agentv 4.35.1 → 4.37.0-next.1

package/dist/{chunk-CRMGUVRZ.js → chunk-P4LSNFZR.js}

@@ -53,7 +53,7 @@ import {
   validateTargetsFile,
   validateWorkspacePaths,
   writeRunTags
-} from "./chunk-INOKS5LF.js";
+} from "./chunk-M7AMFWBZ.js";
 import {
   RESULT_INDEX_FILENAME,
   aggregateRunDir,
@@ -65,7 +65,7 @@ import {
   resolveRunManifestPath,
   toSnakeCaseDeep as toSnakeCaseDeep2,
   writeArtifactsFromResults
-} from "./chunk-KJGYL3M3.js";
+} from "./chunk-N6E5XFOM.js";
 import {
   DEFAULT_CATEGORY,
   addProject,
@@ -86,9 +86,10 @@ import {
   touchProject,
   transpileEvalYamlFile,
   trimBaselineResult
-} from "./chunk-KNF3AGCI.js";
+} from "./chunk-OYI35QFW.js";
 import {
   DEFAULT_THRESHOLD,
+  buildTraceFromMessages,
   createBuiltinRegistry,
   discoverCopilotSessions,
   executeScript,
@@ -116,7 +117,7 @@ import {
   runStartsWithAssertion,
   toCamelCaseDeep,
   toSnakeCaseDeep
-} from "./chunk-6QEIZ33V.js";
+} from "./chunk-RL4S2FBZ.js";
 import {
   __commonJS,
   __require,
@@ -4325,16 +4326,25 @@ var evalAssertCommand = command({
       );
       process.exit(1);
     }
+    const messages = [{ role: "assistant", content: resolvedOutput }];
+    const inputMessages = [{ role: "user", content: resolvedInput }];
+    const trace = buildTraceFromMessages({
+      input: inputMessages,
+      output: messages,
+      finalOutput: resolvedOutput
+    });
     const payload = JSON.stringify(
       {
-        output: [{ role: "assistant", content: resolvedOutput }],
-        input: [{ role: "user", content: resolvedInput }],
+        output: resolvedOutput,
+        answer: resolvedOutput,
+        messages,
+        input: inputMessages,
         question: resolvedInput,
         criteria: "",
         expected_output: [],
         reference_answer: "",
         input_files: [],
-        trace: null,
+        trace,
         token_usage: null,
         cost_usd: null,
         duration_ms: null,
@@ -4581,11 +4591,21 @@ var evalRunCommand = command({
       type: optional(string),
       long: "transcript",
       description: "Grade a pre-recorded transcript JSONL instead of invoking a live provider. Ignores targets."
+    }),
+    recordReplay: option({
+      type: optional(string),
+      long: "record-replay",
+      description: "Append live target outputs to a replay fixture JSONL file. Graders still run normally."
+    }),
+    recordReplayVariant: option({
+      type: optional(string),
+      long: "record-replay-variant",
+      description: "Optional variant key to store with --record-replay fixture rows."
     })
   },
   handler: async (args) => {
     if (args.evalPaths.length === 0 && process.stdin.isTTY) {
-      const { launchInteractiveWizard } = await import("./interactive-VYQ5SYMR.js");
+      const { launchInteractiveWizard } = await import("./interactive-CQELHITQ.js");
       await launchInteractiveWizard();
       return;
     }
@@ -4634,7 +4654,9 @@ var evalRunCommand = command({
       budgetUsd: args.budgetUsd,
       tag: args.tag,
       excludeTag: args.excludeTag,
-      transcript: args.transcript
+      transcript: args.transcript,
+      recordReplay: args.recordReplay,
+      recordReplayVariant: args.recordReplayVariant
     };
     const result = await runEvalCommand({ testFiles: resolvedPaths, rawOptions });
     if (result?.allExecutionErrors) {
@@ -6730,9 +6752,16 @@ async function runScore(results, evaluatorConfig, testIdFilter) {
   const scored = [];
   for (const raw of results) {
     if (testIdFilter && raw.test_id !== testIdFilter) continue;
-    const trace = toTraceSummary(raw);
     const candidate = extractCandidate(raw);
-    const output = raw.output;
+    const output = raw.trace?.messages ?? (Array.isArray(raw.output) ? raw.output : void 0);
+    const outputMessages = Array.isArray(output) ? toCamelCaseDeep(output) : void 0;
+    const trace = raw.trace && Array.isArray(raw.trace.messages) && Array.isArray(raw.trace.events) ? toCamelCaseDeep(raw.trace) : buildTraceFromMessages({
+      output: outputMessages,
+      finalOutput: candidate,
+      summary: toTraceSummary(raw),
+      target: raw.target,
+      testId: raw.test_id
+    });
     const evalContext = {
       evalCase: buildTestCase(raw),
       candidate,
@@ -6741,7 +6770,7 @@ async function runScore(results, evaluatorConfig, testIdFilter) {
       attempt: 1,
       promptInputs: { question: "" },
       now: /* @__PURE__ */ new Date(),
-      output: Array.isArray(output) ? output : void 0,
+      output: outputMessages,
       trace,
       tokenUsage: raw.token_usage ? toCamelCaseDeep(raw.token_usage) : void 0,
       costUsd: raw.cost_usd,
@@ -7102,7 +7131,7 @@ function renderScores(scores) {
   }).join(" | ");
 }
 function renderTree(result) {
-  const messages = result.output;
+  const messages = result.trace?.messages ?? (Array.isArray(result.output) ? result.output : void 0);
   const spans = getTraceSpans(result);
   if (!messages || messages.length === 0) {
     if (spans.length > 0) {
@@ -7759,13 +7788,22 @@ async function runCodeGraders(tasks, concurrency) {
     const { testId, resultsDir, responseText, inputData } = task;
     const graderName = graderConfig.name;
     const inputText = extractInputText(inputData.input);
+    const messages = [{ role: "assistant", content: responseText }];
+    const trace = buildTraceFromMessages({
+      input: inputData.input,
+      output: messages,
+      finalOutput: responseText,
+      testId
+    });
     const payload = JSON.stringify({
-      output: [{ role: "assistant", content: responseText }],
+      output: responseText,
+      answer: responseText,
+      messages,
       input: inputData.input,
       criteria: "",
       expected_output: [],
       input_files: inputData.input_files ?? [],
-      trace: null,
+      trace,
       token_usage: null,
       cost_usd: null,
       duration_ms: null,
@@ -10939,7 +10977,7 @@ function renderResultsReport(results, sourceFile, records, benchmarkEvalFile) {
     (result, index) => serializeReportResult(result, sourceFile, records[index], benchmarkEvalFile)
   );
   const dataJson = JSON.stringify(rows).replace(/<\//g, "<\\/");
-  return RESULTS_REPORT_TEMPLATE.replace("__DATA_PLACEHOLDER__", dataJson);
+  return RESULTS_REPORT_TEMPLATE.replace("__DATA_PLACEHOLDER__", () => dataJson);
 }
 async function writeResultsReport(source, outputPath, cwd) {
   const { sourceFile, results, records, benchmarkEvalFile } = await loadReportSource(source, cwd);
@@ -10999,7 +11037,7 @@ function formatInput(result) {
 }
 function formatOutput(result) {
   if (!result.output || result.output.length === 0) return "(no output)";
-  return result.output.map((msg) => String(msg.content ?? "")).join("\n");
+  return result.output;
 }
 function formatShow(result) {
   const usage = result.tokenUsage;
@@ -12075,6 +12113,16 @@ function inferLanguage(filePath) {
   };
   return langMap[ext] ?? "plaintext";
 }
+function inferRawContentType(filePath) {
+  const ext = path23.extname(filePath).toLowerCase();
+  if (ext === ".json") return "application/json; charset=utf-8";
+  if (ext === ".jsonl") return "text/plain; charset=utf-8";
+  if (ext === ".md") return "text/markdown; charset=utf-8";
+  return "text/plain; charset=utf-8";
+}
+function contentDispositionFilename(filePath) {
+  return path23.basename(filePath).replace(/["\\\r\n]/g, "_");
+}
 function stripHeavyFields(results) {
   return results.map((r) => {
     const { requests, trace, ...rest } = r;
@@ -12475,6 +12523,8 @@ async function handleEvalFiles(c4, { searchDir, projectId }) {
       record.input_path,
       record.output_path,
       record.response_path,
+      record.answer_path,
+      record.transcript_path,
       record.task_dir,
       record.eval_path,
       record.targets_path,
@@ -12502,7 +12552,13 @@ async function handleEvalFileContent(c4, { searchDir, projectId }) {
   if (!meta) return c4.json({ error: "Run not found" }, 404);
   const marker = "/files/";
   const markerIdx = c4.req.path.indexOf(marker);
-  const filePath = markerIdx >= 0 ? c4.req.path.slice(markerIdx + marker.length) : "";
+  const encodedFilePath = markerIdx >= 0 ? c4.req.path.slice(markerIdx + marker.length) : "";
+  let filePath = "";
+  try {
+    filePath = encodedFilePath ? decodeURIComponent(encodedFilePath) : "";
+  } catch {
+    return c4.json({ error: "Invalid file path encoding" }, 400);
+  }
   if (!filePath) return c4.json({ error: "No file path specified" }, 400);
   await ensureRunReadable(searchDir, meta, projectId);
   const baseDir = path23.dirname(meta.path);
@@ -12515,6 +12571,16 @@ async function handleEvalFileContent(c4, { searchDir, projectId }) {
   }
   try {
     const fileContent = readFileSync12(absolutePath, "utf8");
+    if (c4.req.query("raw") === "1" || c4.req.query("download") === "1") {
+      c4.header("Content-Type", inferRawContentType(absolutePath));
+      if (c4.req.query("download") === "1") {
+        c4.header(
+          "Content-Disposition",
+          `attachment; filename="${contentDispositionFilename(absolutePath)}"`
+        );
+      }
+      return c4.body(fileContent);
+    }
     const language = inferLanguage(absolutePath);
     return c4.json({ content: fileContent, language });
   } catch {
@@ -15426,4 +15492,4 @@ export {
   preprocessArgv,
   runCli
 };
-//# sourceMappingURL=chunk-CRMGUVRZ.js.map
+//# sourceMappingURL=chunk-P4LSNFZR.js.map