agentv 4.23.0-next.1 → 4.24.0

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 (27) hide show
  1. package/dist/{artifact-writer-KBYDJNKS.js → artifact-writer-7CKE5ILM.js} +4 -4
  2. package/dist/{chunk-SBI4Z7YY.js → chunk-D5HCHBPT.js} +2 -2
  3. package/dist/{chunk-32VOBCSX.js → chunk-EFA3SBAY.js} +24 -11
  4. package/dist/chunk-EFA3SBAY.js.map +1 -0
  5. package/dist/{chunk-KX2MJDMY.js → chunk-HL3MTSIW.js} +14 -9
  6. package/dist/{chunk-KX2MJDMY.js.map → chunk-HL3MTSIW.js.map} +1 -1
  7. package/dist/{chunk-ZVEESBXF.js → chunk-KTHVVL7Q.js} +3 -3
  8. package/dist/{chunk-4ISLX6RP.js → chunk-NM4NQSUT.js} +6 -6
  9. package/dist/cli.js +5 -5
  10. package/dist/{dist-A7D52JMJ.js → dist-IA3D556U.js} +3 -3
  11. package/dist/index.js +5 -5
  12. package/dist/{interactive-66Z7DNMJ.js → interactive-4NRIFB5R.js} +5 -5
  13. package/dist/studio/assets/index-DY-K3PKr.css +1 -0
  14. package/dist/studio/assets/{index-SZVrc1UE.js → index-DZpM1lzZ.js} +25 -25
  15. package/dist/studio/assets/{index-BVSHI8Eo.js → index-jBBmFdWd.js} +1 -1
  16. package/dist/studio/index.html +2 -2
  17. package/dist/{ts-eval-loader-HT7NCO4M-V63KENQZ.js → ts-eval-loader-SYQYQPMC-FKU23VX3.js} +2 -2
  18. package/package.json +1 -1
  19. package/dist/chunk-32VOBCSX.js.map +0 -1
  20. package/dist/studio/assets/index-DcpbFwI5.css +0 -1
  21. /package/dist/{artifact-writer-KBYDJNKS.js.map → artifact-writer-7CKE5ILM.js.map} +0 -0
  22. /package/dist/{chunk-SBI4Z7YY.js.map → chunk-D5HCHBPT.js.map} +0 -0
  23. /package/dist/{chunk-ZVEESBXF.js.map → chunk-KTHVVL7Q.js.map} +0 -0
  24. /package/dist/{chunk-4ISLX6RP.js.map → chunk-NM4NQSUT.js.map} +0 -0
  25. /package/dist/{dist-A7D52JMJ.js.map → dist-IA3D556U.js.map} +0 -0
  26. /package/dist/{interactive-66Z7DNMJ.js.map → interactive-4NRIFB5R.js.map} +0 -0
  27. /package/dist/{ts-eval-loader-HT7NCO4M-V63KENQZ.js.map → ts-eval-loader-SYQYQPMC-FKU23VX3.js.map} +0 -0
@@ -13,11 +13,11 @@ import {
13
13
  writeArtifacts,
14
14
  writeArtifactsFromResults,
15
15
  writePerTestArtifacts
16
- } from "./chunk-ZVEESBXF.js";
17
- import "./chunk-SBI4Z7YY.js";
16
+ } from "./chunk-KTHVVL7Q.js";
17
+ import "./chunk-D5HCHBPT.js";
18
18
  import "./chunk-QOBQ5XYF.js";
19
19
  import "./chunk-BPGJ4HBU.js";
20
- import "./chunk-KX2MJDMY.js";
20
+ import "./chunk-HL3MTSIW.js";
21
21
  import "./chunk-ZKO2LGRR.js";
22
22
  import "./chunk-NPVGBFF6.js";
23
23
  import "./chunk-HQDCIXVH.js";
@@ -38,4 +38,4 @@ export {
38
38
  writeArtifactsFromResults,
39
39
  writePerTestArtifacts
40
40
  };
41
- //# sourceMappingURL=artifact-writer-KBYDJNKS.js.map
41
+ //# sourceMappingURL=artifact-writer-7CKE5ILM.js.map
@@ -12,7 +12,7 @@ import {
12
12
  parseYamlValue,
13
13
  toCamelCaseDeep,
14
14
  toSnakeCaseDeep
15
- } from "./chunk-KX2MJDMY.js";
15
+ } from "./chunk-HL3MTSIW.js";
16
16
 
17
17
  // ../../packages/core/dist/index.js
18
18
  import { readFileSync } from "node:fs";
@@ -2269,4 +2269,4 @@ export {
2269
2269
  TranscriptProvider,
2270
2270
  createAgentKernel
2271
2271
  };
2272
- //# sourceMappingURL=chunk-SBI4Z7YY.js.map
2272
+ //# sourceMappingURL=chunk-D5HCHBPT.js.map
@@ -15,7 +15,7 @@ import {
15
15
  resolveWorkspaceOrFilePath,
16
16
  toSnakeCaseDeep,
17
17
  writeArtifactsFromResults
18
- } from "./chunk-ZVEESBXF.js";
18
+ } from "./chunk-KTHVVL7Q.js";
19
19
  import {
20
20
  ResponseCache,
21
21
  RunBudgetTracker,
@@ -31,7 +31,7 @@ import {
31
31
  shouldSkipCacheForTemperature,
32
32
  stageResultsArtifacts,
33
33
  syncResultsRepo
34
- } from "./chunk-SBI4Z7YY.js";
34
+ } from "./chunk-D5HCHBPT.js";
35
35
  import {
36
36
  CLI_PLACEHOLDERS,
37
37
  COMMON_TARGET_SETTINGS,
@@ -63,7 +63,7 @@ import {
63
63
  subscribeToCopilotSdkLogEntries,
64
64
  subscribeToPiLogEntries,
65
65
  toCamelCaseDeep
66
- } from "./chunk-KX2MJDMY.js";
66
+ } from "./chunk-HL3MTSIW.js";
67
67
 
68
68
  // src/commands/eval/shared.ts
69
69
  import { constants } from "node:fs";
@@ -179,7 +179,7 @@ async function findRepoRoot(start) {
179
179
  // package.json
180
180
  var package_default = {
181
181
  name: "agentv",
182
- version: "4.23.0-next.1",
182
+ version: "4.24.0",
183
183
  description: "CLI entry point for AgentV",
184
184
  type: "module",
185
185
  repository: {
@@ -2352,6 +2352,17 @@ function formatVerdict(score, verdict) {
2352
2352
  const color = verdict === "PASS" ? ANSI_GREEN2 : verdict === "FAIL" ? ANSI_RED2 : ANSI_YELLOW2;
2353
2353
  return ` | ${color}${ANSI_BOLD}${verdictLabel}${ANSI_RESET2}`;
2354
2354
  }
2355
+ function formatDurations(durationMs, totalDurationMs) {
2356
+ if (durationMs === void 0 && totalDurationMs === void 0) {
2357
+ return "";
2358
+ }
2359
+ if (durationMs !== void 0 && totalDurationMs !== void 0) {
2360
+ const normalizedTotalMs = Math.max(durationMs, totalDurationMs);
2361
+ return ` | ${durationMs}/${normalizedTotalMs}ms`;
2362
+ }
2363
+ const singleDurationMs = durationMs ?? totalDurationMs;
2364
+ return singleDurationMs !== void 0 ? ` | ${singleDurationMs}ms` : "";
2365
+ }
2355
2366
  var ProgressDisplay = class {
2356
2367
  workers = /* @__PURE__ */ new Map();
2357
2368
  totalTests = 0;
@@ -2396,14 +2407,14 @@ var ProgressDisplay = class {
2396
2407
  case "completed": {
2397
2408
  const icon = progress.verdict === "FAIL" ? "\u26A0\uFE0F" : progress.verdict === "ERROR" ? "\u274C" : "\u2705";
2398
2409
  console.log(
2399
- `${countPrefix} ${icon} ${progress.testId}${targetSuffix}${formatVerdict(progress.score, progress.verdict)}`
2410
+ `${countPrefix} ${icon} ${progress.testId}${targetSuffix}${formatVerdict(progress.score, progress.verdict)}${formatDurations(progress.durationMs, progress.totalDurationMs)}`
2400
2411
  );
2401
2412
  break;
2402
2413
  }
2403
2414
  case "failed": {
2404
2415
  const failIcon = progress.verdict === "ERROR" ? "\u274C" : "\u26A0\uFE0F";
2405
2416
  console.log(
2406
- `${countPrefix} ${failIcon} ${progress.testId}${targetSuffix}${formatVerdict(progress.score, progress.verdict)}${progress.error ? `: ${progress.error}` : ""}`
2417
+ `${countPrefix} ${failIcon} ${progress.testId}${targetSuffix}${formatVerdict(progress.score, progress.verdict)}${formatDurations(progress.durationMs, progress.totalDurationMs)}${progress.error ? `: ${progress.error}` : ""}`
2407
2418
  );
2408
2419
  break;
2409
2420
  }
@@ -5294,7 +5305,9 @@ async function runSingleEvalFile(params) {
5294
5305
  error: event.error,
5295
5306
  targetLabel: inlineTargetLabel,
5296
5307
  score: event.score,
5297
- verdict
5308
+ verdict,
5309
+ durationMs: event.durationMs,
5310
+ totalDurationMs: event.evalRunDurationMs
5298
5311
  });
5299
5312
  }
5300
5313
  });
@@ -5431,7 +5444,7 @@ async function runEvalCommand(input) {
5431
5444
  const useFileExport = !!options.otelFile;
5432
5445
  if (options.exportOtel || useFileExport) {
5433
5446
  try {
5434
- const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-A7D52JMJ.js");
5447
+ const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-IA3D556U.js");
5435
5448
  let endpoint = process.env.OTEL_EXPORTER_OTLP_ENDPOINT;
5436
5449
  let headers = {};
5437
5450
  if (options.otelBackend) {
@@ -5630,7 +5643,7 @@ async function runEvalCommand(input) {
5630
5643
  const activeTestFiles = resolvedTestFiles.filter((f) => fileMetadata.has(f));
5631
5644
  let transcriptProviderFactory;
5632
5645
  if (options.transcript) {
5633
- const { TranscriptProvider } = await import("./dist-A7D52JMJ.js");
5646
+ const { TranscriptProvider } = await import("./dist-IA3D556U.js");
5634
5647
  const transcriptProvider = await TranscriptProvider.fromFile(options.transcript);
5635
5648
  const totalTests = [...fileMetadata.values()].reduce(
5636
5649
  (sum, meta) => sum + meta.testCases.length,
@@ -5799,7 +5812,7 @@ async function runEvalCommand(input) {
5799
5812
  if (usesDefaultArtifactWorkspace && allResults.length > 0) {
5800
5813
  const evalFile = activeTestFiles.length === 1 ? activeTestFiles[0] : "";
5801
5814
  if (isResumeAppend) {
5802
- const { writePerTestArtifacts } = await import("./artifact-writer-KBYDJNKS.js");
5815
+ const { writePerTestArtifacts } = await import("./artifact-writer-7CKE5ILM.js");
5803
5816
  await writePerTestArtifacts(allResults, runDir, {
5804
5817
  experiment: normalizeExperimentName(options.experiment)
5805
5818
  });
@@ -6022,4 +6035,4 @@ export {
6022
6035
  getCategories,
6023
6036
  filterByCategory
6024
6037
  };
6025
- //# sourceMappingURL=chunk-32VOBCSX.js.map
6038
+ //# sourceMappingURL=chunk-EFA3SBAY.js.map