agentv 5.2.0-next.1 → 5.3.0-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 (33) hide show
  1. package/dist/{artifact-writer-SN6UNL5G.js → artifact-writer-JFNIPMKW.js} +4 -4
  2. package/dist/{chunk-3S6A2RKR.js → chunk-6ZZCDZPD.js} +1357 -304
  3. package/dist/chunk-6ZZCDZPD.js.map +1 -0
  4. package/dist/{chunk-46K2OET3.js → chunk-BHKQHG26.js} +2 -2
  5. package/dist/{chunk-T4AD3H3Y.js → chunk-FVR4RQFK.js} +74 -16
  6. package/dist/chunk-FVR4RQFK.js.map +1 -0
  7. package/dist/{chunk-3X357HS4.js → chunk-T32NL3E6.js} +62 -6
  8. package/dist/chunk-T32NL3E6.js.map +1 -0
  9. package/dist/{chunk-ALVZQUZP.js → chunk-V52ATPTT.js} +2 -2
  10. package/dist/cli.js +5 -5
  11. package/dist/{dist-DEPJOMCA.js → dist-6Z7U473R.js} +11 -3
  12. package/dist/index.js +5 -5
  13. package/dist/{interactive-36ZNMQB2.js → interactive-RUY3OCBI.js} +5 -5
  14. package/dist/skills/agentv-bench/SKILL.md +1 -1
  15. package/dist/skills/agentv-bench/agents/analyzer.md +1 -1
  16. package/dist/skills/agentv-bench/agents/comparator.md +3 -3
  17. package/dist/skills/agentv-bench/agents/grader.md +4 -2
  18. package/dist/skills/agentv-bench/references/description-optimization.md +5 -6
  19. package/dist/skills/agentv-bench/references/environment-adaptation.md +2 -2
  20. package/dist/skills/agentv-bench/references/eval-yaml-spec.md +20 -9
  21. package/dist/skills/agentv-eval-migrations/references/breaking-changes.md +100 -2
  22. package/dist/skills/agentv-eval-writer/SKILL.md +27 -16
  23. package/dist/{ts-eval-loader-OMG2JBHP-OP5RR7A3.js → ts-eval-loader-DQDYRULE-V3377FOL.js} +2 -2
  24. package/package.json +1 -1
  25. package/dist/chunk-3S6A2RKR.js.map +0 -1
  26. package/dist/chunk-3X357HS4.js.map +0 -1
  27. package/dist/chunk-T4AD3H3Y.js.map +0 -1
  28. /package/dist/{artifact-writer-SN6UNL5G.js.map → artifact-writer-JFNIPMKW.js.map} +0 -0
  29. /package/dist/{chunk-46K2OET3.js.map → chunk-BHKQHG26.js.map} +0 -0
  30. /package/dist/{chunk-ALVZQUZP.js.map → chunk-V52ATPTT.js.map} +0 -0
  31. /package/dist/{dist-DEPJOMCA.js.map → dist-6Z7U473R.js.map} +0 -0
  32. /package/dist/{interactive-36ZNMQB2.js.map → interactive-RUY3OCBI.js.map} +0 -0
  33. /package/dist/{ts-eval-loader-OMG2JBHP-OP5RR7A3.js.map → ts-eval-loader-DQDYRULE-V3377FOL.js.map} +0 -0
@@ -61,11 +61,11 @@ import {
61
61
  validateFileReferences,
62
62
  validateTargetsFile,
63
63
  validateWorkspacePaths
64
- } from "./chunk-T4AD3H3Y.js";
64
+ } from "./chunk-FVR4RQFK.js";
65
65
  import {
66
66
  materializeEvalBundle,
67
67
  writeArtifactsFromResults as writeArtifactsFromResults2
68
- } from "./chunk-46K2OET3.js";
68
+ } from "./chunk-BHKQHG26.js";
69
69
  import {
70
70
  discoverClaudeSessions,
71
71
  discoverCodexSessions,
@@ -80,12 +80,14 @@ import {
80
80
  syncProjects,
81
81
  transpileEvalYamlFile,
82
82
  trimBaselineResult
83
- } from "./chunk-ALVZQUZP.js";
83
+ } from "./chunk-V52ATPTT.js";
84
84
  import {
85
85
  AGENTV_RESULTS_ARTIFACTS_REF,
86
86
  DEFAULT_CATEGORY,
87
87
  DEFAULT_THRESHOLD,
88
88
  EXECUTION_TRACE_SCHEMA_VERSION,
89
+ LEGACY_METRICS_SCHEMA_VERSION,
90
+ METRICS_SCHEMA_VERSION,
89
91
  RESULT_INDEX_FILENAME,
90
92
  addProject,
91
93
  aggregateRunDir,
@@ -110,6 +112,7 @@ import {
110
112
  loadTestSuite,
111
113
  mergeConfigObjects,
112
114
  normalizeCategoryPath,
115
+ normalizeMetricsArtifactWire,
113
116
  normalizeResultRow,
114
117
  omitExternalTraceMetadataKeys,
115
118
  parseYamlValue,
@@ -139,7 +142,7 @@ import {
139
142
  traceEnvelopeToTranscriptMessages,
140
143
  traceFromTranscriptJsonLines,
141
144
  writeArtifactsFromResults
142
- } from "./chunk-3S6A2RKR.js";
145
+ } from "./chunk-6ZZCDZPD.js";
143
146
  import {
144
147
  __commonJS,
145
148
  __require,
@@ -12947,6 +12950,44 @@ function validateNoLegacyGradingFields(value, pathLabel) {
12947
12950
  }
12948
12951
  return errors;
12949
12952
  }
12953
+ function validateMetricsArtifact(value) {
12954
+ if (!isRecord2(value)) {
12955
+ return [{ severity: "error", message: "metrics.json must be an object" }];
12956
+ }
12957
+ const diagnostics = [];
12958
+ const schemaVersion = value.schema_version;
12959
+ const isCurrentSchema = schemaVersion === METRICS_SCHEMA_VERSION;
12960
+ const isLegacySchema = schemaVersion === LEGACY_METRICS_SCHEMA_VERSION;
12961
+ if (isCurrentSchema && Object.hasOwn(value, "trace")) {
12962
+ diagnostics.push({
12963
+ severity: "error",
12964
+ message: "metrics.json must not include trace in agentv.metrics.v2"
12965
+ });
12966
+ } else if (isLegacySchema && Object.hasOwn(value, "trace")) {
12967
+ diagnostics.push({
12968
+ severity: "warning",
12969
+ message: "metrics.json uses legacy trace identity; readers ignore it"
12970
+ });
12971
+ }
12972
+ if (isCurrentSchema && Object.hasOwn(value, "metrics")) {
12973
+ diagnostics.push({
12974
+ severity: "error",
12975
+ message: "metrics.json must not include nested metrics in agentv.metrics.v2"
12976
+ });
12977
+ } else if (isLegacySchema && Object.hasOwn(value, "metrics")) {
12978
+ diagnostics.push({
12979
+ severity: "warning",
12980
+ message: "metrics.json uses legacy nested metrics; readers flatten it"
12981
+ });
12982
+ }
12983
+ if (!normalizeMetricsArtifactWire(value)) {
12984
+ diagnostics.push({
12985
+ severity: "error",
12986
+ message: "metrics.json does not match a supported metrics artifact shape"
12987
+ });
12988
+ }
12989
+ return diagnostics;
12990
+ }
12950
12991
  function checkDirectoryNaming(runDir) {
12951
12992
  const dirName = path24.basename(runDir);
12952
12993
  const pathSegments = path24.normalize(runDir).split(path24.sep).filter(Boolean);
@@ -13163,6 +13204,21 @@ function checkArtifactFiles(runDir, entries2) {
13163
13204
  severity: "warning",
13164
13205
  message: `${testId}: metrics.json not found at '${entry.metrics_path}'`
13165
13206
  });
13207
+ } else {
13208
+ try {
13209
+ const metrics = JSON.parse(readFileSync12(metricsPath, "utf8"));
13210
+ for (const diagnostic of validateMetricsArtifact(metrics)) {
13211
+ diagnostics.push({
13212
+ severity: diagnostic.severity,
13213
+ message: `${testId}: ${diagnostic.message}`
13214
+ });
13215
+ }
13216
+ } catch {
13217
+ diagnostics.push({
13218
+ severity: "error",
13219
+ message: `${testId}: metrics.json is not valid JSON`
13220
+ });
13221
+ }
13166
13222
  }
13167
13223
  }
13168
13224
  }
@@ -14550,7 +14606,7 @@ function buildRepeatTrialReadModels(baseDir, record) {
14550
14606
  const answerPath = caseTrialArtifactPath(resultDir, runPath, "outputs/answer.md");
14551
14607
  const resultPath = caseTrialArtifactPath(resultDir, runPath, "result.json");
14552
14608
  const runResult = readArtifactJsonObject(baseDir, resultPath);
14553
- const metrics = readArtifactJsonObject(baseDir, metricsPath);
14609
+ const metrics = normalizeMetricsArtifactWire(readArtifactJsonObject(baseDir, metricsPath));
14554
14610
  const timing = readArtifactJsonObject(baseDir, timingPath);
14555
14611
  const toolCalls = objectField(metrics, "tool_calls");
14556
14612
  const tokenUsage = objectField(metrics, "tokens") ?? objectField(timing, "token_usage");
@@ -17914,4 +17970,4 @@ export {
17914
17970
  preprocessArgv,
17915
17971
  runCli
17916
17972
  };
17917
- //# sourceMappingURL=chunk-3X357HS4.js.map
17973
+ //# sourceMappingURL=chunk-T32NL3E6.js.map