@wix/pathgrade 1.0.38 → 1.0.39

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 (34) hide show
  1. package/README.md +1 -1
  2. package/dist/agents/claude/sdk-message-projector.js +4 -5
  3. package/dist/agents/claude/tool-permission-bridge.js +2 -1
  4. package/dist/agents/codex-app-server/item-projection.js +2 -2
  5. package/dist/agents/codex-app-server/mcp-approval-correlator.js +3 -2
  6. package/dist/agents/opencode.js +4 -5
  7. package/dist/commands/report.js +5 -25
  8. package/dist/internal/direct-mcp-v2/acp-author-projector.js +20 -7
  9. package/dist/reporters/cli.js +20 -1
  10. package/dist/reporters/github-comment.js +18 -7
  11. package/dist/reporters/loader.d.ts +4 -0
  12. package/dist/reporters/loader.js +26 -9
  13. package/dist/reporting/core.js +17 -7
  14. package/dist/reporting/report-parser.js +57 -2
  15. package/dist/reporting/types.d.ts +2 -1
  16. package/dist/runners/orchestrator.js +3 -4
  17. package/dist/runners/repeated-invocation.js +2 -5
  18. package/dist/runners/report-projection.js +1 -0
  19. package/dist/sdk/agent.js +10 -1
  20. package/dist/sdk/evaluate.js +58 -3
  21. package/dist/sdk/judge-prompt-builder.js +11 -7
  22. package/dist/sdk/mcp-event-input.d.ts +1 -0
  23. package/dist/sdk/mcp-event-input.js +3 -0
  24. package/dist/sdk/mcp-evidence.js +16 -3
  25. package/dist/sdk/mcp-safety.js +2 -2
  26. package/dist/sdk/scripted-mcp-events.js +3 -2
  27. package/dist/sdk/tool-event-log.js +18 -3
  28. package/dist/sdk/tool-event-secrets.d.ts +4 -0
  29. package/dist/sdk/tool-event-secrets.js +33 -2
  30. package/dist/sdk/types.d.ts +2 -0
  31. package/dist/tool-event-results.d.ts +3 -0
  32. package/dist/tool-event-results.js +153 -23
  33. package/dist/types.d.ts +9 -5
  34. package/package.json +2 -2
package/dist/types.d.ts CHANGED
@@ -185,6 +185,10 @@ export interface EvalReport {
185
185
  task: string;
186
186
  /** Canonical status for this report or consolidated group when available. */
187
187
  status?: 'pass' | 'fail';
188
+ /** Whether every reportable runner attempt passed. */
189
+ runner_status?: 'pass' | 'fail';
190
+ /** Whether mean reward meets the configured threshold. */
191
+ threshold_status?: 'pass' | 'fail' | 'not_configured';
188
192
  /** Schema-v2 canonical average score. */
189
193
  mean_reward?: number;
190
194
  /** Binary success frequency, absent when attempts are ineligible. */
@@ -266,11 +270,11 @@ export interface PathgradeReport {
266
270
  overall_mean_reward?: number;
267
271
  attempts_requested?: number;
268
272
  attempts_completed?: number;
269
- /**
270
- * Threshold check result. When `threshold` is set: `'pass'` iff
271
- * `overall_pass_rate >= threshold`. Otherwise: `'pass'` iff every trial
272
- * in every group passed its vitest test.
273
- */
273
+ /** Whether every reportable runner attempt passed. */
274
+ runner_status?: 'pass' | 'fail';
275
+ /** Whether overall mean reward meets the configured threshold. */
276
+ threshold_status?: 'pass' | 'fail' | 'not_configured';
277
+ /** Conjunction of runner assertions and the optional aggregate threshold. */
274
278
  status: 'pass' | 'fail';
275
279
  /** Agent Evals compatibility manifest when changed selection finds no runnable evals. */
276
280
  run_kind?: 'evaluation' | 'no-affected';
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@wix/pathgrade",
3
- "version": "1.0.38",
3
+ "version": "1.0.39",
4
4
  "packageManager": "yarn@4.12.0",
5
5
  "description": "Evaluate whether AI agents discover and use your skills correctly",
6
6
  "exports": {
@@ -142,5 +142,5 @@
142
142
  "typescript": "^5.9.3",
143
143
  "zod": "4.3.6"
144
144
  },
145
- "falconPackageHash": "199b36204420a97fad83bac6e53d8b614d2b6592cb701e7db41d3bad"
145
+ "falconPackageHash": "df08d3b446a37c3224730fca89a408f6238eef420358705733692eee"
146
146
  }