npm - agentv - Versions diffs - 3.13.3 → 3.14.0 - Mend

agentv 3.13.3 → 3.14.0

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 (19) hide show

package/dist/{chunk-D3LNJUUB.js → chunk-3TBDSUYD.js} +659 -320
package/dist/chunk-3TBDSUYD.js.map +1 -0
package/dist/{chunk-TGCWIHBH.js → chunk-W6CGDNQR.js} +166 -18
package/dist/chunk-W6CGDNQR.js.map +1 -0
package/dist/{chunk-PACTPWEN.js → chunk-YYECEMUV.js} +8 -5
package/dist/chunk-YYECEMUV.js.map +1 -0
package/dist/cli.js +3 -3
package/dist/{dist-KPMR7RBT.js → dist-ZGLENPVH.js} +4 -2
package/dist/index.js +3 -3
package/dist/{interactive-OMJAMCQP.js → interactive-AI75XY3X.js} +3 -3
package/dist/templates/.agentv/config.yaml +4 -13
package/dist/templates/.agentv/targets.yaml +0 -16
package/package.json +5 -2
package/dist/chunk-D3LNJUUB.js.map +0 -1
package/dist/chunk-PACTPWEN.js.map +0 -1
package/dist/chunk-TGCWIHBH.js.map +0 -1
package/dist/templates/.agentv/.env.example +0 -23
/package/dist/{dist-KPMR7RBT.js.map → dist-ZGLENPVH.js.map} +0 -0
/package/dist/{interactive-OMJAMCQP.js.map → interactive-AI75XY3X.js.map} +0 -0

package/dist/{chunk-TGCWIHBH.js → chunk-W6CGDNQR.js} RENAMED Viewed

@@ -22,7 +22,7 @@ import {
   validateFileReferences,
   validateTargetsFile,
   writeArtifactsFromResults
-} from "./chunk-PACTPWEN.js";
+} from "./chunk-YYECEMUV.js";
 import {
   createBuiltinRegistry,
   executeScript,
@@ -39,7 +39,7 @@ import {
   toSnakeCaseDeep as toSnakeCaseDeep2,
   transpileEvalYamlFile,
   trimBaselineResult
-} from "./chunk-D3LNJUUB.js";
+} from "./chunk-3TBDSUYD.js";
 import {
   __commonJS,
   __esm,
@@ -4185,7 +4185,7 @@ var evalRunCommand = command({
   },
   handler: async (args) => {
     if (args.evalPaths.length === 0 && process.stdin.isTTY) {
-      const { launchInteractiveWizard } = await import("./interactive-OMJAMCQP.js");
+      const { launchInteractiveWizard } = await import("./interactive-AI75XY3X.js");
       await launchInteractiveWizard();
       return;
     }
@@ -5562,10 +5562,40 @@ function writeFeedback(cwd, data) {
   writeFileSync3(feedbackPath(cwd), `${JSON.stringify(data, null, 2)}
 `, "utf8");
 }
-function createApp(results, resultDir) {
+function createApp(results, resultDir, cwd, sourceFile) {
+  const searchDir = cwd ?? resultDir;
   const app2 = new Hono();
   app2.get("/", (c3) => {
-    return c3.html(generateServeHtml(results));
+    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
+      }))
+    });
+  });
+  app2.get("/api/runs/:filename", (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 lightResults = stripHeavyFields(loaded);
+      return c3.json({ results: lightResults, source: meta.filename });
+    } catch (err2) {
+      return c3.json({ error: "Failed to load run" }, 500);
+    }
   });
   app2.get("/api/feedback", (c3) => {
     const data = readFeedback(resultDir);
@@ -5611,11 +5641,8 @@ function createApp(results, resultDir) {
   });
   return app2;
 }
-function escapeHtml(s) {
-  return s.replace(/&/g, "&amp;").replace(/</g, "&lt;").replace(/>/g, "&gt;").replace(/"/g, "&quot;");
-}
-function generateServeHtml(results) {
-  const lightResults = results.map((r) => {
+function stripHeavyFields(results) {
+  return results.map((r) => {
     const { requests, trace, ...rest } = r;
     const toolCalls = trace?.toolCalls && Object.keys(trace.toolCalls).length > 0 ? trace.toolCalls : void 0;
     const graderDurationMs = (r.scores ?? []).reduce((sum, s) => sum + (s.durationMs ?? 0), 0);
@@ -5625,6 +5652,12 @@ function generateServeHtml(results) {
       ...graderDurationMs > 0 && { _graderDurationMs: graderDurationMs }
     };
   });
+}
+function escapeHtml(s) {
+  return s.replace(/&/g, "&amp;").replace(/</g, "&lt;").replace(/>/g, "&gt;").replace(/"/g, "&quot;");
+}
+function generateServeHtml(results, sourceFile) {
+  const lightResults = stripHeavyFields(results);
   const dataJson = JSON.stringify(lightResults).replace(/</g, "\\u003c").replace(/>/g, "\\u003e").replace(/\u2028/g, "\\u2028").replace(/\u2029/g, "\\u2029");
   return `<!DOCTYPE html>
 <html lang="en">
@@ -5642,6 +5675,11 @@ ${SERVE_STYLES}
             <h1 class="header-title">AgentV</h1>
             <span class="header-subtitle">Results Review</span>
         </div>
+        <div class="header-center">
+            <select id="run-picker" class="run-picker" title="Switch result file">
+                <option value="">Loading runs...</option>
+            </select>
+        </div>
         <div class="header-right">
             <span class="timestamp">${escapeHtml((/* @__PURE__ */ new Date()).toISOString())}</span>
         </div>
@@ -5653,6 +5691,7 @@ ${SERVE_STYLES}
     <main id="app"></main>
     <script>
     var DATA = ${dataJson};
+    var INITIAL_SOURCE = ${sourceFile ? JSON.stringify(path8.basename(sourceFile)).replace(/</g, "\\u003c").replace(/>/g, "\\u003e") : "null"};
 ${SERVE_SCRIPT}
     </script>
 </body>
@@ -5679,6 +5718,10 @@ body{font-family:var(--font);background:var(--bg);color:var(--text);line-height:
 .header-left{display:flex;align-items:baseline;gap:12px}
 .header-title{font-size:18px;font-weight:600}
 .header-subtitle{font-size:14px;color:var(--text-muted)}
+.header-center{flex:1;display:flex;justify-content:center;padding:0 16px}
+.run-picker{padding:6px 10px;border:1px solid var(--border);border-radius:var(--radius);font-size:13px;background:var(--surface);color:var(--text);font-family:var(--font);max-width:400px;width:100%;cursor:pointer}
+.run-picker:hover{border-color:var(--primary)}
+.run-picker:focus{outline:none;border-color:var(--primary);box-shadow:0 0 0 3px var(--primary-bg)}
 .timestamp{font-size:12px;color:var(--text-muted);font-family:var(--mono)}
 /* Tabs */
@@ -5778,6 +5821,11 @@ body{font-family:var(--font);background:var(--bg);color:var(--text);line-height:
 .tool-tag{display:inline-block;padding:2px 10px;font-size:12px;font-family:var(--mono);background:var(--primary-bg);color:var(--primary);border:1px solid var(--border);border-radius:12px}
 .empty-state{text-align:center;padding:48px 24px;color:var(--text-muted)}
 .empty-state h3{font-size:16px;margin-bottom:8px;color:var(--text)}
+.welcome-state{text-align:center;padding:80px 24px;color:var(--text-muted)}
+.welcome-state h2{font-size:24px;margin-bottom:12px;color:var(--text);font-weight:600}
+.welcome-state p{font-size:15px;margin-bottom:8px;max-width:500px;margin-left:auto;margin-right:auto}
+.welcome-state code{font-family:var(--mono);background:var(--surface);border:1px solid var(--border);border-radius:3px;padding:2px 6px;font-size:13px}
+.welcome-state .hint{margin-top:24px;font-size:13px;color:var(--text-muted)}
 /* Feedback */
 .feedback-section{margin-top:16px;padding-top:16px;border-top:1px solid var(--border-light)}
@@ -5935,7 +5983,15 @@ var SERVE_SCRIPT = `
   /* ---- render ---- */
   function render(){
-    if(DATA.length===0){app.innerHTML='<div class="empty-state"><h3>No results</h3><p>No evaluation results to display.</p></div>';return;}
+    if(DATA.length===0){
+      app.innerHTML='<div class="welcome-state">'
+        +'<h2>No results yet</h2>'
+        +'<p>Run an evaluation or mount a results directory to see results here.</p>'
+        +'<p><code>agentv eval &lt;eval-file&gt;</code></p>'
+        +'<p class="hint">The dashboard will automatically detect new result files.</p>'
+        +'</div>';
+      return;
+    }
     if(state.tab==="overview")renderOverview();else renderTests();
   }
@@ -6198,6 +6254,69 @@ var SERVE_SCRIPT = `
     return h;
   }
+  /* ---- run picker ---- */
+  var runPicker=document.getElementById("run-picker");
+  var knownRunFilenames=[];
+  function refreshRunList(){
+    fetch("/api/runs").then(function(r){return r.json();}).then(function(d){
+      if(!d||!d.runs)return;
+      var runs=d.runs;
+      var newFilenames=runs.map(function(r){return r.filename;});
+      /* Detect new runs that appeared since last poll */
+      if(knownRunFilenames.length>0){
+        var hasNew=newFilenames.some(function(f){return knownRunFilenames.indexOf(f)===-1;});
+        if(hasNew&&DATA.length===0){
+          /* Auto-load the first (most recent) run when starting from empty state */
+          loadRun(runs[0].filename);
+        }
+      }
+      knownRunFilenames=newFilenames;
+      /* Rebuild picker options */
+      var h='<option value="">Select a result file...</option>';
+      if(runs.length===0){
+        h='<option value="">No result files</option>';
+      }
+      for(var i=0;i<runs.length;i++){
+        var r=runs[i];
+        var label=r.filename+" ("+r.test_count+" tests, "+(r.pass_rate*100).toFixed(0)+"% pass)";
+        h+='<option value="'+esc(r.filename)+'">'+esc(label)+"</option>";
+      }
+      runPicker.innerHTML=h;
+      /* Pre-select the initially loaded run */
+      if(INITIAL_SOURCE&&runs.length>0){
+        runPicker.value=INITIAL_SOURCE;
+      }
+    }).catch(function(err){console.warn("Failed to refresh run list:",err);});
+  }
+  function loadRun(filename){
+    fetch("/api/runs/"+encodeURIComponent(filename)).then(function(r){return r.json();}).then(function(d){
+      if(d.error){console.error(d.error);return;}
+      DATA=d.results;
+      stats=computeStats(DATA);
+      tgtStats=computeTargets(DATA);
+      tgtNames=tgtStats.map(function(t){return t.target;});
+      state.expanded={};
+      feedbackCache={};
+      loadFeedback();
+      render();
+      /* Update picker selection */
+      runPicker.value=filename;
+    }).catch(function(err){console.error("Failed to load run:",err);});
+  }
+  runPicker.addEventListener("change",function(){
+    var val=runPicker.value;
+    if(val)loadRun(val);
+  });
+  /* Poll for new result files every 5 seconds */
+  refreshRunList();
+  setInterval(refreshRunList,5000);
   /* ---- init ---- */
   loadFeedback();
   render();
@@ -6216,7 +6335,7 @@ var resultsServeCommand = command({
       type: optional(number),
       long: "port",
       short: "p",
-      description: "Port to listen on (default: 3117)"
+      description: "Port to listen on (flag \u2192 PORT env var \u2192 3117)"
     }),
     dir: option({
       type: optional(string),
@@ -6227,14 +6346,43 @@ var resultsServeCommand = command({
   },
   handler: async ({ source, port, dir }) => {
     const cwd = dir ?? process.cwd();
-    const listenPort = port ?? 3117;
+    const listenPort = port ?? (process.env.PORT ? Number(process.env.PORT) : 3117);
     try {
-      const { results, sourceFile } = await loadResults(source, cwd);
-      const resultDir = path8.dirname(path8.resolve(sourceFile));
-      const app2 = createApp(results, resultDir);
-      console.log(`Serving ${results.length} result(s) from ${sourceFile}`);
+      let results = [];
+      let sourceFile;
+      if (source) {
+        const resolved = resolveResultSourcePath(source, cwd);
+        if (!existsSync4(resolved)) {
+          console.error(`Error: Source file not found: ${resolved}`);
+          process.exit(1);
+        }
+        sourceFile = resolved;
+        results = patchTestIds(loadManifestResults(resolved));
+      } else {
+        const cache = await loadRunCache(cwd);
+        const cachedFile = cache ? resolveRunCacheFile(cache) : "";
+        if (cachedFile && existsSync4(cachedFile)) {
+          sourceFile = cachedFile;
+          results = patchTestIds(loadManifestResults(cachedFile));
+        } else {
+          const metas = listResultFiles(cwd, 1);
+          if (metas.length > 0) {
+            sourceFile = metas[0].path;
+            results = patchTestIds(loadManifestResults(metas[0].path));
+          }
+        }
+      }
+      const resultDir = sourceFile ? path8.dirname(path8.resolve(sourceFile)) : cwd;
+      const app2 = createApp(results, resultDir, cwd, sourceFile);
+      if (results.length > 0 && sourceFile) {
+        console.log(`Serving ${results.length} result(s) from ${sourceFile}`);
+      } else {
+        console.log("No results found. Dashboard will show an empty state.");
+        console.log("Run an evaluation to see results: agentv eval <eval-file>");
+      }
       console.log(`Dashboard: http://localhost:${listenPort}`);
       console.log(`Feedback API: http://localhost:${listenPort}/api/feedback`);
+      console.log(`Result picker API: http://localhost:${listenPort}/api/runs`);
       console.log(`Feedback file: ${feedbackPath(resultDir)}`);
       console.log("Press Ctrl+C to stop");
       const { serve: startServer } = await import("@hono/node-server");
@@ -7766,4 +7914,4 @@ export {
   preprocessArgv,
   runCli
 };
-//# sourceMappingURL=chunk-TGCWIHBH.js.map
+//# sourceMappingURL=chunk-W6CGDNQR.js.map