sillyspec 3.29.2 → 3.29.4

package/analyze-transcript.cjs

@@ -0,0 +1,38 @@
+// 转录命令级分析 v2：response.toolCalls 提取
+// 用法: node analyze-transcript.cjs <model-io.jsonl>
+const fs = require('fs');
+const file = process.argv[2];
+const lines = fs.readFileSync(file, 'utf8').split('\n').filter(Boolean);
+const calls = [];
+const slowReqs = [];
+for (const line of lines) {
+  let rec;
+  try { rec = JSON.parse(line); } catch { continue; }
+  const dur = rec.durationMs || 0;
+  const ts = rec.startedAt ? new Date(rec.startedAt).toTimeString().slice(0, 8) : '?';
+  if (dur > 30000) slowReqs.push({ ts, dur: Math.round(dur / 1000), n: (rec.response && rec.response.toolCalls || []).length });
+  for (const tc of (rec.response && rec.response.toolCalls) || []) {
+    const inp = tc.input || {};
+    let arg = '';
+    if (tc.name === 'Bash') arg = String(inp.command || '').replace(/\s+/g, ' ').slice(0, 130);
+    else if (tc.name === 'Read') arg = String(inp.file_path || '').split(/[\\/]/).slice(-2).join('/');
+    else if (tc.name === 'Edit' || tc.name === 'Write') arg = String(inp.file_path || '').split(/[\\/]/).slice(-2).join('/');
+    else if (tc.name === 'Grep') arg = String(inp.pattern || '').slice(0, 40) + ' @' + String(inp.path || '').split(/[\\/]/).slice(-1);
+    else if (tc.name === 'Agent') arg = 'dispatch: ' + String(inp.description || '').slice(0, 50);
+    else arg = JSON.stringify(inp).slice(0, 80);
+    calls.push({ ts, tool: tc.name, arg });
+  }
+}
+const byTool = {};
+const byArg = {};
+for (const c of calls) {
+  byTool[c.tool] = (byTool[c.tool] || 0) + 1;
+  byArg[c.tool + ' :: ' + c.arg] = (byArg[c.tool + ' :: ' + c.arg] || 0) + 1;
+}
+console.log('请求行:', lines.length, '| 工具调用:', calls.length, '| 分布:', JSON.stringify(byTool));
+console.log('\n== 重复 ≥2 次调用 top20 ==');
+for (const [k, n] of Object.entries(byArg).filter(([, n]) => n >= 2).sort((a, b) => b[1] - a[1]).slice(0, 20)) console.log(n + 'x |', k.slice(0, 160));
+console.log('\n== 慢请求 >30s ==');
+for (const s of slowReqs.sort((a, b) => b.dur - a.dur)) console.log(s.ts, '|', s.dur + 's', '| 本轮工具数:', s.n);
+console.log('\n== 全部 Bash 命令时间线 ==');
+for (const c of calls.filter((c) => c.tool === 'Bash')) console.log(c.ts, '|', c.arg);

package/asset-audit-2026-09-20/README.md

@@ -0,0 +1,42 @@
+# 资产审计数据（2026-09-20）
+
+机器生成，非手抄。每个数字都可复算：`node asset-audit.cjs` 重跑即得（脚本即方法）。
+
+覆盖两仓：
+- `sillyspec` = C:/Users/qinyi/IdeaProjects/sillyspec（本仓）
+- `platform` = C:/Users/qinyi/IdeaProjects/multi-agent-platform（SillyHub 平台侧）
+
+## 文件清单
+
+| 文件 | 内容 |
+|---|---|
+| summary.json | 两仓全部汇总数字（唯一事实源） |
+| sillyspec-archive.csv / platform-archive.csv | 归档变更逐条：名称、六件套文档齐缺 |
+| sillyspec-quicklog.csv / platform-quicklog.csv | quicklog 逐条：ql-ID、时间、状态、标题 |
+| sillyspec-fr.csv / platform-fr.csv | FR 需求索引逐条：来源变更、状态、最近确认 commit、commit 是否在 git 中存在 |
+| sillyspec-modules.csv / platform-modules.csv | 模块文档逐文件行数 |
+| sillyspec-gotchas.csv / platform-gotchas.csv | knowledge/ 下坑目/模式账本逐文件（gotchas 族+patterns/conventions/uncategorized）——**平台仓的真实坑目基础设施**（此前误按文件名找 troubleshooting.md 断言 0，已纠正） |
+| sillyspec-troubleshooting.csv / platform-troubleshooting.csv | 排坑条目逐条标题（仅按 troubleshooting.md 文件名采集；平台侧真身在 gotchas 族） |
+| sillyspec-knownissues.csv / platform-knownissues.csv | 已知问题账本逐条 |
+| asset-audit.cjs | 生成上述一切的脚本（只读仓库，不改动任何文件） |
+| extract-candidates.cjs | 从平台仓 quicklog 提取事故候选（根因含实证/坑/假红等信号） |
+| platform-troubleshooting-candidates.csv | 87 条事故候选 backlog（倒序，分诊冷启动用）；mentions_test 为弱指示器 |
+
+## 口径说明（防夸大）
+
+- 「回归钉指标」= test 文件内容含 `回归钉|回归锁定|回归锁` 的文件数，是**指示器**不是精确计数（commit message 里的回归钉不在此列）。
+- 测试文件数 = 全仓 `*.test.* / *.spec.*`，剔除 node_modules、构建产物、`.sillyspec`（运行态）、`.worktrees`（worktree 存储副本）、`.claude`（测试夹具仓）——副本不计入。
+- FR「确认有效」= `最近确认` 的 commit 哈希经 `git cat-file -t` 验证真实存在；不验证场景语义本身。
+- quicklog 状态按条目内 `状态：` 行原样统计。
+- 归档「六件套」= proposal/requirements/design/tasks/verify-result/verify-facts.json 六文件齐备（平台仓多数走轻量档，缺属常态而非数据错误）。
+- troubleshooting「编号坑目」= 标题形如 `## N.` 的条目；另有少量非编号小节。
+- 本目录为分析产物，未纳入 git 跟踪，可整目录删除。
+
+## 历史口径修正记录（防再犯）
+
+1. v1：平台测试计 0 —— 采集面只扫根 `test/`，实际 monorepo 测试在 frontend/sillyhub-daemon 下。
+2. v1：quicklog 460→459、模块文档 37→29（changelog 误计）、troubleshooting 67→66（非编号小节）。
+3. v2：测试数混入 `.worktrees` 存储副本（2746 个）与 `.claude` 夹具仓（302 个），已加排除规则。
+4. v3：「平台归档只 16/293 全档」为误读——212/293 五件套齐，缺的 verify-facts.json 是 2026-09-07 后新版才产的工件（版本时差非缺陷）。
+5. v3：「平台无坑目账本」为误诊——真身在 `knowledge/`（sillyspec-gotchas 10 + testing-gotchas 6 + patterns 10 + uncategorized 39 = 65 条，另有 INDEX 关键词路由 74 行），按 troubleshooting.md 文件名采集致漏。
+6. v4：「事故闭环薄、需补沉淀纪律」为误诊——①uncategorized.md 头部自述为设计内暂存箱（成熟后迁出分类文件+加 INDEX，INDEX 不索引本文件），39 条是流水线在制品非债务；②平台 .runtime 有 knowledge-hit-report.json / knowledge-hits.jsonl / friction-tally 与名为 knowledge-precipitation 的变更——工具的知识沉淀回路在平台仓已在跑；③候选抽样 7~24 多为一次性产品 bug（正确归宿=quicklog+测试+commit），机制坑（sillyspec CLI 坑）平台当日即登记。「蒸馏率低」是产品仓与工具仓的经济差异，非缺陷。

package/asset-audit-2026-09-20/asset-audit.cjs

@@ -0,0 +1,253 @@
+#!/usr/bin/env node
+// 资产审计：对 sillyspec 本仓与 SillyHub 平台仓做只读盘点，产出 CSV/JSON 数据文件。
+// 方法即证据：本脚本只读不写仓库内容，输出全部落在本目录。重跑：node asset-audit.cjs
+'use strict';
+const fs = require('fs');
+const path = require('path');
+const { execSync } = require('child_process');
+
+const OUT = __dirname;
+const REPOS = [
+  { key: 'sillyspec', root: 'C:/Users/qinyi/IdeaProjects/sillyspec' },
+  { key: 'platform', root: 'C:/Users/qinyi/IdeaProjects/multi-agent-platform' },
+];
+
+// ---------- 工具 ----------
+const readText = (p) => { try { return fs.readFileSync(p, 'utf8'); } catch { return null; } };
+const walk = (dir, pred, acc = [], exclude = []) => {
+  let ents; try { ents = fs.readdirSync(dir, { withFileTypes: true }); } catch { return acc; }
+  for (const e of ents) {
+    if (e.isDirectory() && exclude.includes(e.name)) continue;
+    const p = path.join(dir, e.name);
+    if (e.isDirectory()) walk(p, pred, acc, exclude);
+    else if (pred(p)) acc.push(p);
+  }
+  return acc;
+};
+const csvEscape = (s) => `"${String(s == null ? '' : s).replace(/"/g, '""').replace(/\r?\n/g, ' ')}"`;
+const writeCsv = (name, header, rows) => {
+  const body = [header, ...rows].map((r) => r.map(csvEscape).join(',')).join('\r\n');
+  fs.writeFileSync(path.join(OUT, name), '\ufeff' + body, 'utf8'); // BOM 便于 Excel 直接开
+};
+const lineCount = (p) => { const t = readText(p); return t ? t.split(/\r?\n/).length : 0; };
+
+// ---------- 采集器 ----------
+function auditArchive(repo) {
+  const dir = path.join(repo.root, '.sillyspec/changes/archive');
+  const names = fs.existsSync(dir) ? fs.readdirSync(dir, { withFileTypes: true }).filter((e) => e.isDirectory()).map((e) => e.name) : [];
+  const DOCS = ['proposal.md', 'requirements.md', 'design.md', 'tasks.md', 'verify-result.md', 'verify-facts.json'];
+  const rows = []; const complete = []; const dated = [];
+  for (const n of names) {
+    const flags = DOCS.map((d) => (fs.existsSync(path.join(dir, n, d)) ? 1 : 0));
+    rows.push([n, ...flags, flags.reduce((a, b) => a + b, 0)]);
+    if (flags.every((f) => f === 1)) complete.push(n);
+    const m = n.match(/^(\d{4}-\d{2}-\d{2})/); if (m) dated.push(m[1]);
+  }
+  dated.sort();
+  const active = (() => {
+    const cd = path.join(repo.root, '.sillyspec/changes');
+    try { return fs.readdirSync(cd, { withFileTypes: true }).filter((e) => e.isDirectory() && e.name !== 'archive').map((e) => e.name); } catch { return []; }
+  })();
+  writeCsv(`${repo.key}-archive.csv`, ['change', ...DOCS, 'doc_count'], rows);
+  return {
+    total: names.length,
+    fullSixDocs: complete.length,
+    dateRange: dated.length ? `${dated[0]} ~ ${dated[dated.length - 1]}` : '(无日期前缀)',
+    datedNamed: dated.length,
+    activeChanges: active,
+  };
+}
+
+function auditQuicklog(repo) {
+  const dir = path.join(repo.root, '.sillyspec/quicklog');
+  const files = fs.existsSync(dir) ? fs.readdirSync(dir).filter((f) => /^QUICKLOG-.*\.md$/.test(f)).sort() : [];
+  const rows = []; const status = {}; const byMonth = {}; const sessions = new Set();
+  let firstTs = null, lastTs = null;
+  for (const f of files) {
+    const mUser = f.match(/^QUICKLOG-(.+?)(-\d{4}-\d{2}-\d{2})?\.md$/);
+    if (mUser) sessions.add(f.replace(/-\d{4}-\d{2}-\d{2}\.md$/, '').replace(/^QUICKLOG-/, ''));
+    const lines = (readText(path.join(dir, f)) || '').split(/\r?\n/);
+    for (let i = 0; i < lines.length; i++) {
+      const h = lines[i].match(/^## (ql-[A-Za-z0-9-]+) \| ([^|]+) \| (.*)$/);
+      if (!h) continue;
+      const [, id, ts, title] = h;
+      let st = '(未标)';
+      for (let j = i + 1; j < Math.min(i + 6, lines.length); j++) {
+        const s = lines[j].match(/^状态：(.+)$/); if (s) { st = s[1].trim(); break; }
+      }
+      rows.push([id, ts.trim(), st, title.trim()]);
+      status[st] = (status[st] || 0) + 1;
+      const mo = (ts.trim().match(/^(\d{4}-\d{2})/) || [])[1]; if (mo) byMonth[mo] = (byMonth[mo] || 0) + 1;
+      if (!firstTs || ts.trim() < firstTs) firstTs = ts.trim();
+      if (!lastTs || ts.trim() > lastTs) lastTs = ts.trim();
+    }
+  }
+  writeCsv(`${repo.key}-quicklog.csv`, ['ql_id', 'timestamp', 'status', 'title'], rows);
+  return { files: files.length, entries: rows.length, status, byMonth, firstTs, lastTs };
+}
+
+function collectDocFiles(repo) {
+  // 模块文档与 troubleshooting 只认 docs/ 与 .sillyspec/docs/ 两个根，避开 .runtime/worktrees 副本
+  const roots = [path.join(repo.root, 'docs'), path.join(repo.root, '.sillyspec/docs')];
+  const moduleFiles = [], tsFiles = [];
+  for (const r of roots) {
+    walk(r, (p) => { if (/\.md$/.test(p) && path.dirname(p).split(path.sep).pop() === 'modules' && !/\.changelog\.md$/.test(p)) moduleFiles.push(p); });
+    walk(r, (p) => { if (path.basename(p) === 'troubleshooting.md') tsFiles.push(p); });
+  }
+  return { moduleFiles, tsFiles };
+}
+
+function auditModules(repo, files) {
+  const rows = files.map((p) => [path.relative(repo.root, p).replace(/\\/g, '/'), lineCount(p)]).sort((a, b) => a[0].localeCompare(b[0]));
+  writeCsv(`${repo.key}-modules.csv`, ['file', 'lines'], rows);
+  const byDir = {};
+  for (const r of rows) { const d = r[0].split('/').slice(0, -1).join('/'); byDir[d] = (byDir[d] || 0) + 1; }
+  return { count: rows.length, totalLines: rows.reduce((a, b) => a + b[1], 0), byDir };
+}
+
+function auditTroubleshooting(repo, files) {
+  const rows = []; let total = 0; let numbered = 0;
+  for (const p of files) {
+    const lines = (readText(p) || '').split(/\r?\n/);
+    for (const ln of lines) {
+      const h = ln.match(/^##\s+(.+)$/);
+      if (h && !/^##\s*#/.test(ln) && !/^## (参考|附录|索引)/.test(h[1])) {
+        rows.push([path.relative(repo.root, p).replace(/\\/g, '/'), ++total, h[1].trim()]);
+        if (/^\d+[.、]/.test(h[1].trim())) numbered++;
+      }
+    }
+  }
+  writeCsv(`${repo.key}-troubleshooting.csv`, ['file', 'seq', 'title'], rows);
+  return { files: files.map((p) => path.relative(repo.root, p).replace(/\\/g, '/')), entries: total, numberedEntries: numbered };
+}
+
+function auditKnownIssues(repo) {
+  const p = path.join(repo.root, '.sillyspec/knowledge/known-issues.md');
+  const t = readText(p);
+  if (!t) return { exists: false };
+  const rows = []; const tally = {};
+  let cur = null;
+  for (const ln of t.split(/\r?\n/)) {
+    const h = ln.match(/^##\s+(.+)$/);
+    if (h) { if (cur) rows.push(cur); cur = [h[1].trim(), '(未标)']; continue; }
+    const s = ln.match(/^状态[:：]\s*(.+)$/);
+    if (s && cur) { cur[1] = s[1].trim(); tally[cur[1]] = (tally[cur[1]] || 0) + 1; }
+  }
+  if (cur) rows.push(cur);
+  writeCsv(`${repo.key}-knownissues.csv`, ['title', 'status'], rows);
+  return { exists: true, entries: rows.length, statusTally: tally, lines: t.split(/\r?\n/).length };
+}
+
+function auditFr(repo) {
+  const dir = path.join(repo.root, '.sillyspec/knowledge/fr');
+  const files = fs.existsSync(dir) ? fs.readdirSync(dir).filter((f) => f.endsWith('.md')) : [];
+  const rows = []; const uniqHash = new Set();
+  for (const f of files) {
+    const t = readText(path.join(dir, f)) || '';
+    let cur = null;
+    for (const ln of t.split(/\r?\n/)) {
+      const h = ln.match(/^##\s+(FR-.+)$/);
+      if (h) { if (cur) rows.push(cur); cur = [h[1].trim(), '', '', '', '']; continue; }
+      if (!cur) continue;
+      let m = ln.match(/^变更[:：]\s*(.+)$/); if (m) cur[1] = m[1].trim();
+      m = ln.match(/^状态[:：]\s*(.+)$/); if (m) cur[2] = m[1].trim();
+      m = ln.match(/^最近确认[:：]\s*([0-9a-f]{7,40})/); if (m) { cur[3] = m[1]; uniqHash.add(m[1]); }
+    }
+    if (cur) rows.push(cur);
+  }
+  const hashOk = {};
+  for (const h of uniqHash) {
+    try { execSync(`git cat-file -t ${h}`, { cwd: repo.root, stdio: ['ignore', 'ignore', 'ignore'] }); hashOk[h] = 1; }
+    catch { hashOk[h] = 0; }
+  }
+  for (const r of rows) r[4] = r[3] ? String(hashOk[r[3]] === 1) : '(无确认哈希)';
+  writeCsv(`${repo.key}-fr.csv`, ['fr_id', 'source_change', 'status', 'last_confirm_commit', 'commit_exists_in_git'], rows);
+  return {
+    entries: rows.length,
+    withConfirm: rows.filter((r) => r[3]).length,
+    confirmVerified: rows.filter((r) => r[4] === 'true').length,
+  };
+}
+
+function auditTests(repo) {
+  // monorepo：全仓扫（避开 node_modules/.git/构建产物/.sillyspec 运行态/.worktrees 存储副本/.claude 夹具仓），匹配 *.test.* / *.spec.*
+  const EXCLUDE = new Set(['node_modules', '.git', 'dist', 'build', 'coverage', '.sillyspec', '.next', 'target', 'out', '.worktrees', '.claude']);
+  const files = walk(repo.root, (p) => /\.(test|spec)\.[cm]?[jt]s$/.test(p), [], [...EXCLUDE]);
+  const pin = files.filter((p) => { const t = readText(p); return t && /回归钉|回归锁定|回归锁/.test(t); }).length;
+  const byTop = {};
+  for (const p of files) {
+    const rel = path.relative(repo.root, p).replace(/\\/g, '/');
+    const seg = rel.split('/')[0]; byTop[seg] = (byTop[seg] || 0) + 1;
+  }
+  return { testFiles: files.length, testLines: files.reduce((a, p) => a + lineCount(p), 0), regressionPinIndicator: pin, byTopLevelDir: byTop };
+}
+
+function auditGit(repo) {
+  try {
+    const count = execSync('git rev-list --count HEAD', { cwd: repo.root }).toString().trim();
+    const head = execSync('git rev-parse --short HEAD', { cwd: repo.root }).toString().trim();
+    const lastDate = execSync('git log -1 --format=%cs', { cwd: repo.root }).toString().trim();
+    return { commits: Number(count), head, lastCommitDate: lastDate };
+  } catch (e) { return { error: String(e.message) }; }
+}
+
+function auditGotchas(repo) {
+  // knowledge/ 下的坑目/模式账本（gotchas 族）：平台仓的真实坑目基础设施，条目带 frontmatter + INDEX 关键词路由
+  const dir = path.join(repo.root, '.sillyspec/knowledge');
+  const files = fs.existsSync(dir) ? fs.readdirSync(dir).filter((f) => /gotchas|patterns|uncategorized|conventions/.test(f) && f.endsWith('.md')) : [];
+  const rows = []; const tally = {};
+  for (const f of files) {
+    const t = readText(path.join(dir, f)) || '';
+    const entries = (t.match(/^##\s+(?![#])/gm) || []).length;
+    rows.push([`knowledge/${f}`, t.split(/\r?\n/).length, entries]);
+    tally[f] = entries;
+  }
+  const index = readText(path.join(dir, 'INDEX.md')) || '';
+  const routes = (index.match(/^\s*-\s+.+→/gm) || []).length;
+  writeCsv(`${repo.key}-gotchas.csv`, ['file', 'lines', 'entries'], rows);
+  return { files: rows.length, entries: rows.reduce((a, b) => a + b[2], 0), byFile: tally, indexRouteLines: routes };
+}
+
+function auditMisc(repo) {
+  const roadmap = path.join(repo.root, '.sillyspec/ROADMAP.md');
+  const rt = readText(roadmap);
+  const index = path.join(repo.root, '.sillyspec/knowledge/INDEX.md');
+  const it = readText(index);
+  return {
+    roadmap: rt ? { exists: true, lines: rt.split(/\r?\n/).length, items: (rt.match(/^###\s+/gm) || []).length } : { exists: false },
+    knowledgeIndex: it ? { exists: true, lines: it.split(/\r?\n/).length } : { exists: false },
+    redlinesYaml: fs.existsSync(path.join(repo.root, '.sillyspec/redlines.yaml')),
+  };
+}
+
+// ---------- 汇总 ----------
+const summary = { generatedAt: new Date().toISOString(), repos: {} };
+for (const repo of REPOS) {
+  const docFiles = collectDocFiles(repo);
+  summary.repos[repo.key] = {
+    root: repo.root,
+    archive: auditArchive(repo),
+    quicklog: auditQuicklog(repo),
+    moduleDocs: auditModules(repo, docFiles.moduleFiles),
+    troubleshooting: auditTroubleshooting(repo, docFiles.tsFiles),
+    knownIssues: auditKnownIssues(repo),
+    frIndex: auditFr(repo),
+    gotchas: auditGotchas(repo),
+    tests: auditTests(repo),
+    git: auditGit(repo),
+    misc: auditMisc(repo),
+  };
+}
+fs.writeFileSync(path.join(OUT, 'summary.json'), JSON.stringify(summary, null, 2), 'utf8');
+
+// 控制台速览
+for (const [k, v] of Object.entries(summary.repos)) {
+  console.log(`== ${k} (${v.root})`);
+  console.log(`  归档变更 ${v.archive.total}（六件套齐全 ${v.archive.fullSixDocs}，${v.archive.dateRange}）｜活跃 ${v.archive.activeChanges.length}`);
+  console.log(`  quicklog ${v.quicklog.entries} 条 / ${v.quicklog.files} 文件（${v.quicklog.firstTs} ~ ${v.quicklog.lastTs}）状态分布 ${JSON.stringify(v.quicklog.status)}`);
+  console.log(`  模块文档 ${v.moduleDocs.count} 个 / ${v.moduleDocs.totalLines} 行｜troubleshooting ${v.troubleshooting.entries} 条`);
+  console.log(`  known-issues ${v.knownIssues.entries ?? 0} 条｜FR 索引 ${v.frIndex.entries} 条（带确认 ${v.frIndex.withConfirm}，哈希验真 ${v.frIndex.confirmVerified}）`);
+  console.log(`  gotchas 族 ${v.gotchas.entries} 条（${JSON.stringify(v.gotchas.byFile)}，INDEX 路由行 ${v.gotchas.indexRouteLines}）`);
+  console.log(`  测试文件 ${v.tests.testFiles} 个 / ${v.tests.testLines} 行（回归钉指示器 ${v.tests.regressionPinIndicator}）｜git 提交 ${v.git.commits}（HEAD ${v.git.head} @ ${v.git.lastCommitDate}）`);
+}
+console.log('输出目录:', OUT);

package/asset-audit-2026-09-20/extract-candidates.cjs

@@ -0,0 +1,37 @@
+#!/usr/bin/env node
+// 从平台仓 quicklog 提取「事故候选」：根因段含实证/坑/假红/误拦等信号的条目，
+// 作为 troubleshooting.md 冷启动的分诊 backlog。只读，输出 CSV。
+// 口径：mentions_test = 条目块内出现 test/spec/pytest 字样（事故是否钉了测试的弱指示器）。
+'use strict';
+const fs = require('fs');
+const path = require('path');
+const OUT = __dirname;
+const DIR = 'C:/Users/qinyi/IdeaProjects/multi-agent-platform/.sillyspec/quicklog';
+
+const MARKER = /实证|踩坑|两坑|三坑|四坑|五坑|假红|假败|误拦|误报|误伤|事故|意外|教训|摩擦|坑位|翻车|卡死|挂死/;
+const csvEscape = (s) => `"${String(s == null ? '' : s).replace(/"/g, '""').replace(/\r?\n/g, ' ')}"`;
+
+const files = fs.readdirSync(DIR).filter((f) => /^QUICKLOG-.*\.md$/.test(f)).sort();
+const rows = []; let total = 0;
+for (const f of files) {
+  const text = fs.readFileSync(path.join(DIR, f), 'utf8');
+  const blocks = text.split(/^(?=## ql-)/m);
+  for (const b of blocks) {
+    const h = b.match(/^## (ql-[A-Za-z0-9-]+) \| ([^|]+) \| ([^\r\n]*)/);
+    if (!h) continue;
+    total++;
+    const rootCause = ((b.match(/^根因[:：]\s*(.+)$/m) || [])[1] || '').trim();
+    if (!MARKER.test(rootCause) && !MARKER.test(h[3])) continue;
+    const mentionsTest = /test|spec|pytest/i.test(b);
+    rows.push([h[1], h[2].trim(), h[3].trim().slice(0, 90), rootCause.slice(0, 160), String(mentionsTest)]);
+  }
+}
+rows.sort((a, b) => (a[1] < b[1] ? 1 : -1)); // 新的在前，分诊从最近的开始
+fs.writeFileSync(
+  path.join(OUT, 'platform-troubleshooting-candidates.csv'),
+  '\ufeff' + [['ql_id', 'timestamp', 'title', 'root_cause_excerpt', 'mentions_test'], ...rows].map((r) => r.map(csvEscape).join(',')).join('\r\n'),
+  'utf8'
+);
+const withTest = rows.filter((r) => r[4] === 'true').length;
+console.log(`平台 quicklog 总条目 ${total}，命中事故信号 ${rows.length} 条（其中带 test 字样 ${withTest} 条，链接率 ${(rows.length ? ((withTest / rows.length) * 100).toFixed(0) : 0)}%）`);
+console.log('输出: platform-troubleshooting-candidates.csv（按时间倒序，分诊从最近的做起）');