agent-inspect 6.7.2 → 6.7.3

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 (39) hide show
  1. package/CHANGELOG.md +6 -0
  2. package/README.md +1 -1
  3. package/docs/EXPORTS.md +9 -0
  4. package/docs/PERFORMANCE.md +18 -0
  5. package/docs/STANDARDS.md +4 -0
  6. package/package.json +2 -2
  7. package/packages/cli/dist/{chunk-M6PC5KUT.mjs → chunk-EMYDJREA.mjs} +26 -10
  8. package/packages/cli/dist/chunk-EMYDJREA.mjs.map +1 -0
  9. package/packages/cli/dist/index.cjs +46 -21
  10. package/packages/cli/dist/index.cjs.map +1 -1
  11. package/packages/cli/dist/index.mjs +23 -14
  12. package/packages/cli/dist/index.mjs.map +1 -1
  13. package/packages/cli/dist/{src-27LC2BB3.mjs → src-IPJNFV37.mjs} +3 -3
  14. package/packages/cli/dist/{src-27LC2BB3.mjs.map → src-IPJNFV37.mjs.map} +1 -1
  15. package/packages/core/dist/advanced.cjs +24 -8
  16. package/packages/core/dist/advanced.cjs.map +1 -1
  17. package/packages/core/dist/advanced.mjs +25 -9
  18. package/packages/core/dist/advanced.mjs.map +1 -1
  19. package/packages/core/dist/checks.cjs +56 -4
  20. package/packages/core/dist/checks.cjs.map +1 -1
  21. package/packages/core/dist/checks.mjs +1 -1
  22. package/packages/core/dist/{chunk-EWHN6NA4.mjs → chunk-EQVXDGGU.mjs} +58 -6
  23. package/packages/core/dist/chunk-EQVXDGGU.mjs.map +1 -0
  24. package/packages/core/dist/diff.cjs +4 -0
  25. package/packages/core/dist/diff.cjs.map +1 -1
  26. package/packages/core/dist/diff.mjs +4 -0
  27. package/packages/core/dist/diff.mjs.map +1 -1
  28. package/packages/core/dist/exporters.cjs +9 -4
  29. package/packages/core/dist/exporters.cjs.map +1 -1
  30. package/packages/core/dist/exporters.d.cts +3 -2
  31. package/packages/core/dist/exporters.d.ts +3 -2
  32. package/packages/core/dist/exporters.mjs +9 -4
  33. package/packages/core/dist/exporters.mjs.map +1 -1
  34. package/packages/core/dist/logs.cjs +3 -2
  35. package/packages/core/dist/logs.cjs.map +1 -1
  36. package/packages/core/dist/logs.mjs +3 -2
  37. package/packages/core/dist/logs.mjs.map +1 -1
  38. package/packages/cli/dist/chunk-M6PC5KUT.mjs.map +0 -1
  39. package/packages/core/dist/chunk-EWHN6NA4.mjs.map +0 -1
@@ -1,4 +1,4 @@
1
- import { runTraceChecks, createRunStatusRule, createToolUsageRule, createRunDurationRule, createLlmUsageRule, createObservedOutcomeRule } from './chunk-EWHN6NA4.mjs';
1
+ import { runTraceChecks, createRunStatusRule, createToolUsageRule, createRunDurationRule, createLlmUsageRule, createObservedOutcomeRule } from './chunk-EQVXDGGU.mjs';
2
2
  export { MAX_TERMINAL_DEPTH, MAX_TERMINAL_NAME_LENGTH, TERMINAL_INDENT, TraceDirectory, createInspector, createInspectorRuntime, formatTerminalName, getCurrentContext, getCurrentCorrelationMetadata, getCurrentDepth, getCurrentRunId, getCurrentRunName, getCurrentStepId, getIndent, getParentStepId, getTraceDirFromContext, getTraceSafetyFromContext, hasActiveContext, isAgentInspectEnabled, isSilentContext, maybeInspectRun, printError, printFailedAt, printRunComplete, printRunStart, printStepComplete, printStepStart, renderErrorLine, renderRunSummary, renderStepLine, resolveTraceDir, runWithContext, runWithStepContext } from './chunk-ZWA5RWMV.mjs';
3
3
  import { buildRunSummary, extractMetadata } from './chunk-3USJVBLA.mjs';
4
4
  export { buildRunSummary, buildRunTimeline, buildRunWhatSummary, extractMetadata, renderRunWhat, renderTimeline } from './chunk-3USJVBLA.mjs';
@@ -2417,7 +2417,9 @@ function compareCohortAggregates(groups, options) {
2417
2417
  metric,
2418
2418
  "Average duration (ms)",
2419
2419
  baselineAgg?.avgDurationMs,
2420
- candidateAgg?.avgDurationMs
2420
+ candidateAgg?.avgDurationMs,
2421
+ true,
2422
+ tolerance
2421
2423
  );
2422
2424
  if (item) comparisons.push(item);
2423
2425
  break;
@@ -2427,7 +2429,9 @@ function compareCohortAggregates(groups, options) {
2427
2429
  metric,
2428
2430
  "Average LLM calls",
2429
2431
  baselineAgg?.avgLlmCallCount,
2430
- candidateAgg?.avgLlmCallCount
2432
+ candidateAgg?.avgLlmCallCount,
2433
+ true,
2434
+ tolerance
2431
2435
  );
2432
2436
  if (item) comparisons.push(item);
2433
2437
  break;
@@ -2437,7 +2441,9 @@ function compareCohortAggregates(groups, options) {
2437
2441
  metric,
2438
2442
  "Average token usage",
2439
2443
  baselineAgg?.avgTokenUsage,
2440
- candidateAgg?.avgTokenUsage
2444
+ candidateAgg?.avgTokenUsage,
2445
+ true,
2446
+ tolerance
2441
2447
  );
2442
2448
  if (item) comparisons.push(item);
2443
2449
  break;
@@ -2447,7 +2453,9 @@ function compareCohortAggregates(groups, options) {
2447
2453
  metric,
2448
2454
  "Average retries",
2449
2455
  baselineAgg?.avgRetryCount,
2450
- candidateAgg?.avgRetryCount
2456
+ candidateAgg?.avgRetryCount,
2457
+ true,
2458
+ tolerance
2451
2459
  );
2452
2460
  if (item) comparisons.push(item);
2453
2461
  break;
@@ -2457,7 +2465,9 @@ function compareCohortAggregates(groups, options) {
2457
2465
  metric,
2458
2466
  "Observation failure rate",
2459
2467
  baselineAgg?.observationFailureRate,
2460
- candidateAgg?.observationFailureRate
2468
+ candidateAgg?.observationFailureRate,
2469
+ true,
2470
+ tolerance
2461
2471
  );
2462
2472
  if (item) comparisons.push(item);
2463
2473
  break;
@@ -2493,7 +2503,9 @@ function compareCohortAggregates(groups, options) {
2493
2503
  metric,
2494
2504
  "Guardrail failures",
2495
2505
  baselineAgg?.avgGuardrailFailures,
2496
- candidateAgg?.avgGuardrailFailures
2506
+ candidateAgg?.avgGuardrailFailures,
2507
+ true,
2508
+ tolerance
2497
2509
  );
2498
2510
  if (item) comparisons.push(item);
2499
2511
  break;
@@ -2503,7 +2515,9 @@ function compareCohortAggregates(groups, options) {
2503
2515
  metric,
2504
2516
  "Circuit violations",
2505
2517
  baselineAgg?.avgCircuitViolations,
2506
- candidateAgg?.avgCircuitViolations
2518
+ candidateAgg?.avgCircuitViolations,
2519
+ true,
2520
+ tolerance
2507
2521
  );
2508
2522
  if (item) comparisons.push(item);
2509
2523
  break;
@@ -2513,7 +2527,9 @@ function compareCohortAggregates(groups, options) {
2513
2527
  metric,
2514
2528
  "Redaction warnings",
2515
2529
  baselineAgg?.avgRedactionWarnings,
2516
- candidateAgg?.avgRedactionWarnings
2530
+ candidateAgg?.avgRedactionWarnings,
2531
+ true,
2532
+ tolerance
2517
2533
  );
2518
2534
  if (item) comparisons.push(item);
2519
2535
  break;