agentv 3.14.5 → 4.0.0

package/dist/{chunk-UBLKP2F4.js → chunk-E3VSJJI4.js}

@@ -10,6 +10,7 @@ import {
   loadManifestResults,
   loadRunCache,
   package_default,
+  parseResultManifest,
   resolveEvalPaths,
   resolveExistingRunPrimaryPath,
   resolveResultSourcePath,
@@ -23,9 +24,11 @@ import {
   validateFileReferences,
   validateTargetsFile,
   writeArtifactsFromResults
-} from "./chunk-GUXXTOYK.js";
+} from "./chunk-OT2J474N.js";
 import {
+  DEFAULT_CATEGORY,
   createBuiltinRegistry,
+  deriveCategory,
   executeScript,
   getAgentvHome,
   getOutputFilenames,
@@ -40,7 +43,7 @@ import {
   toSnakeCaseDeep as toSnakeCaseDeep2,
   transpileEvalYamlFile,
   trimBaselineResult
-} from "./chunk-ELQEFMGO.js";
+} from "./chunk-OXBBWZOY.js";
 import {
   __commonJS,
   __esm,
@@ -3479,9 +3482,23 @@ var ASSERTION_TEMPLATES = {
   default: `#!/usr/bin/env bun
 import { defineAssertion } from '@agentv/eval';
 
-export default defineAssertion(({ outputText }) => {
+/** Extract text from the last message with the given role. */
+function getMessageText(messages: Array<{ role: string; content?: unknown }>, role = 'assistant'): string {
+  for (let i = messages.length - 1; i >= 0; i--) {
+    const msg = messages[i];
+    if (msg.role !== role) continue;
+    if (typeof msg.content === 'string') return msg.content;
+    if (Array.isArray(msg.content)) {
+      return msg.content.filter((b: any) => b.type === 'text').map((b: any) => b.text).join('\\n');
+    }
+  }
+  return '';
+}
+
+export default defineAssertion(({ output }) => {
   // TODO: Implement your assertion logic
-  const pass = outputText.length > 0;
+  const text = getMessageText(output ?? []);
+  const pass = text.length > 0;
   return {
     pass,
     reasoning: pass ? 'Output has content' : 'Output is empty',
@@ -3491,9 +3508,23 @@ export default defineAssertion(({ outputText }) => {
   score: `#!/usr/bin/env bun
 import { defineAssertion } from '@agentv/eval';
 
-export default defineAssertion(({ outputText }) => {
+/** Extract text from the last message with the given role. */
+function getMessageText(messages: Array<{ role: string; content?: unknown }>, role = 'assistant'): string {
+  for (let i = messages.length - 1; i >= 0; i--) {
+    const msg = messages[i];
+    if (msg.role !== role) continue;
+    if (typeof msg.content === 'string') return msg.content;
+    if (Array.isArray(msg.content)) {
+      return msg.content.filter((b: any) => b.type === 'text').map((b: any) => b.text).join('\\n');
+    }
+  }
+  return '';
+}
+
+export default defineAssertion(({ output }) => {
   // TODO: Implement your scoring logic (0.0 to 1.0)
-  const score = outputText.length > 0 ? 1.0 : 0.0;
+  const text = getMessageText(output ?? []);
+  const score = text.length > 0 ? 1.0 : 0.0;
   return {
     pass: score >= 0.5,
     score,
@@ -4186,7 +4217,7 @@ var evalRunCommand = command({
   },
   handler: async (args) => {
     if (args.evalPaths.length === 0 && process.stdin.isTTY) {
-      const { launchInteractiveWizard } = await import("./interactive-WUIEXGWM.js");
+      const { launchInteractiveWizard } = await import("./interactive-D5UTP72M.js");
       await launchInteractiveWizard();
       return;
     }
@@ -4421,6 +4452,9 @@ var evalBenchCommand = command({
     const manifest = JSON.parse(await readFile(join(exportDir, "manifest.json"), "utf8"));
     const testIds = manifest.test_ids;
     const targetName = manifest.target?.name ?? "unknown";
+    const evalSet = manifest.dataset ?? "";
+    const experiment = manifest.experiment;
+    const safeEvalSet = evalSet ? evalSet.replace(/[\/\\:*?"<>|]/g, "_") : "";
     let stdinData;
     if (llmScoresPath) {
       stdinData = await readFile(llmScoresPath, "utf8");
@@ -4431,7 +4465,9 @@ var evalBenchCommand = command({
     const indexLines = [];
     const allPassRates = [];
     for (const testId of testIds) {
-      const testDir = join(exportDir, testId);
+      const subpath = safeEvalSet ? [safeEvalSet, testId] : [testId];
+      const testDir = join(exportDir, ...subpath);
+      const artifactSubdir = subpath.join("/");
       const evaluators = [];
       const allAssertions = [];
       const codeResultsDir = join(testDir, "code_grader_results");
@@ -4527,13 +4563,15 @@ var evalBenchCommand = command({
         JSON.stringify({
           timestamp: manifest.timestamp,
           test_id: testId,
+          dataset: evalSet || void 0,
+          experiment: experiment || void 0,
           score: Math.round(weightedScore * 1e3) / 1e3,
           target: targetName,
           scores,
           execution_status: executionStatus,
-          grading_path: `${testId}/grading.json`,
-          timing_path: `${testId}/timing.json`,
-          response_path: hasResponse ? `${testId}/response.md` : null
+          grading_path: `${artifactSubdir}/grading.json`,
+          timing_path: `${artifactSubdir}/timing.json`,
+          response_path: hasResponse ? `${artifactSubdir}/response.md` : void 0
         })
       );
     }
@@ -4548,6 +4586,7 @@ var evalBenchCommand = command({
       metadata: {
         eval_file: manifest.eval_file,
         timestamp: manifest.timestamp,
+        experiment: experiment || void 0,
         targets: [targetName],
         tests_run: testIds
       },
@@ -4589,6 +4628,12 @@ function computeStats(values) {
 // src/commands/pipeline/grade.ts
 import { mkdir as mkdir2, readFile as readFile2, readdir as readdir2, writeFile as writeFile3 } from "node:fs/promises";
 import { join as join2 } from "node:path";
+function extractInputText(input) {
+  if (!input || input.length === 0) return "";
+  if (input.length === 1) return input[0].content;
+  return input.map((m) => `@[${m.role}]:
+${m.content}`).join("\n\n");
+}
 var evalGradeCommand = command({
   name: "grade",
   description: "Run code-grader assertions on responses in an export directory",
@@ -4603,10 +4648,13 @@ var evalGradeCommand = command({
     const manifestPath = join2(exportDir, "manifest.json");
     const manifest = JSON.parse(await readFile2(manifestPath, "utf8"));
     const testIds = manifest.test_ids;
+    const evalSet = manifest.dataset ?? "";
+    const safeEvalSet = evalSet ? evalSet.replace(/[\/\\:*?"<>|]/g, "_") : "";
     let totalGraders = 0;
     let totalPassed = 0;
     for (const testId of testIds) {
-      const testDir = join2(exportDir, testId);
+      const subpath = safeEvalSet ? [safeEvalSet, testId] : [testId];
+      const testDir = join2(exportDir, ...subpath);
       const codeGradersDir = join2(testDir, "code_graders");
       const resultsDir = join2(testDir, "code_grader_results");
       let graderFiles;
@@ -4622,14 +4670,13 @@ var evalGradeCommand = command({
       for (const graderFile of graderFiles) {
         const graderConfig = JSON.parse(await readFile2(join2(codeGradersDir, graderFile), "utf8"));
         const graderName = graderConfig.name;
+        const inputText = extractInputText(inputData.input);
         const payload = JSON.stringify({
           output: [{ role: "assistant", content: responseText }],
-          input: inputData.input_messages,
-          question: inputData.input_text,
+          input: inputData.input,
           criteria: "",
           expected_output: [],
-          reference_answer: "",
-          input_files: [],
+          input_files: inputData.input_files ?? [],
           trace: null,
           token_usage: null,
           cost_usd: null,
@@ -4639,8 +4686,8 @@ var evalGradeCommand = command({
           file_changes: null,
           workspace_path: null,
           config: graderConfig.config ?? null,
-          metadata: {},
-          input_text: inputData.input_text,
+          metadata: inputData.metadata ?? {},
+          input_text: inputText,
           output_text: responseText,
           expected_output_text: ""
         });
@@ -4698,10 +4745,10 @@ var evalGradeCommand = command({
 // src/commands/pipeline/input.ts
 import { readFile as readFile3 } from "node:fs/promises";
 import { mkdir as mkdir3, writeFile as writeFile4 } from "node:fs/promises";
-import { dirname, join as join3, resolve } from "node:path";
+import { dirname, join as join3, relative, resolve } from "node:path";
 var evalInputCommand = command({
   name: "input",
-  description: "Extract eval inputs, target commands, and grader prompts for agent-mode runs",
+  description: "Extract eval inputs, target commands, and grader prompts for subagent-mode runs",
   args: {
     evalPath: positional({
       type: string,
@@ -4711,15 +4758,21 @@ var evalInputCommand = command({
     out: option({
       type: optional(string),
       long: "out",
-      description: "Output directory for extracted inputs (default: .agentv/results/runs/eval_<timestamp>)"
+      description: "Output directory for extracted inputs (default: .agentv/results/runs/<timestamp>)"
+    }),
+    experiment: option({
+      type: optional(string),
+      long: "experiment",
+      description: "Experiment label (e.g. with_skills, without_skills)"
     })
   },
-  handler: async ({ evalPath, out }) => {
+  handler: async ({ evalPath, out, experiment }) => {
     const resolvedEvalPath = resolve(evalPath);
     const outDir = resolve(out ?? buildDefaultRunDir(process.cwd()));
     const repoRoot = await findRepoRoot(dirname(resolvedEvalPath));
     const evalDir = dirname(resolvedEvalPath);
-    const suite = await loadTestSuite(resolvedEvalPath, repoRoot);
+    const category = deriveCategory(relative(process.cwd(), resolvedEvalPath));
+    const suite = await loadTestSuite(resolvedEvalPath, repoRoot, { category });
     const tests = suite.tests;
     if (tests.length === 0) {
       console.error("No tests found in eval file.");
@@ -4728,6 +4781,7 @@ var evalInputCommand = command({
     let targetInfo = null;
     let targetName = "agent";
     let targetKind = "agent";
+    let subagentModeAllowed = true;
     try {
       const selection = await selectTarget({
         testFilePath: resolvedEvalPath,
@@ -4740,32 +4794,38 @@ var evalInputCommand = command({
         env: process.env
       });
       targetName = selection.targetName;
-      if (selection.resolvedTarget.kind === "cli") {
+      const resolved = selection.resolvedTarget;
+      subagentModeAllowed = resolved.subagentModeAllowed !== false;
+      if (resolved.kind === "cli") {
         targetKind = "cli";
-        const config = selection.resolvedTarget.config;
+        subagentModeAllowed = false;
+        const config = resolved.config;
         targetInfo = {
           kind: "cli",
           command: config.command,
           cwd: config.cwd ?? evalDir,
           timeoutMs: config.timeoutMs ?? 3e4
         };
+      } else {
+        targetKind = resolved.kind;
       }
     } catch {
     }
+    const evalSetName = suite.metadata?.name?.trim() ?? "";
+    const safeEvalSet = evalSetName ? evalSetName.replace(/[\/\\:*?"<>|]/g, "_") : "";
     const testIds = [];
     for (const test of tests) {
-      const testDir = join3(outDir, test.id);
+      const subpath = safeEvalSet ? [safeEvalSet, test.id] : [test.id];
+      const testDir = join3(outDir, ...subpath);
       await mkdir3(testDir, { recursive: true });
       testIds.push(test.id);
-      const inputText = test.question;
       const inputMessages = test.input.map((m) => ({
         role: m.role,
         content: typeof m.content === "string" ? m.content : m.content
       }));
       await writeJson(join3(testDir, "input.json"), {
-        input_text: inputText,
-        input_messages: inputMessages,
-        file_paths: test.file_paths,
+        input: inputMessages,
+        input_files: test.file_paths,
         metadata: test.metadata ?? {}
       });
       if (targetInfo) {
@@ -4793,10 +4853,13 @@ var evalInputCommand = command({
     }
     await writeJson(join3(outDir, "manifest.json"), {
       eval_file: resolvedEvalPath,
+      dataset: evalSetName || void 0,
+      experiment: experiment || void 0,
       timestamp: (/* @__PURE__ */ new Date()).toISOString(),
       target: {
         name: targetName,
-        kind: targetKind
+        kind: targetKind,
+        subagent_mode_allowed: subagentModeAllowed
       },
       test_ids: testIds
     });
@@ -4858,7 +4921,13 @@ import { execSync } from "node:child_process";
 import { existsSync as existsSync3, readFileSync as readFileSync4, unlinkSync } from "node:fs";
 import { mkdir as mkdir4, readFile as readFile4, readdir as readdir3, writeFile as writeFile5 } from "node:fs/promises";
 import { tmpdir } from "node:os";
-import { dirname as dirname2, join as join4, resolve as resolve2 } from "node:path";
+import { dirname as dirname2, join as join4, relative as relative2, resolve as resolve2 } from "node:path";
+function extractInputText2(input) {
+  if (!input || input.length === 0) return "";
+  if (input.length === 1) return input[0].content;
+  return input.map((m) => `@[${m.role}]:
+${m.content}`).join("\n\n");
+}
 function loadEnvFile(dir) {
   let current = resolve2(dir);
   while (true) {
@@ -4892,20 +4961,26 @@ var evalRunCommand2 = command({
     out: option({
       type: optional(string),
       long: "out",
-      description: "Output directory for results (default: .agentv/results/runs/eval_<timestamp>)"
+      description: "Output directory for results (default: .agentv/results/runs/<timestamp>)"
     }),
     workers: option({
       type: optional(number),
       long: "workers",
       description: "Parallel workers for target invocation (default: all tests)"
+    }),
+    experiment: option({
+      type: optional(string),
+      long: "experiment",
+      description: "Experiment label (e.g. with_skills, without_skills)"
     })
   },
-  handler: async ({ evalPath, out, workers }) => {
+  handler: async ({ evalPath, out, workers, experiment }) => {
     const resolvedEvalPath = resolve2(evalPath);
     const outDir = resolve2(out ?? buildDefaultRunDir(process.cwd()));
     const repoRoot = await findRepoRoot(dirname2(resolvedEvalPath));
     const evalDir = dirname2(resolvedEvalPath);
-    const suite = await loadTestSuite(resolvedEvalPath, repoRoot);
+    const category = deriveCategory(relative2(process.cwd(), resolvedEvalPath));
+    const suite = await loadTestSuite(resolvedEvalPath, repoRoot, { category });
     const tests = suite.tests;
     if (tests.length === 0) {
       console.error("No tests found in eval file.");
@@ -4938,20 +5013,21 @@ var evalRunCommand2 = command({
       }
     } catch {
     }
+    const evalSetName = suite.metadata?.name?.trim() ?? "";
+    const safeEvalSet = evalSetName ? evalSetName.replace(/[\/\\:*?"<>|]/g, "_") : "";
     const testIds = [];
     for (const test of tests) {
-      const testDir = join4(outDir, test.id);
+      const subpath = safeEvalSet ? [safeEvalSet, test.id] : [test.id];
+      const testDir = join4(outDir, ...subpath);
       await mkdir4(testDir, { recursive: true });
       testIds.push(test.id);
-      const inputText = test.question;
       const inputMessages = test.input.map((m) => ({
         role: m.role,
         content: typeof m.content === "string" ? m.content : m.content
       }));
       await writeJson2(join4(testDir, "input.json"), {
-        input_text: inputText,
-        input_messages: inputMessages,
-        file_paths: test.file_paths,
+        input: inputMessages,
+        input_files: test.file_paths,
         metadata: test.metadata ?? {}
       });
       if (targetInfo) {
@@ -4979,6 +5055,8 @@ var evalRunCommand2 = command({
     }
     await writeJson2(join4(outDir, "manifest.json"), {
       eval_file: resolvedEvalPath,
+      dataset: evalSetName || void 0,
+      experiment: experiment || void 0,
       timestamp: (/* @__PURE__ */ new Date()).toISOString(),
       target: { name: targetName, kind: targetKind },
       test_ids: testIds
@@ -4993,7 +5071,8 @@ var evalRunCommand2 = command({
       const maxWorkers = workers ?? testIds.length;
       console.log(`Invoking ${testIds.length} CLI target(s) (${maxWorkers} workers)...`);
       const invokeTarget = async (testId) => {
-        const testDir = join4(outDir, testId);
+        const subpath = safeEvalSet ? [safeEvalSet, testId] : [testId];
+        const testDir = join4(outDir, ...subpath);
         const invoke = JSON.parse(await readFile4(join4(testDir, "invoke.json"), "utf8"));
         if (invoke.kind !== "cli") return;
         const inputData = JSON.parse(await readFile4(join4(testDir, "input.json"), "utf8"));
@@ -5002,11 +5081,12 @@ var evalRunCommand2 = command({
         const timeoutMs = invoke.timeout_ms ?? 12e4;
         const promptFile = join4(tmpdir(), `agentv-prompt-${testId}-${Date.now()}.txt`);
         const outputFile = join4(tmpdir(), `agentv-output-${testId}-${Date.now()}.txt`);
-        await writeFile5(promptFile, inputData.input_text, "utf8");
+        const inputText = extractInputText2(inputData.input);
+        await writeFile5(promptFile, inputText, "utf8");
         let rendered = template;
         rendered = rendered.replace("{PROMPT_FILE}", promptFile);
         rendered = rendered.replace("{OUTPUT_FILE}", outputFile);
-        rendered = rendered.replace("{PROMPT}", inputData.input_text);
+        rendered = rendered.replace("{PROMPT}", inputText);
         const start = performance.now();
         try {
           execSync(rendered, {
@@ -5061,12 +5141,13 @@ var evalRunCommand2 = command({
       }
       await Promise.all(pending);
     } else {
-      console.log("Agent-as-target mode \u2014 skipping CLI invocation.");
+      console.log("Subagent-as-target mode \u2014 skipping CLI invocation.");
     }
     let totalGraders = 0;
     let totalPassed = 0;
     for (const testId of testIds) {
-      const testDir = join4(outDir, testId);
+      const subpath = safeEvalSet ? [safeEvalSet, testId] : [testId];
+      const testDir = join4(outDir, ...subpath);
       const codeGradersDir = join4(testDir, "code_graders");
       const resultsDir = join4(testDir, "code_grader_results");
       let graderFiles;
@@ -5082,14 +5163,13 @@ var evalRunCommand2 = command({
       for (const graderFile of graderFiles) {
         const graderConfig = JSON.parse(await readFile4(join4(codeGradersDir, graderFile), "utf8"));
         const graderName = graderConfig.name;
+        const inputText = extractInputText2(inputData.input);
         const payload = JSON.stringify({
           output: [{ role: "assistant", content: responseText }],
-          input: inputData.input_messages,
-          question: inputData.input_text,
+          input: inputData.input,
           criteria: "",
           expected_output: [],
-          reference_answer: "",
-          input_files: [],
+          input_files: inputData.input_files ?? [],
           trace: null,
           token_usage: null,
           cost_usd: null,
@@ -5099,8 +5179,8 @@ var evalRunCommand2 = command({
           file_changes: null,
           workspace_path: null,
           config: graderConfig.config ?? null,
-          metadata: {},
-          input_text: inputData.input_text,
+          metadata: inputData.metadata ?? {},
+          input_text: inputText,
           output_text: responseText,
           expected_output_text: ""
         });
@@ -5288,7 +5368,7 @@ function toRawResult(result) {
   return {
     timestamp: result.timestamp,
     test_id: result.testId,
-    eval_set: result.eval_set,
+    dataset: result.dataset,
     conversation_id: result.conversationId,
     score: result.score,
     assertions: result.assertions?.map((assertion) => ({
@@ -5411,7 +5491,7 @@ function loadOtlpTraceFile(filePath) {
     }
     return {
       test_id: stringAttr(rootAttrs.agentv_test_id) ?? stringAttr(rootAttrs.agentv_eval_id) ?? `trace-${index + 1}`,
-      eval_set: stringAttr(rootAttrs.agentv_eval_set),
+      dataset: stringAttr(rootAttrs.agentv_dataset),
       target: stringAttr(rootAttrs.agentv_target),
       score,
       error: root.status?.code === 2 ? root.status.message : void 0,
@@ -5684,9 +5764,11 @@ function patchTestIds(results) {
 // src/commands/results/export.ts
 function deriveOutputDir(cwd, sourceFile) {
   const parentDir = path7.basename(path7.dirname(sourceFile));
+  if (/^\d{4}-\d{2}-\d{2}T/.test(parentDir)) {
+    return path7.join(cwd, ".agentv", "results", "export", parentDir);
+  }
   if (parentDir.startsWith("eval_")) {
-    const dirName2 = parentDir.slice(5);
-    return path7.join(cwd, ".agentv", "results", "export", dirName2);
+    return path7.join(cwd, ".agentv", "results", "export", parentDir.slice(5));
   }
   const basename = path7.basename(sourceFile, ".jsonl");
   const dirName = basename.startsWith("eval_") ? basename.slice(5) : basename;
@@ -5939,10 +6021,12 @@ function checkDirectoryNaming(runDir) {
       message: `Directory is not under a 'runs/' parent (found '${parentName}/'). Expected: .agentv/results/runs/<run-dir>`
     });
   }
-  if (!/^eval_\d{4}-\d{2}-\d{2}T\d{2}-\d{2}-\d{2}-\d{3}Z$/.test(dirName)) {
+  const isNewFormat = /^\d{4}-\d{2}-\d{2}T\d{2}-\d{2}-\d{2}-\d{3}Z$/.test(dirName);
+  const isLegacyFormat = /^eval_\d{4}-\d{2}-\d{2}T\d{2}-\d{2}-\d{2}-\d{3}Z$/.test(dirName);
+  if (!isNewFormat && !isLegacyFormat) {
     diagnostics.push({
       severity: "warning",
-      message: `Directory name '${dirName}' does not match the expected pattern 'eval_<ISO-timestamp>'. Example: eval_2026-03-27T12-42-24-429Z`
+      message: `Directory name '${dirName}' does not match the expected pattern '<ISO-timestamp>'. Example: 2026-03-27T12-42-24-429Z`
     });
   }
   return diagnostics;
@@ -6151,8 +6235,9 @@ var resultsCommand = subcommands({
 });
 
 // src/commands/results/serve.ts
-import { existsSync as existsSync7, readFileSync as readFileSync8, writeFileSync as writeFileSync3 } from "node:fs";
+import { existsSync as existsSync7, readFileSync as readFileSync8, readdirSync as readdirSync3, statSync as statSync4, writeFileSync as writeFileSync3 } from "node:fs";
 import path9 from "node:path";
+import { fileURLToPath as fileURLToPath2 } from "node:url";
 import { Hono } from "hono";
 function feedbackPath(resultDir) {
   return path9.join(resultDir, "feedback.json");
@@ -6173,24 +6258,45 @@ function writeFeedback(cwd, data) {
   writeFileSync3(feedbackPath(cwd), `${JSON.stringify(data, null, 2)}
 `, "utf8");
 }
-function createApp(results, resultDir, cwd, sourceFile) {
+function createApp(results, resultDir, cwd, sourceFile, options) {
   const searchDir = cwd ?? resultDir;
   const app2 = new Hono();
+  const studioDistPath = options?.studioDir === false ? void 0 : options?.studioDir ?? resolveStudioDistDir();
   app2.get("/", (c3) => {
+    if (studioDistPath) {
+      const indexPath = path9.join(studioDistPath, "index.html");
+      if (existsSync7(indexPath)) {
+        return c3.html(readFileSync8(indexPath, "utf8"));
+      }
+    }
     return c3.html(generateServeHtml(results, sourceFile));
   });
   app2.get("/api/runs", (c3) => {
     const metas = listResultFiles(searchDir);
     return c3.json({
-      runs: metas.map((m) => ({
-        filename: m.filename,
-        path: m.path,
-        timestamp: m.timestamp,
-        test_count: m.testCount,
-        pass_rate: m.passRate,
-        avg_score: m.avgScore,
-        size_bytes: m.sizeBytes
-      }))
+      runs: metas.map((m) => {
+        let target;
+        let experiment;
+        try {
+          const records = loadLightweightResults(m.path);
+          if (records.length > 0) {
+            target = records[0].target;
+            experiment = records[0].experiment;
+          }
+        } catch {
+        }
+        return {
+          filename: m.filename,
+          path: m.path,
+          timestamp: m.timestamp,
+          test_count: m.testCount,
+          pass_rate: m.passRate,
+          avg_score: m.avgScore,
+          size_bytes: m.sizeBytes,
+          ...target && { target },
+          ...experiment && { experiment }
+        };
+      })
     });
   });
   app2.get("/api/runs/:filename", (c3) => {
@@ -6250,8 +6356,393 @@ function createApp(results, resultDir, cwd, sourceFile) {
     writeFeedback(resultDir, existing);
     return c3.json(existing);
   });
+  app2.get("/api/runs/:filename/datasets", (c3) => {
+    const filename = c3.req.param("filename");
+    const metas = listResultFiles(searchDir);
+    const meta = metas.find((m) => m.filename === filename);
+    if (!meta) {
+      return c3.json({ error: "Run not found" }, 404);
+    }
+    try {
+      const loaded = patchTestIds(loadManifestResults(meta.path));
+      const datasetMap = /* @__PURE__ */ new Map();
+      for (const r of loaded) {
+        const ds = r.dataset ?? r.target ?? "default";
+        const entry = datasetMap.get(ds) ?? { total: 0, passed: 0, scoreSum: 0 };
+        entry.total++;
+        if (r.score >= 1) entry.passed++;
+        entry.scoreSum += r.score;
+        datasetMap.set(ds, entry);
+      }
+      const datasets = [...datasetMap.entries()].map(([name, entry]) => ({
+        name,
+        total: entry.total,
+        passed: entry.passed,
+        failed: entry.total - entry.passed,
+        avg_score: entry.total > 0 ? entry.scoreSum / entry.total : 0
+      }));
+      return c3.json({ datasets });
+    } catch {
+      return c3.json({ error: "Failed to load datasets" }, 500);
+    }
+  });
+  app2.get("/api/runs/:filename/categories", (c3) => {
+    const filename = c3.req.param("filename");
+    const metas = listResultFiles(searchDir);
+    const meta = metas.find((m) => m.filename === filename);
+    if (!meta) {
+      return c3.json({ error: "Run not found" }, 404);
+    }
+    try {
+      const loaded = patchTestIds(loadManifestResults(meta.path));
+      const categoryMap = /* @__PURE__ */ new Map();
+      for (const r of loaded) {
+        const cat = r.category ?? DEFAULT_CATEGORY;
+        const entry = categoryMap.get(cat) ?? {
+          total: 0,
+          passed: 0,
+          scoreSum: 0,
+          datasets: /* @__PURE__ */ new Set()
+        };
+        entry.total++;
+        if (r.score >= 1) entry.passed++;
+        entry.scoreSum += r.score;
+        entry.datasets.add(r.dataset ?? r.target ?? "default");
+        categoryMap.set(cat, entry);
+      }
+      const categories = [...categoryMap.entries()].map(([name, entry]) => ({
+        name,
+        total: entry.total,
+        passed: entry.passed,
+        failed: entry.total - entry.passed,
+        avg_score: entry.total > 0 ? entry.scoreSum / entry.total : 0,
+        dataset_count: entry.datasets.size
+      }));
+      return c3.json({ categories });
+    } catch {
+      return c3.json({ error: "Failed to load categories" }, 500);
+    }
+  });
+  app2.get("/api/runs/:filename/categories/:category/datasets", (c3) => {
+    const filename = c3.req.param("filename");
+    const category = decodeURIComponent(c3.req.param("category"));
+    const metas = listResultFiles(searchDir);
+    const meta = metas.find((m) => m.filename === filename);
+    if (!meta) {
+      return c3.json({ error: "Run not found" }, 404);
+    }
+    try {
+      const loaded = patchTestIds(loadManifestResults(meta.path));
+      const filtered = loaded.filter((r) => (r.category ?? DEFAULT_CATEGORY) === category);
+      const datasetMap = /* @__PURE__ */ new Map();
+      for (const r of filtered) {
+        const ds = r.dataset ?? r.target ?? "default";
+        const entry = datasetMap.get(ds) ?? { total: 0, passed: 0, scoreSum: 0 };
+        entry.total++;
+        if (r.score >= 1) entry.passed++;
+        entry.scoreSum += r.score;
+        datasetMap.set(ds, entry);
+      }
+      const datasets = [...datasetMap.entries()].map(([name, entry]) => ({
+        name,
+        total: entry.total,
+        passed: entry.passed,
+        failed: entry.total - entry.passed,
+        avg_score: entry.total > 0 ? entry.scoreSum / entry.total : 0
+      }));
+      return c3.json({ datasets });
+    } catch {
+      return c3.json({ error: "Failed to load datasets" }, 500);
+    }
+  });
+  app2.get("/api/runs/:filename/evals/:evalId", (c3) => {
+    const filename = c3.req.param("filename");
+    const evalId = c3.req.param("evalId");
+    const metas = listResultFiles(searchDir);
+    const meta = metas.find((m) => m.filename === filename);
+    if (!meta) {
+      return c3.json({ error: "Run not found" }, 404);
+    }
+    try {
+      const loaded = patchTestIds(loadManifestResults(meta.path));
+      const result = loaded.find((r) => r.testId === evalId);
+      if (!result) {
+        return c3.json({ error: "Eval not found" }, 404);
+      }
+      return c3.json({ eval: result });
+    } catch {
+      return c3.json({ error: "Failed to load eval" }, 500);
+    }
+  });
+  app2.get("/api/index", (c3) => {
+    const metas = listResultFiles(searchDir);
+    const entries2 = metas.map((m) => {
+      let totalCostUsd = 0;
+      try {
+        const loaded = patchTestIds(loadManifestResults(m.path));
+        totalCostUsd = loaded.reduce((sum, r) => sum + (r.costUsd ?? 0), 0);
+      } catch {
+      }
+      return {
+        run_filename: m.filename,
+        test_count: m.testCount,
+        pass_rate: m.passRate,
+        avg_score: m.avgScore,
+        total_cost_usd: totalCostUsd,
+        timestamp: m.timestamp
+      };
+    });
+    return c3.json({ entries: entries2 });
+  });
+  function buildFileTree(dirPath, relativeTo) {
+    if (!existsSync7(dirPath) || !statSync4(dirPath).isDirectory()) {
+      return [];
+    }
+    const entries2 = readdirSync3(dirPath, { withFileTypes: true });
+    return entries2.sort((a, b) => {
+      if (a.isDirectory() !== b.isDirectory()) return a.isDirectory() ? -1 : 1;
+      return a.name.localeCompare(b.name);
+    }).map((entry) => {
+      const fullPath = path9.join(dirPath, entry.name);
+      const relPath = path9.relative(relativeTo, fullPath);
+      if (entry.isDirectory()) {
+        return {
+          name: entry.name,
+          path: relPath,
+          type: "dir",
+          children: buildFileTree(fullPath, relativeTo)
+        };
+      }
+      return { name: entry.name, path: relPath, type: "file" };
+    });
+  }
+  function inferLanguage(filePath) {
+    const ext = path9.extname(filePath).toLowerCase();
+    const langMap = {
+      ".json": "json",
+      ".jsonl": "json",
+      ".ts": "typescript",
+      ".tsx": "typescript",
+      ".js": "javascript",
+      ".jsx": "javascript",
+      ".md": "markdown",
+      ".yaml": "yaml",
+      ".yml": "yaml",
+      ".log": "plaintext",
+      ".txt": "plaintext",
+      ".py": "python",
+      ".sh": "shell",
+      ".bash": "shell",
+      ".css": "css",
+      ".html": "html",
+      ".xml": "xml",
+      ".svg": "xml",
+      ".toml": "toml",
+      ".diff": "diff",
+      ".patch": "diff"
+    };
+    return langMap[ext] ?? "plaintext";
+  }
+  app2.get("/api/runs/:filename/evals/:evalId/files", (c3) => {
+    const filename = c3.req.param("filename");
+    const evalId = c3.req.param("evalId");
+    const metas = listResultFiles(searchDir);
+    const meta = metas.find((m) => m.filename === filename);
+    if (!meta) {
+      return c3.json({ error: "Run not found" }, 404);
+    }
+    try {
+      const content = readFileSync8(meta.path, "utf8");
+      const records = parseResultManifest(content);
+      const record = records.find((r) => (r.test_id ?? r.eval_id) === evalId);
+      if (!record) {
+        return c3.json({ error: "Eval not found" }, 404);
+      }
+      const baseDir = path9.dirname(meta.path);
+      const knownPaths = [
+        record.grading_path,
+        record.timing_path,
+        record.input_path,
+        record.output_path,
+        record.response_path
+      ].filter((p) => !!p);
+      if (knownPaths.length === 0) {
+        return c3.json({ files: [] });
+      }
+      const artifactDirs = knownPaths.map((p) => path9.dirname(p));
+      let commonDir = artifactDirs[0];
+      for (const dir of artifactDirs) {
+        while (!dir.startsWith(commonDir)) {
+          commonDir = path9.dirname(commonDir);
+        }
+      }
+      const artifactAbsDir = path9.join(baseDir, commonDir);
+      const files = buildFileTree(artifactAbsDir, baseDir);
+      return c3.json({ files });
+    } catch {
+      return c3.json({ error: "Failed to load file tree" }, 500);
+    }
+  });
+  app2.get("/api/runs/:filename/evals/:evalId/files/*", (c3) => {
+    const filename = c3.req.param("filename");
+    const evalId = c3.req.param("evalId");
+    const metas = listResultFiles(searchDir);
+    const meta = metas.find((m) => m.filename === filename);
+    if (!meta) {
+      return c3.json({ error: "Run not found" }, 404);
+    }
+    const requestPath = c3.req.path;
+    const prefix = `/api/runs/${filename}/evals/${evalId}/files/`;
+    const filePath = requestPath.slice(prefix.length);
+    if (!filePath) {
+      return c3.json({ error: "No file path specified" }, 400);
+    }
+    const baseDir = path9.dirname(meta.path);
+    const absolutePath = path9.resolve(baseDir, filePath);
+    if (!absolutePath.startsWith(path9.resolve(baseDir) + path9.sep) && absolutePath !== path9.resolve(baseDir)) {
+      return c3.json({ error: "Path traversal not allowed" }, 403);
+    }
+    if (!existsSync7(absolutePath) || !statSync4(absolutePath).isFile()) {
+      return c3.json({ error: "File not found" }, 404);
+    }
+    try {
+      const fileContent = readFileSync8(absolutePath, "utf8");
+      const language = inferLanguage(absolutePath);
+      return c3.json({ content: fileContent, language });
+    } catch {
+      return c3.json({ error: "Failed to read file" }, 500);
+    }
+  });
+  app2.get("/api/experiments", (c3) => {
+    const metas = listResultFiles(searchDir);
+    const experimentMap = /* @__PURE__ */ new Map();
+    for (const m of metas) {
+      try {
+        const records = loadLightweightResults(m.path);
+        for (const r of records) {
+          const experiment = r.experiment ?? "default";
+          const entry = experimentMap.get(experiment) ?? {
+            targets: /* @__PURE__ */ new Set(),
+            runFilenames: /* @__PURE__ */ new Set(),
+            evalCount: 0,
+            passedCount: 0,
+            lastTimestamp: ""
+          };
+          entry.runFilenames.add(m.filename);
+          if (r.target) entry.targets.add(r.target);
+          entry.evalCount++;
+          if (r.score >= 1) entry.passedCount++;
+          if (r.timestamp && r.timestamp > entry.lastTimestamp) {
+            entry.lastTimestamp = r.timestamp;
+          }
+          experimentMap.set(experiment, entry);
+        }
+      } catch {
+      }
+    }
+    const experiments = [...experimentMap.entries()].map(([name, entry]) => ({
+      name,
+      run_count: entry.runFilenames.size,
+      target_count: entry.targets.size,
+      eval_count: entry.evalCount,
+      passed_count: entry.passedCount,
+      pass_rate: entry.evalCount > 0 ? entry.passedCount / entry.evalCount : 0,
+      last_run: entry.lastTimestamp || null
+    }));
+    return c3.json({ experiments });
+  });
+  app2.get("/api/targets", (c3) => {
+    const metas = listResultFiles(searchDir);
+    const targetMap = /* @__PURE__ */ new Map();
+    for (const m of metas) {
+      try {
+        const records = loadLightweightResults(m.path);
+        for (const r of records) {
+          const target = r.target ?? "default";
+          const entry = targetMap.get(target) ?? {
+            experiments: /* @__PURE__ */ new Set(),
+            runFilenames: /* @__PURE__ */ new Set(),
+            evalCount: 0,
+            passedCount: 0
+          };
+          entry.runFilenames.add(m.filename);
+          if (r.experiment) entry.experiments.add(r.experiment);
+          entry.evalCount++;
+          if (r.score >= 1) entry.passedCount++;
+          targetMap.set(target, entry);
+        }
+      } catch {
+      }
+    }
+    const targets = [...targetMap.entries()].map(([name, entry]) => ({
+      name,
+      run_count: entry.runFilenames.size,
+      experiment_count: entry.experiments.size,
+      eval_count: entry.evalCount,
+      passed_count: entry.passedCount,
+      pass_rate: entry.evalCount > 0 ? entry.passedCount / entry.evalCount : 0
+    }));
+    return c3.json({ targets });
+  });
+  if (studioDistPath) {
+    app2.get("/assets/*", (c3) => {
+      const assetPath = c3.req.path;
+      const filePath = path9.join(studioDistPath, assetPath);
+      if (!existsSync7(filePath)) {
+        return c3.notFound();
+      }
+      const content = readFileSync8(filePath);
+      const ext = path9.extname(filePath);
+      const mimeTypes = {
+        ".js": "application/javascript",
+        ".css": "text/css",
+        ".html": "text/html",
+        ".json": "application/json",
+        ".svg": "image/svg+xml",
+        ".png": "image/png",
+        ".woff2": "font/woff2",
+        ".woff": "font/woff"
+      };
+      const contentType = mimeTypes[ext] ?? "application/octet-stream";
+      return new Response(content, {
+        headers: {
+          "Content-Type": contentType,
+          "Cache-Control": "public, max-age=31536000, immutable"
+        }
+      });
+    });
+    app2.get("*", (c3) => {
+      if (c3.req.path.startsWith("/api/")) {
+        return c3.json({ error: "Not found" }, 404);
+      }
+      const indexPath = path9.join(studioDistPath, "index.html");
+      if (existsSync7(indexPath)) {
+        return c3.html(readFileSync8(indexPath, "utf8"));
+      }
+      return c3.notFound();
+    });
+  }
   return app2;
 }
+function resolveStudioDistDir() {
+  const currentDir = typeof __dirname !== "undefined" ? __dirname : path9.dirname(fileURLToPath2(import.meta.url));
+  const candidates = [
+    // From src/commands/results/ → sibling apps/studio/dist
+    path9.resolve(currentDir, "../../../../studio/dist"),
+    // From dist/ → sibling apps/studio/dist (monorepo dev)
+    path9.resolve(currentDir, "../../studio/dist"),
+    // Bundled inside CLI dist (published package)
+    path9.resolve(currentDir, "../studio"),
+    // From dist/ in monorepo root context
+    path9.resolve(currentDir, "../../../apps/studio/dist")
+  ];
+  for (const candidate of candidates) {
+    if (existsSync7(candidate) && existsSync7(path9.join(candidate, "index.html"))) {
+      return candidate;
+    }
+  }
+  return void 0;
+}
 function stripHeavyFields(results) {
   return results.map((r) => {
     const { requests, trace, ...rest } = r;
@@ -6934,8 +7425,8 @@ var SERVE_SCRIPT = `
 })();
 `;
 var resultsServeCommand = command({
-  name: "serve",
-  description: "Start a local HTTP server to review evaluation results",
+  name: "studio",
+  description: "Start AgentV Studio \u2014 a local dashboard for reviewing evaluation results",
   args: {
     source: positional({
       type: optional(string),
@@ -7572,7 +8063,7 @@ function formatResultDetail(result, index, tree) {
   }
   const scoreColor = result.score >= 0.9 ? c2.green : result.score >= 0.5 ? c2.yellow : c2.red;
   lines.push(
-    `${c2.bold}${testId}${c2.reset}  ${scoreColor}${formatScore(result.score)}${c2.reset}${result.target ? `  ${c2.dim}target: ${result.target}${c2.reset}` : ""}${result.eval_set ? `  ${c2.dim}eval-set: ${result.eval_set}${c2.reset}` : ""}`
+    `${c2.bold}${testId}${c2.reset}  ${scoreColor}${formatScore(result.score)}${c2.reset}${result.target ? `  ${c2.dim}target: ${result.target}${c2.reset}` : ""}${result.dataset ? `  ${c2.dim}dataset: ${result.dataset}${c2.reset}` : ""}`
   );
   if (result.error) {
     lines.push(`  ${c2.red}Error: ${result.error}${c2.reset}`);
@@ -7746,8 +8237,8 @@ function groupResults(results, groupBy2) {
       case "target":
         key = result.target ?? "unknown";
         break;
-      case "eval-set":
-        key = result.eval_set ?? "unknown";
+      case "dataset":
+        key = result.dataset ?? "unknown";
         break;
       case "test-id":
         key = result.test_id ?? result.eval_id ?? "unknown";
@@ -8460,7 +8951,9 @@ var app = subcommands({
     pipeline: pipelineCommand,
     results: resultsCommand,
     self: selfCommand,
+    studio: resultsServeCommand,
     serve: resultsServeCommand,
+    // hidden alias for backward compatibility
     trace: traceCommand,
     transpile: transpileCommand,
     trim: trimCommand,
@@ -8479,6 +8972,7 @@ var TOP_LEVEL_COMMANDS = /* @__PURE__ */ new Set([
   "results",
   "self",
   "serve",
+  "studio",
   "trace",
   "transpile",
   "trim",
@@ -8525,4 +9019,4 @@ export {
   preprocessArgv,
   runCli
 };
-//# sourceMappingURL=chunk-UBLKP2F4.js.map
+//# sourceMappingURL=chunk-E3VSJJI4.js.map