agentv 4.32.0-next.1 → 4.33.0-next.1

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 (37) hide show
  1. package/README.md +5 -5
  2. package/dist/{artifact-writer-VDF7KRWL.js → artifact-writer-PLUZAGYD.js} +4 -4
  3. package/dist/chunk-2TDFJ2FZ.js +1234 -0
  4. package/dist/chunk-2TDFJ2FZ.js.map +1 -0
  5. package/dist/{chunk-TAZBCVEZ.js → chunk-6QEIZ33V.js} +1340 -279
  6. package/dist/chunk-6QEIZ33V.js.map +1 -0
  7. package/dist/{chunk-5JMFFG36.js → chunk-HE3TRGCU.js} +786 -260
  8. package/dist/chunk-HE3TRGCU.js.map +1 -0
  9. package/dist/{chunk-LX5AK3P7.js → chunk-QKZLKDXB.js} +568 -191
  10. package/dist/chunk-QKZLKDXB.js.map +1 -0
  11. package/dist/{chunk-IGTRNQAM.js → chunk-VASYLLBA.js} +1650 -453
  12. package/dist/chunk-VASYLLBA.js.map +1 -0
  13. package/dist/cli.js +5 -5
  14. package/dist/dashboard/assets/index-DBg02hJh.js +118 -0
  15. package/dist/dashboard/assets/index-F8mrCRqc.css +1 -0
  16. package/dist/dashboard/assets/{index-BdoQWnyM.js → index-w2eILJoK.js} +1 -1
  17. package/dist/dashboard/index.html +3 -3
  18. package/dist/{dist-GICSKMNP.js → dist-W63D37ZR.js} +58 -6
  19. package/dist/index.js +5 -5
  20. package/dist/{interactive-GIDBBDYZ.js → interactive-T6F2GRDI.js} +5 -5
  21. package/dist/skills/agentv-eval-writer/SKILL.md +2 -1
  22. package/dist/skills/agentv-eval-writer/references/eval-schema.json +104 -0
  23. package/dist/skills/agentv-eval-writer/references/rubric-evaluator.md +20 -0
  24. package/dist/{ts-eval-loader-Z6IUSDNA-YBOE4JIQ.js → ts-eval-loader-EQJX3OLT-THE7D3GR.js} +2 -2
  25. package/package.json +2 -2
  26. package/dist/chunk-2ZEY3WBH.js +0 -729
  27. package/dist/chunk-2ZEY3WBH.js.map +0 -1
  28. package/dist/chunk-5JMFFG36.js.map +0 -1
  29. package/dist/chunk-IGTRNQAM.js.map +0 -1
  30. package/dist/chunk-LX5AK3P7.js.map +0 -1
  31. package/dist/chunk-TAZBCVEZ.js.map +0 -1
  32. package/dist/dashboard/assets/index-DcPH8PyS.css +0 -1
  33. package/dist/dashboard/assets/index-EXkiwqam.js +0 -116
  34. /package/dist/{artifact-writer-VDF7KRWL.js.map → artifact-writer-PLUZAGYD.js.map} +0 -0
  35. /package/dist/{dist-GICSKMNP.js.map → dist-W63D37ZR.js.map} +0 -0
  36. /package/dist/{interactive-GIDBBDYZ.js.map → interactive-T6F2GRDI.js.map} +0 -0
  37. /package/dist/{ts-eval-loader-Z6IUSDNA-YBOE4JIQ.js.map → ts-eval-loader-EQJX3OLT-THE7D3GR.js.map} +0 -0
package/README.md CHANGED
@@ -77,9 +77,9 @@ agentv compare .agentv/results/runs/<timestamp>/index.jsonl
77
77
  ## Output formats
78
78
 
79
79
  ```bash
80
- agentv eval evals/my-eval.yaml # JSONL (default)
81
- agentv eval evals/my-eval.yaml -o report.html # HTML dashboard
82
- agentv eval evals/my-eval.yaml -o results.xml # JUnit XML for CI
80
+ agentv eval evals/my-eval.yaml --output ./run # writes ./run/index.jsonl
81
+ agentv eval evals/my-eval.yaml --export report.html
82
+ agentv eval evals/my-eval.yaml --export results.xml # JUnit XML for CI
83
83
  ```
84
84
 
85
85
  ## TypeScript SDK
@@ -135,8 +135,8 @@ AGENTV_RESULTS_REPO=EntityProcess/agentv-evalresults \
135
135
  ```
136
136
 
137
137
  The script clones AgentV examples into `~/agentv-dashboard`, clones the results
138
- repo, writes the Dashboard project registry, builds the Docker image, and starts
139
- Dashboard at `http://localhost:3117`.
138
+ repo, writes the Dashboard project registry under `$AGENTV_HOME/config.yaml`,
139
+ builds the Docker image, and starts Dashboard at `http://localhost:3117`.
140
140
 
141
141
  ## License
142
142
 
package/dist/{artifact-writer-VDF7KRWL.js → artifact-writer-PLUZAGYD.js} RENAMED
@@ -14,11 +14,11 @@ import {
14
14
  writeArtifactsFromResults,
15
15
  writeInitialBenchmarkArtifact,
16
16
  writePerTestArtifacts
17
- } from "./chunk-2ZEY3WBH.js";
18
- import "./chunk-LX5AK3P7.js";
17
+ } from "./chunk-2TDFJ2FZ.js";
18
+ import "./chunk-QKZLKDXB.js";
19
19
  import "./chunk-QOBQ5XYF.js";
20
20
  import "./chunk-BPGJ4HBU.js";
21
- import "./chunk-TAZBCVEZ.js";
21
+ import "./chunk-6QEIZ33V.js";
22
22
  import "./chunk-NPVGBFF6.js";
23
23
  import "./chunk-M7BUKBAF.js";
24
24
  import "./chunk-5H446C7X.js";
@@ -38,4 +38,4 @@ export {
38
38
  writeInitialBenchmarkArtifact,
39
39
  writePerTestArtifacts
40
40
  };
41
- //# sourceMappingURL=artifact-writer-VDF7KRWL.js.map
41
+ //# sourceMappingURL=artifact-writer-PLUZAGYD.js.map