agentv 4.20.0 → 4.21.0

package/dist/{chunk-KJZ7PZCE.js → chunk-LPSUVXHV.js}

@@ -4,7 +4,11 @@ import {
   TARGET_FILE_CANDIDATES,
   c,
   detectFileType,
+  detectInstallScope,
+  detectPackageManager,
   discoverEvalFiles,
+  enforceRequiredVersion,
+  fetchLatestVersion,
   findRepoRoot,
   findRunById,
   formatCost,
@@ -26,6 +30,7 @@ import {
   padLeft,
   padRight,
   parseResultManifest,
+  performSelfUpdate,
   resolveEvalPaths,
   resolveResultSourcePath,
   resolveRunCacheFile,
@@ -39,7 +44,7 @@ import {
   validateFileReferences,
   validateTargetsFile,
   validateWorkspacePaths
-} from "./chunk-PHGEGHKR.js";
+} from "./chunk-GPGX24OI.js";
 import {
   RESULT_INDEX_FILENAME,
   aggregateRunDir,
@@ -47,12 +52,11 @@ import {
   resolveRunManifestPath,
   toSnakeCaseDeep,
   writeArtifactsFromResults
-} from "./chunk-ZNS74WKH.js";
+} from "./chunk-FZUNMXBT.js";
 import {
   DEFAULT_CATEGORY,
   addBenchmark,
   deriveCategory,
-  discoverBenchmarks,
   discoverClaudeSessions,
   discoverCodexSessions,
   getBenchmark,
@@ -66,7 +70,7 @@ import {
   toTranscriptJsonLines,
   transpileEvalYamlFile,
   trimBaselineResult
-} from "./chunk-36HXBYUY.js";
+} from "./chunk-27VT2KU2.js";
 import {
   DEFAULT_THRESHOLD,
   createBuiltinRegistry,
@@ -76,6 +80,7 @@ import {
   getWorkspacePoolRoot,
   isAgentSkillsFormat,
   listTargetNames,
+  loadConfig,
   loadTestSuite,
   normalizeLineEndings,
   parseAgentSkillsEvals,
@@ -94,7 +99,7 @@ import {
   runStartsWithAssertion,
   toCamelCaseDeep,
   toSnakeCaseDeep as toSnakeCaseDeep2
-} from "./chunk-LP4Y5D2Z.js";
+} from "./chunk-ERSBQAGK.js";
 import {
   __commonJS,
   __require,
@@ -3986,7 +3991,7 @@ var evalRunCommand = command({
   },
   handler: async (args) => {
     if (args.evalPaths.length === 0 && process.stdin.isTTY) {
-      const { launchInteractiveWizard } = await import("./interactive-GLRASSKM.js");
+      const { launchInteractiveWizard } = await import("./interactive-RLMRNXXD.js");
       await launchInteractiveWizard();
       return;
     }
@@ -9732,14 +9737,14 @@ Serving most recent: ${metas[0].path}
   }
   return metas[0].path;
 }
-function resolveDashboardMode(projectCount, options) {
+function resolveDashboardMode(benchmarkCount, options) {
   if (options.single === true) {
-    return { isMultiProject: false, showMultiWarning: options.multi === true };
+    return { isMultiBenchmark: false, showMultiWarning: options.multi === true };
   }
   if (options.multi === true) {
-    return { isMultiProject: true, showMultiWarning: true };
+    return { isMultiBenchmark: true, showMultiWarning: true };
   }
-  return { isMultiProject: projectCount > 1, showMultiWarning: false };
+  return { isMultiBenchmark: benchmarkCount > 1, showMultiWarning: false };
 }
 function feedbackPath(resultDir) {
   return path19.join(resultDir, "feedback.json");
@@ -10241,8 +10246,8 @@ function handleConfig(c4, { agentvDir, searchDir }, options) {
   return c4.json({
     ...loadStudioConfig(agentvDir),
     read_only: options?.readOnly === true,
-    project_name: path19.basename(searchDir),
-    multi_project_dashboard: options?.multiProjectDashboard === true
+    benchmark_name: path19.basename(searchDir),
+    multi_benchmark_dashboard: options?.multiBenchmarkDashboard === true
   });
 }
 function handleFeedbackRead(c4, { searchDir }) {
@@ -10302,7 +10307,7 @@ function createApp(results, resultDir, cwd, sourceFile, options) {
   function withBenchmark(c4, handler) {
     const benchmark = getBenchmark(c4.req.param("benchmarkId") ?? "");
     if (!benchmark || !existsSync14(benchmark.path)) {
-      return c4.json({ error: "Project not found" }, 404);
+      return c4.json({ error: "Benchmark not found" }, 404);
     }
     return handler(c4, {
       searchDir: benchmark.path,
@@ -10360,7 +10365,7 @@ function createApp(results, resultDir, cwd, sourceFile, options) {
         };
       })
     );
-    return c4.json({ projects: benchmarks });
+    return c4.json({ benchmarks });
   });
   app2.post("/api/benchmarks", async (c4) => {
     if (readOnly) {
@@ -10375,17 +10380,9 @@ function createApp(results, resultDir, cwd, sourceFile, options) {
       return c4.json({ error: err2.message }, 400);
     }
   });
-  app2.delete("/api/benchmarks/:benchmarkId", (c4) => {
-    if (readOnly) {
-      return c4.json({ error: "Studio is running in read-only mode" }, 403);
-    }
-    const removed = removeBenchmark(c4.req.param("benchmarkId") ?? "");
-    if (!removed) return c4.json({ error: "Project not found" }, 404);
-    return c4.json({ ok: true });
-  });
   app2.get("/api/benchmarks/:benchmarkId/summary", async (c4) => {
     const benchmark = getBenchmark(c4.req.param("benchmarkId") ?? "");
-    if (!benchmark) return c4.json({ error: "Project not found" }, 404);
+    if (!benchmark) return c4.json({ error: "Benchmark not found" }, 404);
     try {
       const { runs: metas } = await listMergedResultFiles(benchmark.path);
       const runCount = metas.length;
@@ -10400,21 +10397,7 @@ function createApp(results, resultDir, cwd, sourceFile, options) {
         last_run: lastRun
       });
     } catch {
-      return c4.json({ error: "Failed to read project" }, 500);
-    }
-  });
-  app2.post("/api/benchmarks/discover", async (c4) => {
-    if (readOnly) {
-      return c4.json({ error: "Studio is running in read-only mode" }, 403);
-    }
-    try {
-      const body = await c4.req.json();
-      if (!body.path) return c4.json({ error: "Missing path" }, 400);
-      const discovered = discoverBenchmarks(body.path);
-      const registered = discovered.map((p) => benchmarkEntryToWire(addBenchmark(p)));
-      return c4.json({ discovered: registered });
-    } catch (err2) {
-      return c4.json({ error: err2.message }, 400);
+      return c4.json({ error: "Failed to read benchmark" }, 500);
     }
   });
   app2.get("/api/benchmarks/all-runs", async (c4) => {
@@ -10446,8 +10429,8 @@ function createApp(results, resultDir, cwd, sourceFile, options) {
             source: m.source,
             ...target && { target },
             ...experiment && { experiment },
-            project_id: p.id,
-            project_name: p.name
+            benchmark_id: p.id,
+            benchmark_name: p.name
           });
         }
       } catch {
@@ -10456,11 +10439,19 @@ function createApp(results, resultDir, cwd, sourceFile, options) {
     allRuns.sort((a, b) => b.timestamp.localeCompare(a.timestamp));
     return c4.json({ runs: allRuns });
   });
+  app2.delete("/api/benchmarks/:benchmarkId", (c4) => {
+    if (readOnly) {
+      return c4.json({ error: "Studio is running in read-only mode" }, 403);
+    }
+    const removed = removeBenchmark(c4.req.param("benchmarkId") ?? "");
+    if (!removed) return c4.json({ error: "Benchmark not found" }, 404);
+    return c4.json({ ok: true });
+  });
   app2.get(
     "/api/config",
     (c4) => handleConfig(c4, defaultCtx, {
       readOnly,
-      multiProjectDashboard: options?.multiProjectDashboard
+      multiBenchmarkDashboard: options?.multiBenchmarkDashboard
     })
   );
   app2.get("/api/remote/status", async (c4) => c4.json(await getRemoteResultsStatus(searchDir)));
@@ -10563,7 +10554,7 @@ function createApp(results, resultDir, cwd, sourceFile, options) {
       c4,
       (ctx, dataCtx) => handleConfig(ctx, dataCtx, {
         readOnly,
-        multiProjectDashboard: options?.multiProjectDashboard
+        multiBenchmarkDashboard: options?.multiBenchmarkDashboard
       })
     )
   );
@@ -10715,39 +10706,34 @@ var resultsServeCommand = command({
     }),
     multi: flag({
       long: "multi",
-      description: "Launch in multi-project dashboard mode (deprecated; use auto-detect or --single)"
+      description: "Launch in multi-benchmark dashboard mode (deprecated; use auto-detect or --single)"
     }),
     single: flag({
       long: "single",
-      description: "Force single-project dashboard mode"
+      description: "Force single-benchmark dashboard mode"
     }),
     add: option({
       type: optional(string),
       long: "add",
-      description: "Register a project by path"
+      description: "Register a benchmark by path"
     }),
     remove: option({
       type: optional(string),
       long: "remove",
-      description: "Unregister a project by ID"
-    }),
-    discover: option({
-      type: optional(string),
-      long: "discover",
-      description: "Scan a directory tree for repos with .agentv/"
+      description: "Unregister a benchmark by ID"
     }),
     readOnly: flag({
       long: "read-only",
       description: "Disable write operations and launch Studio in read-only leaderboard mode"
     })
   },
-  handler: async ({ source, port, dir, multi, single, add, remove, discover, readOnly }) => {
+  handler: async ({ source, port, dir, multi, single, add, remove, readOnly }) => {
     const cwd = dir ?? process.cwd();
     const listenPort = port ?? (process.env.PORT ? Number(process.env.PORT) : 3117);
     if (add) {
       try {
         const entry = addBenchmark(add);
-        console.log(`Registered project: ${entry.name} (${entry.id}) at ${entry.path}`);
+        console.log(`Registered benchmark: ${entry.name} (${entry.id}) at ${entry.path}`);
       } catch (err2) {
         console.error(`Error: ${err2.message}`);
         process.exit(1);
@@ -10757,32 +10743,23 @@ var resultsServeCommand = command({
     if (remove) {
       const removed = removeBenchmark(remove);
       if (removed) {
-        console.log(`Unregistered project: ${remove}`);
+        console.log(`Unregistered benchmark: ${remove}`);
       } else {
-        console.error(`Project not found: ${remove}`);
+        console.error(`Benchmark not found: ${remove}`);
         process.exit(1);
       }
       return;
     }
-    if (discover) {
-      const discovered = discoverBenchmarks(discover);
-      if (discovered.length === 0) {
-        console.log(`No projects with .agentv/ found under ${discover}`);
-        return;
-      }
-      for (const p of discovered) {
-        const entry = addBenchmark(p);
-        console.log(`Registered: ${entry.name} (${entry.id}) at ${entry.path}`);
-      }
-      console.log(`
-Discovered ${discovered.length} project(s).`);
-      return;
+    const repoRoot = await findRepoRoot(cwd);
+    const yamlConfig = await loadConfig(path19.join(cwd, "_"), repoRoot);
+    if (yamlConfig?.required_version) {
+      await enforceRequiredVersion(yamlConfig.required_version);
     }
     const registry = loadBenchmarkRegistry();
-    const { isMultiProject, showMultiWarning } = resolveDashboardMode(registry.benchmarks.length, {
-      multi,
-      single
-    });
+    const { isMultiBenchmark, showMultiWarning } = resolveDashboardMode(
+      registry.benchmarks.length,
+      { multi, single }
+    );
     try {
       let results = [];
       let sourceFile;
@@ -10806,15 +10783,15 @@ Discovered ${discovered.length} project(s).`);
       const resultDir = sourceFile ? path19.dirname(path19.resolve(sourceFile)) : cwd;
       const app2 = createApp(results, resultDir, cwd, sourceFile, {
         readOnly,
-        multiProjectDashboard: isMultiProject
+        multiBenchmarkDashboard: isMultiBenchmark
       });
       if (showMultiWarning) {
         console.warn(
-          "Warning: --multi is deprecated. Studio now auto-detects multi-project mode when multiple projects are registered. Use --single to force the single-project view."
+          "Warning: --multi is deprecated. Studio now auto-detects multi-benchmark mode when multiple benchmarks are registered. Use --single to force the single-benchmark view."
         );
       }
-      if (isMultiProject) {
-        console.log(`Multi-project mode: ${registry.benchmarks.length} project(s) registered`);
+      if (isMultiBenchmark) {
+        console.log(`Multi-benchmark mode: ${registry.benchmarks.length} benchmark(s) registered`);
       } else if (results.length > 0 && sourceFile) {
         console.log(`Serving ${results.length} result(s) from ${sourceFile}`);
       } else {
@@ -10839,28 +10816,6 @@ Discovered ${discovered.length} project(s).`);
 });
 
 // src/commands/self/index.ts
-import { spawn as spawn2 } from "node:child_process";
-function detectPackageManagerFromPath(scriptPath) {
-  if (scriptPath.includes(".bun")) {
-    return "bun";
-  }
-  return "npm";
-}
-function detectPackageManager() {
-  return detectPackageManagerFromPath(process.argv[1] ?? "");
-}
-function runCommand(cmd, args) {
-  return new Promise((resolve3, reject) => {
-    const child = spawn2(cmd, args, { stdio: ["inherit", "pipe", "inherit"], shell: true });
-    let stdout = "";
-    child.stdout?.on("data", (data) => {
-      process.stdout.write(data);
-      stdout += data.toString();
-    });
-    child.on("error", reject);
-    child.on("close", (code) => resolve3({ exitCode: code ?? 1, stdout }));
-  });
-}
 var updateCommand = command({
   name: "update",
   description: "Update agentv to the latest version",
@@ -10883,38 +10838,29 @@ var updateCommand = command({
     }
     const currentVersion = package_default.version;
     console.log(`Current version: ${currentVersion}`);
-    console.log(`Updating agentv using ${pm}...
+    console.log("Checking for updates...");
+    const latestVersion = await fetchLatestVersion();
+    if (latestVersion && latestVersion === currentVersion) {
+      console.log(`Already up to date (${currentVersion}).`);
+      return;
+    }
+    if (latestVersion) {
+      console.log(`Update available: ${currentVersion} \u2192 ${latestVersion}`);
+    }
+    const scope = detectInstallScope();
+    const scopeLabel = scope === "local" ? "local project install" : "global install";
+    console.log(`Updating agentv using ${pm} (${scopeLabel})...
 `);
-    const args = pm === "npm" ? ["install", "-g", "agentv@latest"] : ["add", "-g", "agentv@latest"];
-    try {
-      const result = await runCommand(pm, args);
-      if (result.exitCode !== 0) {
-        console.error("\nUpdate failed.");
-        process.exit(1);
-      }
-      let newVersion;
-      try {
-        const versionResult = await runCommand("agentv", ["--version"]);
-        newVersion = versionResult.stdout.trim();
-      } catch {
-      }
-      if (newVersion) {
-        console.log(`
-Update complete: ${currentVersion} \u2192 ${newVersion}`);
-      } else {
-        console.log("\nUpdate complete.");
-      }
-    } catch (error) {
-      if (error instanceof Error) {
-        if (error.message.includes("ENOENT") || error.message.includes("not found")) {
-          const alternative = pm === "npm" ? "bun" : "npm";
-          console.error(`Error: ${pm} not found. Try using --${alternative} flag.`);
-        } else {
-          console.error(`Error: ${error.message}`);
-        }
-        process.exit(1);
-      }
-      throw error;
+    const result = await performSelfUpdate({ pm, currentVersion, scope });
+    if (!result.success) {
+      console.error("\nUpdate failed.");
+      process.exit(1);
+    }
+    if (result.newVersion) {
+      console.log(`
+Update complete: ${currentVersion} \u2192 ${result.newVersion}`);
+    } else {
+      console.log("\nUpdate complete.");
     }
   }
 });
@@ -11847,7 +11793,7 @@ var workspaceCommand = subcommands({
 });
 
 // src/update-check.ts
-import { spawn as spawn3 } from "node:child_process";
+import { spawn as spawn2 } from "node:child_process";
 import { readFile as readFile8 } from "node:fs/promises";
 import { join as join5 } from "node:path";
 var CHECK_INTERVAL_MS = 24 * 60 * 60 * 1e3;
@@ -11912,7 +11858,7 @@ function backgroundUpdateCheck() {
     }).on('error', () => process.exit()).on('timeout', function() { this.destroy(); process.exit(); });
   `;
   try {
-    const child = spawn3(process.execPath, ["-e", script], {
+    const child = spawn2(process.execPath, ["-e", script], {
       detached: true,
       stdio: "ignore",
       windowsHide: true
@@ -12016,4 +11962,4 @@ export {
   preprocessArgv,
   runCli
 };
-//# sourceMappingURL=chunk-KJZ7PZCE.js.map
+//# sourceMappingURL=chunk-LPSUVXHV.js.map