agentv 4.5.1 → 4.6.0

package/dist/{chunk-BQC2CDLN.js → chunk-U2LSJ6Y4.js}

@@ -24,7 +24,7 @@ import {
   validateFileReferences,
   validateTargetsFile,
   writeArtifactsFromResults
-} from "./chunk-5DEZ72J3.js";
+} from "./chunk-5GZJIXTY.js";
 import {
   DEFAULT_CATEGORY,
   PASS_THRESHOLD,
@@ -50,7 +50,7 @@ import {
   toSnakeCaseDeep as toSnakeCaseDeep2,
   transpileEvalYamlFile,
   trimBaselineResult
-} from "./chunk-7DRAXDVC.js";
+} from "./chunk-KQQTEWZF.js";
 import {
   __commonJS,
   __require,
@@ -3960,11 +3960,21 @@ var evalRunCommand = command({
       type: optional(number),
       long: "threshold",
       description: "Per-test score threshold (0-1, default 0.8). Exit 1 if any test scores below this value"
+    }),
+    tag: multioption({
+      type: array(string),
+      long: "tag",
+      description: "Only run eval files that have this tag (repeatable, AND logic)"
+    }),
+    excludeTag: multioption({
+      type: array(string),
+      long: "exclude-tag",
+      description: "Skip eval files that have this tag (repeatable, file skipped if any match)"
     })
   },
   handler: async (args) => {
     if (args.evalPaths.length === 0 && process.stdin.isTTY) {
-      const { launchInteractiveWizard } = await import("./interactive-OG7ZJIHG.js");
+      const { launchInteractiveWizard } = await import("./interactive-6D3ULOMN.js");
       await launchInteractiveWizard();
       return;
     }
@@ -4001,7 +4011,9 @@ var evalRunCommand = command({
       graderTarget: args.graderTarget,
       model: args.model,
       outputMessages: args.outputMessages,
-      threshold: args.threshold
+      threshold: args.threshold,
+      tag: args.tag,
+      excludeTag: args.excludeTag
     };
     const result = await runEvalCommand({ testFiles: resolvedPaths, rawOptions });
     if (result?.thresholdFailed) {
@@ -8464,6 +8476,7 @@ var app = subcommands({
     pipeline: pipelineCommand,
     results: resultsCommand,
     self: selfCommand,
+    serve: resultsServeCommand,
     studio: resultsServeCommand,
     trace: traceCommand,
     transpile: transpileCommand,
@@ -8482,6 +8495,7 @@ var TOP_LEVEL_COMMANDS = /* @__PURE__ */ new Set([
   "pipeline",
   "results",
   "self",
+  "serve",
   "studio",
   "trace",
   "transpile",
@@ -8529,4 +8543,4 @@ export {
   preprocessArgv,
   runCli
 };
-//# sourceMappingURL=chunk-BQC2CDLN.js.map
+//# sourceMappingURL=chunk-U2LSJ6Y4.js.map