CAAC;YACH,CAAC;QACH,CAAC;aAAM,IAAI,SAAS,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;YACvC,IAAI,CAAC;gBACH,MAAM,MAAM,GAAG,OAAO,CAAC,UAAU,EAAE,SAAS,CAAC,EAAG,CAAC,CAAC;gBAClD,MAAM,GAAG,GAAG,MAAM,MAAM,CAAC,MAAM,CAA0B,CAAC;gBAC1D,MAAM,EAAE,GAAG,GAAG,CAAC,OAAO,IAAI,GAAG,CAAC,KAAK,CAAC;gBACpC,IAAI,CAAC,EAAE;oBAAE,MAAM,IAAI,KAAK,CAAC,sDAAsD,CAAC,CAAC;gBACjF,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,IAAI,CAAwB;oBACvD,EAAE,CAAC,MAAM,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC;oBACjC,IAAI,OAAO,CAAwB,CAAC,CAAC,EAAE,MAAM,EAAE,EAAE,CAAC,UAAU,CAAC,GAAG,EAAE,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,+BAA+B,2BAA2B,GAAG,IAAI,IAAI,CAAC,CAAC,EAAE,2BAA2B,CAAC,CAAC;iBAC3L,CAAC,CAAC;gBACH,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;gBAC9B,OAAO,GAAG,MAAM,CAAC,OAAO,IAAI,EAAE,CAAC;YACjC,CAAC;YAAC,OAAO,GAAY,EAAE,CAAC;gBACtB,MAAM,GAAG,KAAK,CAAC;gBACf,OAAO,GAAG,2BAA2B,eAAe,CAAC,GAAG,CAAC,EAAE,CAAC;YAC9D,CAAC;QACH,CAAC;QAED,OAAO,CAAC,IAAI,CAAC;YACX,IAAI,EAAE,SAAS,CAAC,IAAI;YACpB,KAAK,EAAE,SAAS,CAAC,SAAS,IAAI,SAAS,CAAC,EAAE,IAAI,EAAE;YAChD,MAAM;YACN,MAAM;YACN,GAAG,CAAC,OAAO,IAAI,EAAE,OAAO,EAAE,CAAC;SAC5B,CAAC,CAAC;IACL,CAAC;IAED,MAAM,WAAW,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;IAC9D,MAAM,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;IACvF,MAAM,WAAW,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC;IAC3D,MAAM,KAAK,GAAG,WAAW,GAAG,CAAC,CAAC,CAAC,CAAC,YAAY,GAAG,WAAW,CAAC,CAAC,CAAC,CAAC,CAAC;IAE/D,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE,KAAK,EAAE,OAAO,CAAC,MAAM,EAAE,KAAK,EAAE,YAAY,CAAC,KAAK,CAAC,EAAE,OAAO,EAAE,YAAY,EAAE,YAAY,EAAE,CAAC;AACzH,CAAC"}
@@ -0,0 +1,26 @@
1
+ /**
2
+ * Mixed grading: deterministic assertions + LLM judge + multi-dimensional scoring.
3
+ */
4
+ import type { GradeResult, ExecutorFn, Sample, ToolCallInfo, TurnInfo } from '../types.js';
5
+ interface GradeOptions {
6
+ output: string;
7
+ sample: Sample;
8
+ executor: ExecutorFn;
9
+ judgeModel: string;
10
+ allowLlmJudge?: boolean;
11
+ execMetrics?: {
12
+ costUSD?: number;
13
+ durationMs?: number;
14
+ numTurns?: number;
15
+ toolCalls?: ToolCallInfo[];
16
+ turns?: TurnInfo[];
17
+ };
18
+ samplesDir?: string;
19
+ }
20
+ /**
21
+ * Grade a model output against a sample's criteria.
22
+ */
23
+ export declare function grade({ output, sample, executor, judgeModel, allowLlmJudge, execMetrics, samplesDir }: GradeOptions): Promise<GradeResult>;
24
+ export { runAssertions, validateJsonSchema } from './assertions.js';
25
+ export { buildTraceSummary } from './judge.js';
26
+ //# sourceMappingURL=index.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/grading/index.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,EAAoB,WAAW,EAAkC,UAAU,EAAE,MAAM,EAAE,YAAY,EAAE,QAAQ,EAAE,MAAM,aAAa,CAAC;AAK7I,UAAU,YAAY;IACpB,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,UAAU,CAAC;IACrB,UAAU,EAAE,MAAM,CAAC;IACnB,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,WAAW,CAAC,EAAE;QAAE,OAAO,CAAC,EAAE,MAAM,CAAC;QAAC,UAAU,CAAC,EAAE,MAAM,CAAC;QAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;QAAC,SAAS,CAAC,EAAE,YAAY,EAAE,CAAC;QAAC,KAAK,CAAC,EAAE,QAAQ,EAAE,CAAA;KAAE,CAAC;IAC3H,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED;;GAEG;AACH,wBAAsB,KAAK,CAAC,EAAE,MAAM,EAAE,MAAM,EAAE,QAAQ,EAAE,UAAU,EAAE,aAAoB,EAAE,WAAgB,EAAE,UAAgB,EAAE,EAAE,YAAY,GAAG,OAAO,CAAC,WAAW,CAAC,CA+FlK;AACD,OAAO,EAAE,aAAa,EAAE,kBAAkB,EAAE,MAAM,iBAAiB,CAAC;AACpE,OAAO,EAAE,iBAAiB,EAAE,MAAM,YAAY,CAAC"}
@@ -0,0 +1,98 @@
1
+ /**
2
+ * Mixed grading: deterministic assertions + LLM judge + multi-dimensional scoring.
3
+ */
4
+ import { ASYNC_ASSERTION_TYPES, ratioToScore, runAssertions, runAsyncAssertions } from './assertions.js';
5
+ import { buildTraceSummary, llmJudge } from './judge.js';
6
+ import { computeLayeredScores } from './layered-scores.js';
7
+ /**
8
+ * Grade a model output against a sample's criteria.
9
+ */
10
+ export async function grade({ output, sample, executor, judgeModel, allowLlmJudge = true, execMetrics = {}, samplesDir = '.' }) {
11
+ const results = { compositeScore: 0 };
12
+ // 1. Deterministic assertions (pure, no LLM)
13
+ const allAssertions = sample.assertions || [];
14
+ const syncAssertions = allAssertions.filter((a) => !ASYNC_ASSERTION_TYPES.has(a.type));
15
+ const asyncAssertions = allAssertions.filter((assertion) => {
16
+ if (!ASYNC_ASSERTION_TYPES.has(assertion.type))
17
+ return false;
18
+ if (allowLlmJudge)
19
+ return true;
20
+ return assertion.type === 'custom';
21
+ });
22
+ if (syncAssertions.length > 0) {
23
+ results.assertions = runAssertions(output, syncAssertions, { ...execMetrics });
24
+ }
25
+ // 1b. Async assertions (custom JS, semantic_similarity)
26
+ if (asyncAssertions.length > 0 && executor != null) {
27
+ const asyncResults = await runAsyncAssertions(output, asyncAssertions, {
28
+ executor, judgeModel, sample, samplesDir,
29
+ });
30
+ if (results.assertions) {
31
+ // Merge async results into sync results
32
+ results.assertions.details.push(...asyncResults.details);
33
+ results.assertions.total += asyncResults.total;
34
+ results.assertions.passed += asyncResults.passed;
35
+ // Recompute score from merged details
36
+ const allDetails = results.assertions.details;
37
+ const totalWeight = allDetails.reduce((s, d) => s + d.weight, 0);
38
+ const passedWeight = allDetails.filter((d) => d.passed).reduce((s, d) => s + d.weight, 0);
39
+ const ratio = totalWeight > 0 ? passedWeight / totalWeight : 0;
40
+ results.assertions.score = ratioToScore(ratio);
41
+ }
42
+ else {
43
+ results.assertions = asyncResults;
44
+ }
45
+ // Accumulate async assertion judge cost
46
+ if (asyncResults.judgeCostUSD && asyncResults.judgeCostUSD > 0) {
47
+ results.judgeCostUSD = (results.judgeCostUSD || 0) + asyncResults.judgeCostUSD;
48
+ }
49
+ }
50
+ // 2. LLM scoring
51
+ // Build execution trace summary for agent-aware judging
52
+ const traceSummary = buildTraceSummary(execMetrics.turns, execMetrics.toolCalls);
53
+ if (allowLlmJudge && sample.dimensions && Object.keys(sample.dimensions).length > 0) {
54
+ // Multi-dimensional scoring
55
+ results.dimensions = {};
56
+ for (const [dim, rubric] of Object.entries(sample.dimensions)) {
57
+ results.dimensions[dim] = await llmJudge({
58
+ output,
59
+ rubric,
60
+ prompt: sample.prompt,
61
+ executor,
62
+ model: judgeModel,
63
+ traceSummary,
64
+ });
65
+ }
66
+ const dimValues = Object.values(results.dimensions);
67
+ const dimScores = dimValues.map((d) => d.score).filter((s) => s > 0);
68
+ if (dimScores.length > 0) {
69
+ results.llmScore = Number((dimScores.reduce((a, b) => a + b, 0) / dimScores.length).toFixed(2));
70
+ }
71
+ // Accumulate judge cost from all dimensions (add to any existing async assertion cost)
72
+ const dimCost = dimValues.reduce((s, d) => s + (d.judgeCostUSD || 0), 0);
73
+ if (dimCost > 0)
74
+ results.judgeCostUSD = (results.judgeCostUSD || 0) + dimCost;
75
+ }
76
+ else if (allowLlmJudge && sample.rubric) {
77
+ // Single rubric scoring
78
+ const judge = await llmJudge({
79
+ output,
80
+ rubric: sample.rubric,
81
+ prompt: sample.prompt,
82
+ executor,
83
+ model: judgeModel,
84
+ traceSummary,
85
+ });
86
+ results.llmScore = judge.score;
87
+ results.llmReason = judge.reason;
88
+ results.judgeCostUSD = (results.judgeCostUSD || 0) + (judge.judgeCostUSD || 0);
89
+ }
90
+ // 3. Layered scores + composite
91
+ const { compositeScore, layeredScores } = computeLayeredScores(results);
92
+ results.compositeScore = compositeScore;
93
+ results.layeredScores = layeredScores;
94
+ return results;
95
+ }
96
+ export { runAssertions, validateJsonSchema } from './assertions.js';
97
+ export { buildTraceSummary } from './judge.js';
98
+ //# sourceMappingURL=index.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/grading/index.ts"],"names":[],"mappings":"AAAA;;GAEG;AAGH,OAAO,EAAE,qBAAqB,EAAE,YAAY,EAAE,aAAa,EAAE,kBAAkB,EAAE,MAAM,iBAAiB,CAAC;AACzG,OAAO,EAAE,iBAAiB,EAAE,QAAQ,EAAE,MAAM,YAAY,CAAC;AACzD,OAAO,EAAE,oBAAoB,EAAE,MAAM,qBAAqB,CAAC;AAY3D;;GAEG;AACH,MAAM,CAAC,KAAK,UAAU,KAAK,CAAC,EAAE,MAAM,EAAE,MAAM,EAAE,QAAQ,EAAE,UAAU,EAAE,aAAa,GAAG,IAAI,EAAE,WAAW,GAAG,EAAE,EAAE,UAAU,GAAG,GAAG,EAAgB;IAC1I,MAAM,OAAO,GAQT,EAAE,cAAc,EAAE,CAAC,EAAE,CAAC;IAE1B,6CAA6C;IAC7C,MAAM,aAAa,GAAG,MAAM,CAAC,UAAU,IAAI,EAAE,CAAC;IAC9C,MAAM,cAAc,GAAG,aAAa,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,qBAAqB,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC;IACvF,MAAM,eAAe,GAAG,aAAa,CAAC,MAAM,CAAC,CAAC,SAAS,EAAE,EAAE;QACzD,IAAI,CAAC,qBAAqB,CAAC,GAAG,CAAC,SAAS,CAAC,IAAI,CAAC;YAAE,OAAO,KAAK,CAAC;QAC7D,IAAI,aAAa;YAAE,OAAO,IAAI,CAAC;QAC/B,OAAO,SAAS,CAAC,IAAI,KAAK,QAAQ,CAAC;IACrC,CAAC,CAAC,CAAC;IAEH,IAAI,cAAc,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC9B,OAAO,CAAC,UAAU,GAAG,aAAa,CAAC,MAAM,EAAE,cAAc,EAAE,EAAE,GAAG,WAAW,EAAE,CAAC,CAAC;IACjF,CAAC;IAED,wDAAwD;IACxD,IAAI,eAAe,CAAC,MAAM,GAAG,CAAC,IAAI,QAAQ,IAAI,IAAI,EAAE,CAAC;QACnD,MAAM,YAAY,GAAG,MAAM,kBAAkB,CAAC,MAAM,EAAE,eAAe,EAAE;YACrE,QAAQ,EAAE,UAAU,EAAE,MAAM,EAAE,UAAU;SACzC,CAAC,CAAC;QACH,IAAI,OAAO,CAAC,UAAU,EAAE,CAAC;YACvB,wCAAwC;YACxC,OAAO,CAAC,UAAU,CAAC,OAAO,CAAC,IAAI,CAAC,GAAG,YAAY,CAAC,OAAO,CAAC,CAAC;YACzD,OAAO,CAAC,UAAU,CAAC,KAAK,IAAI,YAAY,CAAC,KAAK,CAAC;YAC/C,OAAO,CAAC,UAAU,CAAC,MAAM,IAAI,YAAY,CAAC,MAAM,CAAC;YACjD,sCAAsC;YACtC,MAAM,UAAU,GAAG,OAAO,CAAC,UAAU,CAAC,OAAO,CAAC;YAC9C,MAAM,WAAW,GAAG,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;YACjE,MAAM,YAAY,GAAG,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;YAC1F,MAAM,KAAK,GAAG,WAAW,GAAG,CAAC,CAAC,CAAC,CAAC,YAAY,GAAG,WAAW,CAAC,CAAC,CAAC,CAAC,CAAC;YAC/D,OAAO,CAAC,UAAU,CAAC,KAAK,GAAG,YAAY,CAAC,KAAK,CAAC,CAAC;QACjD,CAAC;aAAM,CAAC;YACN,OAAO,CAAC,UAAU,GAAG,YAAY,CAAC;QACpC,CAAC;QACD,wCAAwC;QACxC,IAAI,YAAY,CAAC,YAAY,IAAI,YAAY,CAAC,YAAY,GAAG,CAAC,EAAE,CAAC;YAC/D,OAAO,CAAC,YAAY,GAAG,CAAC,OAAO,CAAC,YAAY,IAAI,CAAC,CAAC,GAAG,YAAY,CAAC,YAAY,CAAC;QACjF,CAAC;IACH,CAAC;IAED,iBAAiB;IACjB,wDAAwD;IACxD,MAAM,YAAY,GAAG,iBAAiB,CAAC,WAAW,CAAC,KAAK,EAAE,WAAW,CAAC,SAAS,CAAC,CAAC;IAEjF,IAAI,aAAa,IAAI,MAAM,CAAC,UAAU,IAAI,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACpF,4BAA4B;QAC5B,OAAO,CAAC,UAAU,GAAG,EAAE,CAAC;QACxB,KAAK,MAAM,CAAC,GAAG,EAAE,MAAM,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,UAAU,CAAC,EAAE,CAAC;YAC9D,OAAO,CAAC,UAAU,CAAC,GAAG,CAAC,GAAG,MAAM,QAAQ,CAAC;gBACvC,MAAM;gBACN,MAAM;gBACN,MAAM,EAAE,MAAM,CAAC,MAAM;gBACrB,QAAQ;gBACR,KAAK,EAAE,UAAU;gBACjB,YAAY;aACb,CAAC,CAAC;QACL,CAAC;QACD,MAAM,SAAS,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC;QACpD,MAAM,SAAS,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;QACrE,IAAI,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACzB,OAAO,CAAC,QAAQ,GAAG,MAAM,CAAC,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC;QAClG,CAAC;QACD,uFAAuF;QACvF,MAAM,OAAO,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,YAAY,IAAI,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;QACzE,IAAI,OAAO,GAAG,CAAC;YAAE,OAAO,CAAC,YAAY,GAAG,CAAC,OAAO,CAAC,YAAY,IAAI,CAAC,CAAC,GAAG,OAAO,CAAC;IAChF,CAAC;SAAM,IAAI,aAAa,IAAI,MAAM,CAAC,MAAM,EAAE,CAAC;QAC1C,wBAAwB;QACxB,MAAM,KAAK,GAAG,MAAM,QAAQ,CAAC;YAC3B,MAAM;YACN,MAAM,EAAE,MAAM,CAAC,MAAM;YACrB,MAAM,EAAE,MAAM,CAAC,MAAM;YACrB,QAAQ;YACR,KAAK,EAAE,UAAU;YACjB,YAAY;SACb,CAAC,CAAC;QACH,OAAO,CAAC,QAAQ,GAAG,KAAK,CAAC,KAAK,CAAC;QAC/B,OAAO,CAAC,SAAS,GAAG,KAAK,CAAC,MAAM,CAAC;QACjC,OAAO,CAAC,YAAY,GAAG,CAAC,OAAO,CAAC,YAAY,IAAI,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,YAAY,IAAI,CAAC,CAAC,CAAC;IACjF,CAAC;IAED,gCAAgC;IAChC,MAAM,EAAE,cAAc,EAAE,aAAa,EAAE,GAAG,oBAAoB,CAAC,OAAO,CAAC,CAAC;IACxE,OAAO,CAAC,cAAc,GAAG,cAAc,CAAC;IACxC,OAAO,CAAC,aAAa,GAAG,aAAa,CAAC;IAEtC,OAAO,OAAO,CAAC;AACjB,CAAC;AACD,OAAO,EAAE,aAAa,EAAE,kBAAkB,EAAE,MAAM,iBAAiB,CAAC;AACpE,OAAO,EAAE,iBAAiB,EAAE,MAAM,YAAY,CAAC"}
@@ -0,0 +1,13 @@
1
+ import type { DimensionResult, ExecutorFn, ToolCallInfo, TurnInfo } from '../types.js';
2
+ interface LlmJudgeOptions {
3
+ output: string;
4
+ rubric: string;
5
+ prompt: string;
6
+ executor: ExecutorFn;
7
+ model: string;
8
+ traceSummary?: string | null;
9
+ }
10
+ export declare function buildTraceSummary(turns?: TurnInfo[], toolCalls?: ToolCallInfo[]): string | null;
11
+ export declare function llmJudge({ output, rubric, prompt, executor, model, traceSummary }: LlmJudgeOptions): Promise<DimensionResult>;
12
+ export {};
13
+ //# sourceMappingURL=judge.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"judge.d.ts","sourceRoot":"","sources":["../../../src/grading/judge.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,eAAe,EAAE,UAAU,EAAE,YAAY,EAAE,QAAQ,EAAE,MAAM,aAAa,CAAC;AAOvF,UAAU,eAAe;IACvB,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,UAAU,CAAC;IACrB,KAAK,EAAE,MAAM,CAAC;IACd,YAAY,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;CAC9B;AAMD,wBAAgB,iBAAiB,CAAC,KAAK,CAAC,EAAE,QAAQ,EAAE,EAAE,SAAS,CAAC,EAAE,YAAY,EAAE,GAAG,MAAM,GAAG,IAAI,CA6C/F;AAED,wBAAsB,QAAQ,CAAC,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,QAAQ,EAAE,KAAK,EAAE,YAAY,EAAE,EAAE,eAAe,GAAG,OAAO,CAAC,eAAe,CAAC,CAiDnI"}
@@ -0,0 +1,98 @@
1
+ function getErrorMessage(err) {
2
+ return err instanceof Error ? err.message : String(err);
3
+ }
4
+ export function buildTraceSummary(turns, toolCalls) {
5
+ if ((!turns || turns.length === 0) && (!toolCalls || toolCalls.length === 0))
6
+ return null;
7
+ const lines = [];
8
+ if (toolCalls && toolCalls.length > 0) {
9
+ lines.push(`共调用 ${toolCalls.length} 个工具:`);
10
+ const successCount = toolCalls.filter((tc) => tc.success).length;
11
+ const failureCount = toolCalls.length - successCount;
12
+ lines.push(` 成功 ${successCount}/${toolCalls.length}`);
13
+ if (failureCount > 0)
14
+ lines.push(` 失败 ${failureCount}/${toolCalls.length}`);
15
+ const dist = {};
16
+ for (const tc of toolCalls) {
17
+ dist[tc.tool] = (dist[tc.tool] || 0) + 1;
18
+ }
19
+ lines.push(` 工具分布:${Object.entries(dist).map(([k, v]) => `${k}(${v})`).join(', ')}`);
20
+ const failedTools = toolCalls.filter((tc) => !tc.success).map((tc) => tc.tool);
21
+ if (failedTools.length > 0) {
22
+ lines.push(` 失败工具:${[...new Set(failedTools)].join(', ')}`);
23
+ }
24
+ }
25
+ if (turns && turns.length > 0) {
26
+ lines.push('');
27
+ lines.push('执行轨迹摘要:');
28
+ const assistantTurns = turns.filter((turn) => turn.role === 'assistant').length;
29
+ const toolTurns = turns.filter((turn) => turn.role === 'tool').length;
30
+ lines.push(` 共 ${turns.length} 步(assistant ${assistantTurns} / tool ${toolTurns})`);
31
+ const maxTurns = Math.min(turns.length, 10);
32
+ for (let i = 0; i < maxTurns; i++) {
33
+ const t = turns[i];
34
+ const preview = t.content.slice(0, 100) + (t.content.length > 100 ? '...' : '');
35
+ if (t.role === 'assistant' && t.toolCalls?.length) {
36
+ lines.push(` [${i + 1}] assistant: 调用 ${t.toolCalls.map((tc) => tc.tool).join(', ')}`);
37
+ }
38
+ else if (t.role === 'tool') {
39
+ lines.push(` [${i + 1}] tool: ${preview}`);
40
+ }
41
+ else {
42
+ lines.push(` [${i + 1}] ${t.role}: ${preview}`);
43
+ }
44
+ }
45
+ if (turns.length > maxTurns)
46
+ lines.push(` ... 还有 ${turns.length - maxTurns} 步`);
47
+ }
48
+ return lines.join('\n');
49
+ }
50
+ export async function llmJudge({ output, rubric, prompt, executor, model, traceSummary }) {
51
+ const traceSection = traceSummary
52
+ ? ['', '## Agent 执行过程', traceSummary, '', '请同时考虑执行过程的合理性(工具选择、步骤效率、错误恢复)。']
53
+ : [];
54
+ const judgePrompt = [
55
+ '请对以下 AI 输出进行质量评分。',
56
+ '',
57
+ '## 原始任务',
58
+ prompt,
59
+ '',
60
+ '## 评分标准',
61
+ rubric,
62
+ '',
63
+ '## AI 输出',
64
+ output,
65
+ ...traceSection,
66
+ '',
67
+ '请返回 JSON(不要包含 markdown 代码块标记):',
68
+ '{"score": <1-5的整数>, "reason": "<简短理由>"}',
69
+ '',
70
+ '评分标准:1=完全不达标, 2=部分涉及, 3=基本达标, 4=较好, 5=优秀',
71
+ ].join('\n');
72
+ const result = await executor({
73
+ model,
74
+ system: '你是一个严格的 AI 输出质量评审员。只返回 JSON,不要其他内容。',
75
+ prompt: judgePrompt,
76
+ });
77
+ if (!result.ok)
78
+ return { score: 0, reason: `judge error: ${result.error}`, judgeCostUSD: result.costUSD };
79
+ try {
80
+ const text = result.output.trim();
81
+ const jsonMatch = text.match(/\{[\s\S]*\}/);
82
+ if (!jsonMatch) {
83
+ process.stderr.write(`[omk] LLM judge returned non-JSON: ${text.slice(0, 100)}\n`);
84
+ return { score: 0, reason: 'judge returned non-JSON', judgeCostUSD: result.costUSD };
85
+ }
86
+ const parsed = JSON.parse(jsonMatch[0]);
87
+ return {
88
+ score: Number(parsed.score) || 0,
89
+ reason: String(parsed.reason || ''),
90
+ judgeCostUSD: result.costUSD,
91
+ };
92
+ }
93
+ catch (parseErr) {
94
+ process.stderr.write(`[omk] LLM judge parse error: ${getErrorMessage(parseErr)}\n`);
95
+ return { score: 0, reason: 'failed to parse judge response', judgeCostUSD: result.costUSD };
96
+ }
97
+ }
98
+ //# sourceMappingURL=judge.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"judge.js","sourceRoot":"","sources":["../../../src/grading/judge.ts"],"names":[],"mappings":"AAgBA,SAAS,eAAe,CAAC,GAAY;IACnC,OAAO,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;AAC1D,CAAC;AAED,MAAM,UAAU,iBAAiB,CAAC,KAAkB,EAAE,SAA0B;IAC9E,IAAI,CAAC,CAAC,KAAK,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC,SAAS,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC,CAAC;QAAE,OAAO,IAAI,CAAC;IAE1F,MAAM,KAAK,GAAa,EAAE,CAAC;IAE3B,IAAI,SAAS,IAAI,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACtC,KAAK,CAAC,IAAI,CAAC,OAAO,SAAS,CAAC,MAAM,OAAO,CAAC,CAAC;QAC3C,MAAM,YAAY,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,EAAE,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC;QACjE,MAAM,YAAY,GAAG,SAAS,CAAC,MAAM,GAAG,YAAY,CAAC;QACrD,KAAK,CAAC,IAAI,CAAC,QAAQ,YAAY,IAAI,SAAS,CAAC,MAAM,EAAE,CAAC,CAAC;QACvD,IAAI,YAAY,GAAG,CAAC;YAAE,KAAK,CAAC,IAAI,CAAC,QAAQ,YAAY,IAAI,SAAS,CAAC,MAAM,EAAE,CAAC,CAAC;QAE7E,MAAM,IAAI,GAA2B,EAAE,CAAC;QACxC,KAAK,MAAM,EAAE,IAAI,SAAS,EAAE,CAAC;YAC3B,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC;QAC3C,CAAC;QACD,KAAK,CAAC,IAAI,CAAC,UAAU,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACtF,MAAM,WAAW,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC,EAAE,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,EAAE,CAAC,IAAI,CAAC,CAAC;QAC/E,IAAI,WAAW,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC3B,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC,GAAG,IAAI,GAAG,CAAC,WAAW,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QAC/D,CAAC;IACH,CAAC;IAED,IAAI,KAAK,IAAI,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC9B,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACf,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QACtB,MAAM,cAAc,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,KAAK,WAAW,CAAC,CAAC,MAAM,CAAC;QAChF,MAAM,SAAS,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,KAAK,MAAM,CAAC,CAAC,MAAM,CAAC;QACtE,KAAK,CAAC,IAAI,CAAC,OAAO,KAAK,CAAC,MAAM,gBAAgB,cAAc,WAAW,SAAS,GAAG,CAAC,CAAC;QACrF,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;QAC5C,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,QAAQ,EAAE,CAAC,EAAE,EAAE,CAAC;YAClC,MAAM,CAAC,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC;YACnB,MAAM,OAAO,GAAG,CAAC,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,OAAO,CAAC,MAAM,GAAG,GAAG,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;YAChF,IAAI,CAAC,CAAC,IAAI,KAAK,WAAW,IAAI,CAAC,CAAC,SAAS,EAAE,MAAM,EAAE,CAAC;gBAClD,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,mBAAmB,CAAC,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,EAAE,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;YAC1F,CAAC;iBAAM,IAAI,CAAC,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;gBAC7B,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,WAAW,OAAO,EAAE,CAAC,CAAC;YAC9C,CAAC;iBAAM,CAAC;gBACN,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,IAAI,KAAK,OAAO,EAAE,CAAC,CAAC;YACnD,CAAC;QACH,CAAC;QACD,IAAI,KAAK,CAAC,MAAM,GAAG,QAAQ;YAAE,KAAK,CAAC,IAAI,CAAC,YAAY,KAAK,CAAC,MAAM,GAAG,QAAQ,IAAI,CAAC,CAAC;IACnF,CAAC;IAED,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,QAAQ,CAAC,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,QAAQ,EAAE,KAAK,EAAE,YAAY,EAAmB;IACvG,MAAM,YAAY,GAAG,YAAY;QAC/B,CAAC,CAAC,CAAC,EAAE,EAAE,eAAe,EAAE,YAAY,EAAE,EAAE,EAAE,gCAAgC,CAAC;QAC3E,CAAC,CAAC,EAAE,CAAC;IAEP,MAAM,WAAW,GAAG;QAClB,mBAAmB;QACnB,EAAE;QACF,SAAS;QACT,MAAM;QACN,EAAE;QACF,SAAS;QACT,MAAM;QACN,EAAE;QACF,UAAU;QACV,MAAM;QACN,GAAG,YAAY;QACf,EAAE;QACF,gCAAgC;QAChC,yCAAyC;QACzC,EAAE;QACF,0CAA0C;KAC3C,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAEb,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC;QAC5B,KAAK;QACL,MAAM,EAAE,qCAAqC;QAC7C,MAAM,EAAE,WAAW;KACpB,CAAC,CAAC;IAEH,IAAI,CAAC,MAAM,CAAC,EAAE;QAAE,OAAO,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,gBAAgB,MAAM,CAAC,KAAK,EAAE,EAAE,YAAY,EAAE,MAAM,CAAC,OAAO,EAAE,CAAC;IAE1G,IAAI,CAAC;QACH,MAAM,IAAI,GAAG,MAAM,CAAC,MAAO,CAAC,IAAI,EAAE,CAAC;QACnC,MAAM,SAAS,GAAG,IAAI,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC;QAC5C,IAAI,CAAC,SAAS,EAAE,CAAC;YACf,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,sCAAsC,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,IAAI,CAAC,CAAC;YACnF,OAAO,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,yBAAyB,EAAE,YAAY,EAAE,MAAM,CAAC,OAAO,EAAE,CAAC;QACvF,CAAC;QACD,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAAkB,CAAC;QACzD,OAAO;YACL,KAAK,EAAE,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC;YAChC,MAAM,EAAE,MAAM,CAAC,MAAM,CAAC,MAAM,IAAI,EAAE,CAAC;YACnC,YAAY,EAAE,MAAM,CAAC,OAAO;SAC7B,CAAC;IACJ,CAAC;IAAC,OAAO,QAAiB,EAAE,CAAC;QAC3B,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,gCAAgC,eAAe,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;QACpF,OAAO,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,gCAAgC,EAAE,YAAY,EAAE,MAAM,CAAC,OAAO,EAAE,CAAC;IAC9F,CAAC;AACH,CAAC"}
@@ -0,0 +1,13 @@
1
+ import type { AssertionDetail, LayeredScores } from '../types.js';
2
+ interface CompositeInput {
3
+ assertions?: {
4
+ details?: AssertionDetail[];
5
+ };
6
+ llmScore?: number;
7
+ }
8
+ export declare function computeLayeredScores(results: CompositeInput): {
9
+ compositeScore: number;
10
+ layeredScores: LayeredScores;
11
+ };
12
+ export {};
13
+ //# sourceMappingURL=layered-scores.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"layered-scores.d.ts","sourceRoot":"","sources":["../../../src/grading/layered-scores.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,eAAe,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AA+ClE,UAAU,cAAc;IACtB,UAAU,CAAC,EAAE;QAAE,OAAO,CAAC,EAAE,eAAe,EAAE,CAAA;KAAE,CAAC;IAC7C,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB;AAED,wBAAgB,oBAAoB,CAAC,OAAO,EAAE,cAAc,GAAG;IAAE,cAAc,EAAE,MAAM,CAAC;IAAC,aAAa,EAAE,aAAa,CAAA;CAAE,CAqBtH"}