agentv 4.15.3 → 4.15.4

package/dist/{chunk-NXZLEIAG.js → chunk-TT36K77B.js}

@@ -42,12 +42,12 @@ import {
   syncResultsRepo,
   toCamelCaseDeep,
   toTranscriptJsonLines
-} from "./chunk-6X2UFVGI.js";
+} from "./chunk-K4QF5TNX.js";
 
 // package.json
 var package_default = {
   name: "agentv",
-  version: "4.15.3",
+  version: "4.15.4",
   description: "CLI entry point for AgentV",
   type: "module",
   repository: {
@@ -4923,6 +4923,8 @@ function normalizeOptions(rawOptions, config, yamlExecution) {
     retryErrors: normalizeString(rawOptions.retryErrors),
     workspaceMode,
     workspacePath,
+    // Precedence: CLI > YAML config > TS config
+    keepWorkspaces: normalizeBoolean(rawOptions.keepWorkspaces) || yamlExecution?.keep_workspaces === true || config?.execution?.keepWorkspaces === true,
     benchmarkJson: normalizeString(rawOptions.benchmarkJson),
     artifacts: normalizeString(rawOptions.artifacts),
     graderTarget: normalizeString(rawOptions.graderTarget),
@@ -5191,6 +5193,7 @@ async function runSingleEvalFile(params) {
     maxConcurrency: resolvedWorkers,
     workspaceMode: options.workspaceMode,
     workspacePath: options.workspacePath,
+    keepWorkspaces: options.keepWorkspaces,
     trials: trialsConfig,
     totalBudgetUsd,
     failOnError,
@@ -5352,7 +5355,7 @@ async function runEvalCommand(input) {
   const useFileExport = !!options.otelFile;
   if (options.exportOtel || useFileExport) {
     try {
-      const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-7ZYP7EBD.js");
+      const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-ZVTERRJS.js");
       let endpoint = process.env.OTEL_EXPORTER_OTLP_ENDPOINT;
       let headers = {};
       if (options.otelBackend) {
@@ -5548,7 +5551,7 @@ async function runEvalCommand(input) {
   const activeTestFiles = resolvedTestFiles.filter((f) => fileMetadata.has(f));
   let transcriptProviderFactory;
   if (options.transcript) {
-    const { TranscriptProvider } = await import("./dist-7ZYP7EBD.js");
+    const { TranscriptProvider } = await import("./dist-ZVTERRJS.js");
     const transcriptProvider = await TranscriptProvider.fromFile(options.transcript);
     const totalTests = [...fileMetadata.values()].reduce(
       (sum, meta) => sum + meta.testCases.length,
@@ -5705,15 +5708,18 @@ async function runEvalCommand(input) {
         `Export file(s) written: ${resolvedExportPaths.map((p) => path17.relative(cwd, p)).join(", ")}`
       );
     }
-    const failedWithWorkspaces = allResults.filter(
-      (r) => r.workspacePath && (r.error || r.score < 0.5)
-    );
-    if (failedWithWorkspaces.length > 0) {
-      console.log("\nWorkspaces preserved for debugging:");
-      for (const result of failedWithWorkspaces) {
-        console.log(`  ${result.testId}: ${result.workspacePath}`);
+    const resultsWithWorkspaces = allResults.filter((r) => r.workspacePath);
+    const preservedWorkspaces = options.keepWorkspaces ? resultsWithWorkspaces : resultsWithWorkspaces.filter((r) => r.error || r.score < 0.5);
+    if (preservedWorkspaces.length > 0) {
+      console.log("\nPreserved workspaces:");
+      for (const result of preservedWorkspaces) {
+        console.log(`  ${result.testId} -> ${result.workspacePath}`);
       }
     }
+    const usedWorkspaces = resultsWithWorkspaces.length > 0 || options.workspaceMode && options.workspaceMode !== "static";
+    if (!options.keepWorkspaces && usedWorkspaces) {
+      console.log("Use --keep-workspaces to preserve all workspaces for inspection.");
+    }
     if (allResults.length > 0) {
       console.log(`
 Results written to: ${outputPath}`);
@@ -5870,4 +5876,4 @@ export {
   getCategories,
   filterByCategory
 };
-//# sourceMappingURL=chunk-NXZLEIAG.js.map
+//# sourceMappingURL=chunk-TT36K77B.js.map