@wix/pathgrade 1.0.2 → 1.0.4

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.
@@ -72,6 +72,9 @@ function recordResult(result, agent, attribution) {
72
72
  meta: {
73
73
  score: result.score,
74
74
  scorers: result.scorers,
75
+ ...(result.evaluationDefinitionKey
76
+ ? { evaluationDefinitionKey: result.evaluationDefinitionKey }
77
+ : {}),
75
78
  trial: result.trial,
76
79
  resultKind: result.resultKind ?? 'evaluated',
77
80
  ...(result.scoringDurationMs !== undefined
@@ -39,7 +39,7 @@ export function emitEvalResult(event) {
39
39
  }
40
40
  }
41
41
  export function resetUserResultObservers() {
42
- for (const subscription of [...observers]) {
42
+ for (const subscription of observers) {
43
43
  if (subscription.owner === 'user') {
44
44
  observers.delete(subscription);
45
45
  }
@@ -49,7 +49,7 @@ export function resetAllResultObserversForTests() {
49
49
  observers.clear();
50
50
  }
51
51
  function removeObserverByOwnerAndKey(owner, key) {
52
- for (const subscription of [...observers]) {
52
+ for (const subscription of observers) {
53
53
  if (subscription.owner === owner && subscription.key === key) {
54
54
  observers.delete(subscription);
55
55
  }
@@ -367,6 +367,8 @@ export interface ScorerResultEntry {
367
367
  export interface PathgradeTestMeta {
368
368
  score: number;
369
369
  scorers: ScorerResultEntry[];
370
+ /** Stable identity for this evaluation definition across separate runs. */
371
+ evaluationDefinitionKey?: string;
370
372
  trial?: TrialResult;
371
373
  diagnostics?: DiagnosticsReport;
372
374
  resultKind?: EvaluationResultKind;
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@wix/pathgrade",
3
- "version": "1.0.2",
3
+ "version": "1.0.4",
4
4
  "packageManager": "yarn@4.12.0",
5
5
  "description": "Evaluate whether AI agents discover and use your skills correctly",
6
6
  "exports": {
@@ -59,14 +59,6 @@
59
59
  "README.md"
60
60
  ],
61
61
  "scripts": {
62
- "lint": "oxlint src tests evals examples scripts --deny-warnings --disable-unicorn-plugin --disable-oxc-plugin --allow no-unused-vars --allow no-underscore-dangle --allow no-shadow --allow preserve-caught-error --allow no-unsafe-optional-chaining --allow no-unreachable --ignore-pattern tests/fixtures --ignore-pattern .worktrees",
63
- "knip": "yarn quality:deps",
64
- "knip:unused": "knip --exclude binaries --no-config-hints --max-show-issues 20",
65
- "quality:deps": "knip --dependencies --exclude binaries --no-config-hints --max-show-issues 20",
66
- "quality:arch": "sentrux check .",
67
- "quality:boundaries": "scripts/check-quality-boundaries.sh",
68
- "quality:max-lines": "scripts/check-max-lines.sh .",
69
- "quality": "yarn lint && yarn quality:deps && yarn quality:arch && yarn quality:boundaries && yarn quality:max-lines",
70
62
  "test": "yarn build && vitest run && yarn test:runner-cli-smoke",
71
63
  "test:runner-cli-smoke": "node tests/runner-cli-smoke/run-smokes.mjs",
72
64
  "test:evals": "vitest run --config evals/vitest.config.mts",
@@ -134,5 +126,5 @@
134
126
  "typescript": "^5.9.3",
135
127
  "zod": "4.3.6"
136
128
  },
137
- "falconPackageHash": "ccc242dcd7240e8c890225f5dd2f45be54f0f3b3426b862b95d847f1"
129
+ "falconPackageHash": "112cde138733c678a01778b225c1b3891d22417d98bbf2c479f8734c"
138
130
  }