agentv 3.14.5 → 4.0.0

package/dist/{chunk-GUXXTOYK.js → chunk-OT2J474N.js}

@@ -1,11 +1,13 @@
 import { createRequire } from 'node:module'; const require = createRequire(import.meta.url);
 import {
   CLI_PLACEHOLDERS,
+  COMMON_TARGET_SETTINGS,
   KNOWN_PROVIDERS,
   PROVIDER_ALIASES,
   ResponseCache,
   buildDirectoryChain,
   buildSearchRoots,
+  deriveCategory,
   ensureVSCodeSubagents,
   findGitRoot,
   interpolateEnv,
@@ -27,12 +29,12 @@ import {
   subscribeToCopilotCliLogEntries,
   subscribeToCopilotSdkLogEntries,
   subscribeToPiLogEntries
-} from "./chunk-ELQEFMGO.js";
+} from "./chunk-OXBBWZOY.js";
 
 // package.json
 var package_default = {
   name: "agentv",
-  version: "3.14.5",
+  version: "4.0.0",
   description: "CLI entry point for AgentV",
   type: "module",
   repository: {
@@ -110,6 +112,27 @@ async function resolveEvalPaths(evalPaths, cwd) {
         results.add(candidatePath);
         continue;
       }
+      if (stats.isDirectory()) {
+        const dirGlob = path.posix.join(
+          candidatePath.replace(/\\/g, "/"),
+          "**/*.eval.{yaml,yml}"
+        );
+        const dirMatches = await fg(dirGlob, {
+          absolute: true,
+          onlyFiles: true,
+          unique: true,
+          dot: true,
+          followSymbolicLinks: true
+        });
+        if (dirMatches.length === 0) {
+          unmatched.push(pattern);
+        } else {
+          for (const filePath of dirMatches) {
+            results.add(path.normalize(filePath));
+          }
+        }
+        continue;
+      }
     } catch {
     }
     const globPattern = pattern.includes("\\") ? pattern.replace(/\\/g, "/") : pattern;
@@ -304,7 +327,7 @@ import path3 from "node:path";
 var RESULT_INDEX_FILENAME = "index.jsonl";
 var RESULT_RUNS_DIRNAME = "runs";
 function createRunDirName(timestamp = /* @__PURE__ */ new Date()) {
-  return `eval_${timestamp.toISOString().replace(/[:.]/g, "-")}`;
+  return timestamp.toISOString().replace(/[:.]/g, "-");
 }
 function buildDefaultRunDir(cwd) {
   return path3.join(cwd, ".agentv", "results", RESULT_RUNS_DIRNAME, createRunDirName());
@@ -562,20 +585,16 @@ function safeArtifactPathSegment(value, fallback) {
 function safeTestId(testId) {
   return safeArtifactPathSegment(testId, "unknown");
 }
-function safeTargetId(target) {
-  return safeArtifactPathSegment(target, "default");
-}
-function getEvalSet(result) {
-  const record = result;
-  return result.eval_set ?? record.evalSet;
+function getDataset(result) {
+  return result.dataset;
 }
 function buildArtifactSubdir(result) {
   const segments = [];
-  const evalSet = getEvalSet(result);
+  const evalSet = getDataset(result);
   if (evalSet) {
     segments.push(safeArtifactPathSegment(evalSet, "default"));
   }
-  segments.push(safeTestId(result.testId), safeTargetId(result.target));
+  segments.push(safeTestId(result.testId));
   return path4.posix.join(...segments);
 }
 function formatOutputMarkdown(output) {
@@ -598,7 +617,8 @@ function buildResultIndexArtifact(result) {
   return {
     timestamp: result.timestamp,
     test_id: result.testId ?? "unknown",
-    eval_set: getEvalSet(result),
+    dataset: getDataset(result),
+    category: result.category,
     conversation_id: result.conversationId,
     score: result.score,
     target: result.target ?? "unknown",
@@ -1667,7 +1687,7 @@ var JunitWriter = class _JunitWriter {
     this.closed = true;
     const grouped = /* @__PURE__ */ new Map();
     for (const result of this.results) {
-      const suite = result.eval_set ?? "default";
+      const suite = result.dataset ?? "default";
       const existing = grouped.get(suite);
       if (existing) {
         existing.push(result);
@@ -1995,7 +2015,8 @@ function hydrateManifestRecord(baseDir, record) {
   return {
     timestamp: record.timestamp,
     testId,
-    eval_set: record.eval_set,
+    dataset: record.dataset,
+    category: record.category,
     target: record.target,
     score: record.score,
     executionStatus: record.execution_status,
@@ -2053,6 +2074,7 @@ function loadLightweightResults(sourceFile) {
     return parseResultManifest(content).map((record) => ({
       testId: record.test_id ?? record.eval_id ?? "unknown",
       target: record.target,
+      experiment: record.experiment,
       score: record.score,
       scores: record.scores,
       executionStatus: record.execution_status,
@@ -3020,7 +3042,7 @@ function validateContentForRoleMarkers(content, location, filePath, errors) {
 function isObject2(value) {
   return typeof value === "object" && value !== null && !Array.isArray(value);
 }
-var COMMON_SETTINGS = /* @__PURE__ */ new Set(["provider_batching", "providerBatching"]);
+var COMMON_SETTINGS = new Set(COMMON_TARGET_SETTINGS);
 var RETRY_SETTINGS = /* @__PURE__ */ new Set([
   "max_retries",
   "maxRetries",
@@ -4140,9 +4162,12 @@ async function prepareFileMetadata(params) {
     repoRoot,
     verbose: options.verbose
   });
+  const relativePath = path15.relative(cwd, testFilePath);
+  const category = deriveCategory(relativePath);
   const suite = await loadTestSuite(testFilePath, repoRoot, {
     verbose: options.verbose,
-    filter: options.filter
+    filter: options.filter,
+    category
   });
   const filteredIds = suite.tests.map((value) => value.id);
   const cliTargets = options.cliTargets;
@@ -4421,7 +4446,7 @@ async function runEvalCommand(input) {
   const useFileExport = !!options.otelFile;
   if (options.exportOtel || useFileExport) {
     try {
-      const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-5EEXTTC3.js");
+      const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-3Z22B6SU.js");
       let endpoint = process.env.OTEL_EXPORTER_OTLP_ENDPOINT;
       let headers = {};
       if (options.otelBackend) {
@@ -4794,6 +4819,7 @@ export {
   resolveExistingRunPrimaryPath,
   resolveWorkspaceOrFilePath,
   writeArtifactsFromResults,
+  parseResultManifest,
   resolveResultSourcePath,
   loadManifestResults,
   loadLightweightResults,
@@ -4812,4 +4838,4 @@ export {
   selectTarget,
   runEvalCommand
 };
-//# sourceMappingURL=chunk-GUXXTOYK.js.map
+//# sourceMappingURL=chunk-OT2J474N.js.map