agentv 3.13.1 → 3.13.2

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.
@@ -22,7 +22,7 @@ import {
22
22
  validateFileReferences,
23
23
  validateTargetsFile,
24
24
  writeArtifactsFromResults
25
- } from "./chunk-LSXO22CF.js";
25
+ } from "./chunk-4Z5E5CYT.js";
26
26
  import {
27
27
  createBuiltinRegistry,
28
28
  executeScript,
@@ -39,7 +39,7 @@ import {
39
39
  toSnakeCaseDeep as toSnakeCaseDeep2,
40
40
  transpileEvalYamlFile,
41
41
  trimBaselineResult
42
- } from "./chunk-K747KGDP.js";
42
+ } from "./chunk-D3LNJUUB.js";
43
43
  import {
44
44
  __commonJS,
45
45
  __esm,
@@ -4176,11 +4176,16 @@ var evalRunCommand = command({
4176
4176
  type: optional(string),
4177
4177
  long: "output-messages",
4178
4178
  description: 'Number of trailing messages to include in results output (default: 1, or "all")'
4179
+ }),
4180
+ threshold: option({
4181
+ type: optional(number),
4182
+ long: "threshold",
4183
+ description: "Suite-level quality gate: exit 1 if mean score falls below this value (0-1)"
4179
4184
  })
4180
4185
  },
4181
4186
  handler: async (args) => {
4182
4187
  if (args.evalPaths.length === 0 && process.stdin.isTTY) {
4183
- const { launchInteractiveWizard } = await import("./interactive-76ZJVPI7.js");
4188
+ const { launchInteractiveWizard } = await import("./interactive-HVKLYGRX.js");
4184
4189
  await launchInteractiveWizard();
4185
4190
  return;
4186
4191
  }
@@ -4216,9 +4221,13 @@ var evalRunCommand = command({
4216
4221
  artifacts: args.artifacts,
4217
4222
  graderTarget: args.graderTarget,
4218
4223
  model: args.model,
4219
- outputMessages: args.outputMessages
4224
+ outputMessages: args.outputMessages,
4225
+ threshold: args.threshold
4220
4226
  };
4221
- await runEvalCommand({ testFiles: resolvedPaths, rawOptions });
4227
+ const result = await runEvalCommand({ testFiles: resolvedPaths, rawOptions });
4228
+ if (result?.thresholdFailed) {
4229
+ process.exit(1);
4230
+ }
4222
4231
  }
4223
4232
  });
4224
4233
 
@@ -7756,4 +7765,4 @@ export {
7756
7765
  preprocessArgv,
7757
7766
  runCli
7758
7767
  };
7759
- //# sourceMappingURL=chunk-UK7UMQOX.js.map
7768
+ //# sourceMappingURL=chunk-X2343WOK.js.map