npm - agentv - Versions diffs - 4.31.3 → 4.31.4-next.1 - Mend

agentv 4.31.3 → 4.31.4-next.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (29) hide show

package/dist/{chunk-4POCDPC2.js → chunk-Y35UPYKH.js} RENAMED Viewed

@@ -46,7 +46,7 @@ import {
   validateFileReferences,
   validateTargetsFile,
   validateWorkspacePaths
-} from "./chunk-CO23UZAA.js";
+} from "./chunk-PDGEPSPO.js";
 import {
   RESULT_INDEX_FILENAME,
   aggregateRunDir,
@@ -54,7 +54,7 @@ import {
   resolveRunManifestPath,
   toSnakeCaseDeep as toSnakeCaseDeep2,
   writeArtifactsFromResults
-} from "./chunk-IJPWTVDU.js";
+} from "./chunk-3Y24WATS.js";
 import {
   DEFAULT_CATEGORY,
   addProject,
@@ -75,7 +75,7 @@ import {
   touchProject,
   transpileEvalYamlFile,
   trimBaselineResult
-} from "./chunk-LOYPSIE7.js";
+} from "./chunk-6Y3W7HJP.js";
 import {
   DEFAULT_THRESHOLD,
   createBuiltinRegistry,
@@ -105,7 +105,7 @@ import {
   runStartsWithAssertion,
   toCamelCaseDeep,
   toSnakeCaseDeep
-} from "./chunk-377GONL7.js";
+} from "./chunk-MIEYJ3YP.js";
 import {
   __commonJS,
   __require,
@@ -4055,7 +4055,7 @@ var evalRunCommand = command({
   },
   handler: async (args) => {
     if (args.evalPaths.length === 0 && process.stdin.isTTY) {
-      const { launchInteractiveWizard } = await import("./interactive-QBLQNYDH.js");
+      const { launchInteractiveWizard } = await import("./interactive-RCBFUZUL.js");
       await launchInteractiveWizard();
       return;
     }
@@ -8995,7 +8995,7 @@ var RESULTS_REPORT_TEMPLATE = `<!DOCTYPE html>
       <div class="header-title-wrap">
         <div class="eyebrow">AgentV static export</div>
         <h1 class="header-title">Evaluation Report</h1>
-        <p class="header-subtitle">Studio-themed HTML generated from an existing AgentV results workspace.</p>
+        <p class="header-subtitle">Dashboard-themed HTML generated from an existing AgentV results workspace.</p>
       </div>
       <div class="header-meta" id="header-meta"></div>
     </header>
@@ -10344,7 +10344,7 @@ function generateRunId() {
   const pad = (n, w = 2) => String(n).padStart(w, "0");
   const ts = `${now.getFullYear()}${pad(now.getMonth() + 1)}${pad(now.getDate())}-${pad(now.getHours())}${pad(now.getMinutes())}${pad(now.getSeconds())}`;
   const rand = Math.random().toString(36).slice(2, 6);
-  return `studio-${ts}-${rand}`;
+  return `dashboard-${ts}-${rand}`;
 }
 function pruneFinishedRuns() {
   const finished = [...activeRuns.entries()].filter(([, r]) => r.status === "finished" || r.status === "failed").sort((a, b) => (b[1].finishedAt ?? "").localeCompare(a[1].finishedAt ?? ""));
@@ -10521,7 +10521,7 @@ function registerEvalRoutes(app2, getCwd, options) {
   });
   app2.post("/api/eval/run", async (c4) => {
     if (readOnly) {
-      return c4.json({ error: "Studio is running in read-only mode" }, 403);
+      return c4.json({ error: "Dashboard is running in read-only mode" }, 403);
     }
     const cwd = getCwd(c4);
     let body;
@@ -10620,7 +10620,7 @@ Process error: ${err2.message}
   });
   app2.post("/api/eval/run/:id/stop", (c4) => {
     if (readOnly) {
-      return c4.json({ error: "Studio is running in read-only mode" }, 403);
+      return c4.json({ error: "Dashboard is running in read-only mode" }, 403);
     }
     const id = c4.req.param("id");
     const run2 = activeRuns.get(id ?? "");
@@ -10699,7 +10699,7 @@ Process error: ${err2.message}
   });
   app2.post("/api/projects/:projectId/eval/run", async (c4) => {
     if (readOnly) {
-      return c4.json({ error: "Studio is running in read-only mode" }, 403);
+      return c4.json({ error: "Dashboard is running in read-only mode" }, 403);
     }
     const cwd = getCwd(c4);
     let body;
@@ -10786,7 +10786,7 @@ Process error: ${err2.message}
   });
   app2.post("/api/projects/:projectId/eval/run/:id/stop", (c4) => {
     if (readOnly) {
-      return c4.json({ error: "Studio is running in read-only mode" }, 403);
+      return c4.json({ error: "Dashboard is running in read-only mode" }, 403);
     }
     const id = c4.req.param("id");
     const run2 = activeRuns.get(id ?? "");
@@ -10935,22 +10935,24 @@ function loadStudioConfig(agentvDir) {
   if (!parsed || typeof parsed !== "object") {
     return { ...DEFAULTS };
   }
-  const studio = parsed.studio;
-  let threshold = DEFAULTS.threshold;
-  if (studio && typeof studio === "object" && !Array.isArray(studio)) {
-    const studioObj = studio;
-    if (typeof studioObj.threshold === "number") {
-      threshold = studioObj.threshold;
-    } else if (typeof studioObj.pass_threshold === "number") {
-      threshold = studioObj.pass_threshold;
-    }
-  } else if (typeof parsed.pass_threshold === "number") {
-    threshold = parsed.pass_threshold;
-  }
+  const config = parsed;
+  const threshold = [
+    readThreshold(config.dashboard),
+    readThreshold(config.studio),
+    typeof config.pass_threshold === "number" ? config.pass_threshold : void 0,
+    DEFAULTS.threshold
+  ].find((value) => value !== void 0);
   return {
     threshold: Math.min(1, Math.max(0, threshold))
   };
 }
+function readThreshold(section) {
+  if (!section || typeof section !== "object" || Array.isArray(section)) return void 0;
+  const values = section;
+  if (typeof values.threshold === "number") return values.threshold;
+  if (typeof values.pass_threshold === "number") return values.pass_threshold;
+  return void 0;
+}
 function saveStudioConfig(agentvDir, config) {
   if (!existsSync14(agentvDir)) {
     mkdirSync4(agentvDir, { recursive: true });
@@ -10966,13 +10968,17 @@ function saveStudioConfig(agentvDir, config) {
   }
   const { pass_threshold: _, ...rest } = existing;
   existing = rest;
+  const existingDashboard = existing.dashboard;
   const existingStudio = existing.studio;
-  if (existingStudio && typeof existingStudio === "object" && !Array.isArray(existingStudio)) {
-    const { pass_threshold: __, ...studioRest } = existingStudio;
-    existing.studio = { ...studioRest, ...config };
-  } else {
-    existing.studio = { ...config };
-  }
+  const dashboardRest = existingDashboard && typeof existingDashboard === "object" && !Array.isArray(existingDashboard) ? (({ pass_threshold: _2, ...rest2 }) => rest2)(existingDashboard) : {};
+  const studioRest = existingStudio && typeof existingStudio === "object" && !Array.isArray(existingStudio) ? (({ threshold: _2, pass_threshold: __, ...rest2 }) => rest2)(
+    existingStudio
+  ) : {};
+  const { studio: _legacyStudio, ...withoutStudio } = existing;
+  existing = {
+    ...withoutStudio,
+    dashboard: { ...studioRest, ...dashboardRest, ...config }
+  };
   const yamlStr = stringifyYaml3(existing);
   writeFileSync5(configPath, yamlStr, "utf-8");
 }
@@ -11174,12 +11180,14 @@ async function handleRuns(c4, { searchDir, agentvDir }) {
       let target;
       let experiment = inferExperimentFromRunId(m.raw_filename);
       let passRate = m.passRate;
+      let avgScore = m.avgScore;
       try {
         const records = await loadLightweightResultsForMeta(searchDir, m);
         if (records.length > 0) {
           target = records[0].target;
           experiment = records[0].experiment ?? experiment;
           passRate = records.filter((r) => r.score >= passThreshold).length / records.length;
+          avgScore = records.reduce((sum, r) => sum + r.score, 0) / records.length;
         } else {
           target = getActiveRunTarget(m.path);
         }
@@ -11194,7 +11202,7 @@ async function handleRuns(c4, { searchDir, agentvDir }) {
         timestamp: m.timestamp,
         test_count: m.testCount,
         pass_rate: passRate,
-        avg_score: m.avgScore,
+        avg_score: avgScore,
         size_bytes: m.sizeBytes,
         source: m.source,
         ...target && { target },
@@ -11710,7 +11718,7 @@ function createApp(results, resultDir, cwd, sourceFile, options) {
   }
   app2.post("/api/config", async (c4) => {
     if (readOnly) {
-      return c4.json({ error: "Studio is running in read-only mode" }, 403);
+      return c4.json({ error: "Dashboard is running in read-only mode" }, 403);
     }
     try {
       const body = await c4.req.json();
@@ -11763,7 +11771,7 @@ function createApp(results, resultDir, cwd, sourceFile, options) {
   });
   app2.post("/api/projects", async (c4) => {
     if (readOnly) {
-      return c4.json({ error: "Studio is running in read-only mode" }, 403);
+      return c4.json({ error: "Dashboard is running in read-only mode" }, 403);
     }
     try {
       const body = await c4.req.json();
@@ -11835,7 +11843,7 @@ function createApp(results, resultDir, cwd, sourceFile, options) {
   });
   app2.delete("/api/projects/:projectId", (c4) => {
     if (readOnly) {
-      return c4.json({ error: "Studio is running in read-only mode" }, 403);
+      return c4.json({ error: "Dashboard is running in read-only mode" }, 403);
     }
     const removed = removeProject(c4.req.param("projectId") ?? "");
     if (!removed) return c4.json({ error: "Project not found" }, 404);
@@ -11854,13 +11862,13 @@ function createApp(results, resultDir, cwd, sourceFile, options) {
   app2.get("/api/runs", (c4) => handleRuns(c4, defaultCtx));
   app2.put("/api/runs/:filename/tags", (c4) => {
     if (readOnly) {
-      return c4.json({ error: "Studio is running in read-only mode" }, 403);
+      return c4.json({ error: "Dashboard is running in read-only mode" }, 403);
     }
     return handleRunTagsPut(c4, defaultCtx);
   });
   app2.delete("/api/runs/:filename/tags", (c4) => {
     if (readOnly) {
-      return c4.json({ error: "Studio is running in read-only mode" }, 403);
+      return c4.json({ error: "Dashboard is running in read-only mode" }, 403);
     }
     return handleRunTagsDelete(c4, defaultCtx);
   });
@@ -11884,7 +11892,7 @@ function createApp(results, resultDir, cwd, sourceFile, options) {
   });
   app2.post("/api/feedback", async (c4) => {
     if (readOnly) {
-      return c4.json({ error: "Studio is running in read-only mode" }, 403);
+      return c4.json({ error: "Dashboard is running in read-only mode" }, 403);
     }
     let body;
     try {
@@ -11970,13 +11978,13 @@ function createApp(results, resultDir, cwd, sourceFile, options) {
   app2.get("/api/projects/:projectId/runs", (c4) => withProject(c4, handleRuns));
   app2.put("/api/projects/:projectId/runs/:filename/tags", (c4) => {
     if (readOnly) {
-      return c4.json({ error: "Studio is running in read-only mode" }, 403);
+      return c4.json({ error: "Dashboard is running in read-only mode" }, 403);
     }
     return withProject(c4, handleRunTagsPut);
   });
   app2.delete("/api/projects/:projectId/runs/:filename/tags", (c4) => {
     if (readOnly) {
-      return c4.json({ error: "Studio is running in read-only mode" }, 403);
+      return c4.json({ error: "Dashboard is running in read-only mode" }, 403);
     }
     return withProject(c4, handleRunTagsDelete);
   });
@@ -12021,7 +12029,9 @@ function createApp(results, resultDir, cwd, sourceFile, options) {
   );
   const studioDistPath = options?.studioDir ?? resolveStudioDistDir();
   if (!studioDistPath || !existsSync15(path21.join(studioDistPath, "index.html"))) {
-    throw new Error('Studio dist not found. Run "bun run build" in apps/studio/ to build the SPA.');
+    throw new Error(
+      'Dashboard dist not found. Run "bun run build" in apps/studio/ to build the SPA.'
+    );
   }
   app2.get("/", (c4) => {
     const indexPath = path21.join(studioDistPath, "index.html");
@@ -12080,8 +12090,8 @@ function resolveStudioDistDir() {
   return void 0;
 }
 var resultsServeCommand = command({
-  name: "studio",
-  description: "Start AgentV Studio \u2014 a local dashboard for reviewing evaluation results",
+  name: "dashboard",
+  description: "Start AgentV Dashboard \u2014 a local dashboard for reviewing evaluation results",
   args: {
     source: positional({
       type: optional(string),
@@ -12116,7 +12126,7 @@ var resultsServeCommand = command({
     }),
     readOnly: flag({
       long: "read-only",
-      description: "Disable write operations and launch Studio in read-only leaderboard mode"
+      description: "Disable write operations and launch Dashboard in read-only leaderboard mode"
     })
   },
   handler: async ({ source, port, dir, single, add, remove, readOnly }) => {
@@ -13572,6 +13582,7 @@ var app = subcommands({
   description: "AgentV CLI",
   version: package_default.version,
   cmds: {
+    dashboard: resultsServeCommand,
     eval: evalCommand,
     import: importCommand,
     compare: compareCommand,
@@ -13584,7 +13595,6 @@ var app = subcommands({
     self: selfCommand,
     skills: skillsCommand,
     serve: resultsServeCommand,
-    studio: resultsServeCommand,
     inspect: inspectCommand,
     trend: trendCommand,
     transpile: transpileCommand,
@@ -13600,6 +13610,7 @@ var TOP_LEVEL_COMMANDS = /* @__PURE__ */ new Set([
   "compare",
   "convert",
   "create",
+  "dashboard",
   "doctor",
   "init",
   "pipeline",
@@ -13614,8 +13625,14 @@ var TOP_LEVEL_COMMANDS = /* @__PURE__ */ new Set([
   "validate",
   "workspace"
 ]);
+function usesDeprecatedStudioAlias(argv) {
+  return argv[2] === "studio";
+}
 function preprocessArgv(argv) {
   const result = [...argv];
+  if (result[2] === "studio") {
+    result[2] = "dashboard";
+  }
   for (let i = 0; i < result.length; i++) {
     if (result[i] === "--eval-id") {
       result[i] = "--test-id";
@@ -13648,6 +13665,11 @@ ${updateNotice}
     updateNotice = n;
   });
   const processedArgv = preprocessArgv(argv);
+  if (usesDeprecatedStudioAlias(argv)) {
+    process.stderr.write(
+      "Warning: `agentv studio` is deprecated and will be removed in a future release. Use `agentv dashboard` instead.\n"
+    );
+  }
   const cwd = process.cwd();
   const repoRoot = await findRepoRoot(cwd);
   const sessionConfig = await loadConfig(path29.join(cwd, "_"), repoRoot);
@@ -13660,7 +13682,8 @@ ${updateNotice}
 export {
   app,
+  usesDeprecatedStudioAlias,
   preprocessArgv,
   runCli
 };
-//# sourceMappingURL=chunk-4POCDPC2.js.map
+//# sourceMappingURL=chunk-Y35UPYKH.js.map