fix: SVG className crash + viewport expansion + test suites

autoresearch/browse-tasks.json

@@ -88,7 +88,8 @@
     "name": "extract-npm-description",
     "steps": [
       "opencli operate open https://www.npmjs.com/package/express",
-      "opencli operate eval \"document.querySelector('p[class*=description], [data-testid=package-description], #readme p')?.textContent?.trim()\""
+      "opencli operate wait time 2",
+      "opencli operate eval \"(function(){var ps=document.querySelectorAll('p');for(var i=0;i<ps.length;i++){var t=ps[i].textContent.trim();if(t.length>10&&t.length<200)return t;}return '';})()\""
     ],
     "judge": {
       "type": "nonEmpty"
@@ -294,7 +295,7 @@
       "opencli operate open https://example.com",
       "opencli operate eval \"document.querySelector('a')?.click();'clicked'\"",
       "opencli operate wait time 2",
-      "opencli operate eval \"document.title\""
+      "opencli operate eval \"document.title + ' ' + location.href\""
     ],
     "judge": {
       "type": "contains",
@@ -580,7 +581,8 @@
     "name": "bench-imdb-matrix",
     "steps": [
       "opencli operate open https://www.imdb.com/title/tt0133093/",
-      "opencli operate eval \"JSON.stringify({title:document.querySelector('h1 span, [data-testid=hero__pageTitle] span')?.textContent,year:document.querySelector('a[href*=releaseinfo], [data-testid=hero-title-block__metadata] a')?.textContent})\""
+      "opencli operate wait time 3",
+      "opencli operate eval \"(function(){var title=document.querySelector('h1')?.textContent?.trim()||'';var year='';var links=document.querySelectorAll('a');for(var i=0;i<links.length;i++){if(links[i].textContent.trim()==='1999'){year='1999';break;}}var rating=document.querySelector('[data-testid=hero-rating-bar__aggregate-rating__score] span, .sc-bde20123-1')?.textContent?.trim()||'';return JSON.stringify({title:title,year:year,rating:rating});})()\""
     ],
     "judge": {
       "type": "contains",

autoresearch/eval-all.ts

@@ -0,0 +1,127 @@
+#!/usr/bin/env npx tsx
+/**
+ * Combined Test Suite Runner — runs browse + V2EX + Zhihu tasks.
+ * Reports combined score for AutoResearch iteration.
+ *
+ * Usage:
+ *   npx tsx autoresearch/eval-all.ts              # Run all
+ *   npx tsx autoresearch/eval-all.ts --suite v2ex  # Run one suite
+ */
+
+import { execSync } from 'node:child_process';
+import { readFileSync, writeFileSync, mkdirSync, readdirSync } from 'node:fs';
+import { join, dirname } from 'node:path';
+import { fileURLToPath } from 'node:url';
+
+const __dirname = dirname(fileURLToPath(import.meta.url));
+const ROOT = join(__dirname, '..');
+const RESULTS_DIR = join(__dirname, 'results');
+
+interface SuiteResult {
+  name: string;
+  passed: number;
+  total: number;
+  failures: string[];
+  duration: number;
+}
+
+function runSuite(name: string, script: string): SuiteResult {
+  const start = Date.now();
+  try {
+    const output = execSync(`npx tsx ${script}`, {
+      cwd: ROOT,
+      timeout: 600_000,
+      encoding: 'utf-8',
+      env: process.env,
+      stdio: ['pipe', 'pipe', 'pipe'],
+    });
+
+    // Parse SCORE=X/Y from output
+    const scoreMatch = output.match(/SCORE=(\d+)\/(\d+)/);
+    const passed = scoreMatch ? parseInt(scoreMatch[1], 10) : 0;
+    const total = scoreMatch ? parseInt(scoreMatch[2], 10) : 0;
+
+    // Parse failures
+    const failures: string[] = [];
+    const failLines = output.match(/✗.*$/gm) || [];
+    for (const line of failLines) {
+      const m = line.match(/✗\s+(?:\[.*?\]\s+)?(\S+)/);
+      if (m) failures.push(m[1].replace(/:$/, ''));
+    }
+
+    return { name, passed, total, failures, duration: Date.now() - start };
+  } catch (err: any) {
+    const output = err.stdout ?? '';
+    const scoreMatch = output.match(/SCORE=(\d+)\/(\d+)/);
+    const passed = scoreMatch ? parseInt(scoreMatch[1], 10) : 0;
+    const total = scoreMatch ? parseInt(scoreMatch[2], 10) : 0;
+    const failures: string[] = [];
+    const failLines = output.match(/✗.*$/gm) || [];
+    for (const line of failLines) {
+      const m = line.match(/✗\s+(?:\[.*?\]\s+)?(\S+)/);
+      if (m) failures.push(m[1].replace(/:$/, ''));
+    }
+    return { name, passed, total, failures, duration: Date.now() - start };
+  }
+}
+
+function main() {
+  const args = process.argv.slice(2);
+  const singleSuite = args.includes('--suite') ? args[args.indexOf('--suite') + 1] : null;
+
+  const suites = [
+    { name: 'browse', script: 'autoresearch/eval-browse.ts' },
+    { name: 'v2ex', script: 'autoresearch/eval-v2ex.ts' },
+    { name: 'zhihu', script: 'autoresearch/eval-zhihu.ts' },
+  ].filter(s => !singleSuite || s.name === singleSuite);
+
+  console.log(`\n🔬 Combined AutoResearch — ${suites.length} suites\n`);
+
+  const results: SuiteResult[] = [];
+  for (const suite of suites) {
+    console.log(`  Running ${suite.name}...`);
+    const result = runSuite(suite.name, suite.script);
+    results.push(result);
+    const icon = result.passed === result.total ? '✓' : '✗';
+    console.log(`    ${icon} ${result.name}: ${result.passed}/${result.total} (${Math.round(result.duration / 1000)}s)`);
+    if (result.failures.length > 0) {
+      for (const f of result.failures.slice(0, 5)) {
+        console.log(`      ✗ ${f}`);
+      }
+    }
+  }
+
+  // Summary
+  const totalPassed = results.reduce((s, r) => s + r.passed, 0);
+  const totalTasks = results.reduce((s, r) => s + r.total, 0);
+  const totalDuration = results.reduce((s, r) => s + r.duration, 0);
+  const allFailures = results.flatMap(r => r.failures.map(f => `${r.name}:${f}`));
+
+  console.log(`\n${'━'.repeat(50)}`);
+  console.log(`  Combined: ${totalPassed}/${totalTasks}`);
+  for (const r of results) {
+    console.log(`    ${r.name}: ${r.passed}/${r.total}`);
+  }
+  console.log(`  Time: ${Math.round(totalDuration / 60000)}min`);
+  if (allFailures.length > 0) {
+    console.log(`\n  All failures:`);
+    for (const f of allFailures) console.log(`    ✗ ${f}`);
+  }
+
+  // Save result
+  mkdirSync(RESULTS_DIR, { recursive: true });
+  const existing = readdirSync(RESULTS_DIR).filter(f => f.startsWith('all-')).length;
+  const roundNum = String(existing + 1).padStart(3, '0');
+  const resultPath = join(RESULTS_DIR, `all-${roundNum}.json`);
+  writeFileSync(resultPath, JSON.stringify({
+    timestamp: new Date().toISOString(),
+    score: `${totalPassed}/${totalTasks}`,
+    suites: Object.fromEntries(results.map(r => [r.name, `${r.passed}/${r.total}`])),
+    failures: allFailures,
+    duration: `${Math.round(totalDuration / 60000)}min`,
+  }, null, 2), 'utf-8');
+  console.log(`\n  Results saved to: ${resultPath}`);
+  console.log(`\nSCORE=${totalPassed}/${totalTasks}`);
+}
+
+main();

autoresearch/eval-browse.ts

@@ -76,7 +76,7 @@ function runCommand(cmd: string): string {
       stdio: ['pipe', 'pipe', 'pipe'],
     }).trim();
   } catch (err: any) {
-    return err.stdout?.trim() ?? '';
+    return err.stdout?.trim() || err.stderr?.trim() || '';
   }
 }
 

autoresearch/eval-skill.ts

@@ -18,7 +18,6 @@ import { join, dirname } from 'node:path';
 import { fileURLToPath } from 'node:url';
 
 const __dirname = dirname(fileURLToPath(import.meta.url));
-const TASKS_FILE = join(__dirname, 'skill-tasks.yaml');
 const RESULTS_DIR = join(__dirname, 'results');
 const SKILL_PATH = join(__dirname, '..', 'skills', 'opencli-operate', 'SKILL.md');
 
@@ -160,13 +159,20 @@ Always close the browser with 'opencli operate close' when done.`;
 }
 
 function extractVerdict(text: string): { success: boolean; explanation: string } {
-  // Try to find {"success": ...} JSON in the text
-  const jsonMatches = text.match(/\{"success"\s*:\s*(true|false)\s*,\s*"explanation"\s*:\s*"([^"]*)"\s*\}/g);
-  if (jsonMatches) {
-    const last = jsonMatches[jsonMatches.length - 1];
-    try {
-      return JSON.parse(last);
-    } catch { /* fall through */ }
+  // Try to find and parse {"success": ...} JSON from the last occurrence
+  const idx = text.lastIndexOf('{"success"');
+  if (idx !== -1) {
+    // Find the matching closing brace (handle escaped quotes in explanation)
+    const sub = text.slice(idx);
+    let braceCount = 0;
+    let end = -1;
+    for (let i = 0; i < sub.length; i++) {
+      if (sub[i] === '{') braceCount++;
+      else if (sub[i] === '}') { braceCount--; if (braceCount === 0) { end = i + 1; break; } }
+    }
+    if (end > 0) {
+      try { return JSON.parse(sub.slice(0, end)); } catch { /* fall through */ }
+    }
   }
 
   // Fallback: check for success indicators in text