@jaimevalasek/aioson 1.22.0 → 1.23.1

package/src/lib/retro/retro-sources.js

@@ -0,0 +1,624 @@
+'use strict';
+
+/**
+ * Leitores de fonte para `aioson harness:retro` (RHO-lite, requirements §3.2/§5.1).
+ *
+ * Um leitor por fonte; cada um é best-effort independente (padrão de
+ * `attempt-artifacts.js`): nunca propaga exceção, sempre devolve
+ * `{ findings, warnings, count, ... }`. Fonte ausente, vazia, ilegível ou DB
+ * lockado vira linha de aviso na "Trilha minerada", nunca erro fatal (REQ-3).
+ *
+ * Mineração 100% determinística (REQ-1): nenhuma chamada LLM, nenhuma
+ * classificação semântica — só regex e chaves exatas. Leitura-apenas: este
+ * módulo NUNCA escreve no filesystem.
+ */
+
+const fs = require('node:fs');
+const path = require('node:path');
+const crypto = require('node:crypto');
+
+const FINDING_ID_RE = /\b([A-Z]{1,2}-\d{1,2})\b/;
+const TRAIL_ENTRY_RE = /^\*\*([^*]+)\*\*\s*\|\s*@?([\w.-]+)\s*\|\s*_([^_]+)_\s*$/;
+const VERDICT_RE = /\b(?:verdict|veredicto)\b[^\n]*?\b(PASS|FAIL)\b/i;
+const VERDICT_FALLBACK_RE = /\b(PASS|FAIL)\b/i;
+const ISO_RE = /^\d{4}-\d{2}-\d{2}T\d{2}:\d{2}/;
+
+const SEVERITY_ALIASES = {
+  critical: 'critical',
+  crit: 'critical',
+  high: 'high',
+  med: 'medium',
+  medium: 'medium',
+  low: 'low',
+  info: 'info',
+  informational: 'info'
+};
+
+/** Normaliza severidade (case-insensitive); desconhecida → `unknown` (nunca promove). */
+function normalizeSeverity(raw) {
+  if (raw === null || raw === undefined) return 'unknown';
+  const key = String(raw).trim().toLowerCase();
+  return SEVERITY_ALIASES[key] || 'unknown';
+}
+
+/** Path relativo com separador POSIX (determinismo cross-OS — AC-4, EC Windows). */
+function relPath(rootDir, p) {
+  return path.relative(rootDir, p).replaceAll('\\', '/');
+}
+
+function readTextSafe(p) {
+  try {
+    return fs.readFileSync(p, 'utf8');
+  } catch {
+    return null;
+  }
+}
+
+function listFilesSafe(dir) {
+  try {
+    return fs.readdirSync(dir, { withFileTypes: true });
+  } catch {
+    return [];
+  }
+}
+
+function statSizeSafe(p) {
+  try {
+    return fs.statSync(p).size;
+  } catch {
+    return 0;
+  }
+}
+
+/** Parser de frontmatter YAML simples (`key: value`, valor escalar de 1 linha). */
+function parseFrontmatter(text) {
+  if (typeof text !== 'string' || !text.startsWith('---')) return { data: {}, body: text || '' };
+  const end = text.indexOf('\n---', 3);
+  if (end === -1) return { data: {}, body: text };
+  const block = text.slice(3, end).replace(/^\r?\n/, '');
+  const body = text.slice(end + 4);
+  const data = {};
+  for (const line of block.split(/\r?\n/)) {
+    const m = line.match(/^([A-Za-z0-9_]+):\s*(.*)$/);
+    if (!m) continue;
+    // Remove comentário inline (ex.: `status: resolved   # open | ...`).
+    let value = m[2].replace(/\s+#.*$/, '').trim();
+    value = value.replace(/^["']|["']$/g, '');
+    data[m[1]] = value;
+  }
+  return { data, body };
+}
+
+function isoDate(raw) {
+  if (!raw) return null;
+  const m = String(raw).match(/\d{4}-\d{2}-\d{2}(?:T[\d:.+Z-]+)?/);
+  return m ? m[0] : null;
+}
+
+function makeFinding(partial) {
+  return {
+    source_type: partial.source_type,
+    feature_slug: partial.feature_slug,
+    finding_id: partial.finding_id || null,
+    severity: normalizeSeverity(partial.severity),
+    title: (partial.title ? String(partial.title) : '').slice(0, 200),
+    file_ref: partial.file_ref || null,
+    date: partial.date || null,
+    status: partial.status || 'unknown',
+    source_path: partial.source_path,
+    signature: partial.signature || null
+  };
+}
+
+// --- 1. QA reports ----------------------------------------------------------
+
+const KNOWN_SEVERITIES = new Set(['critical', 'high', 'medium', 'low', 'info']);
+
+/** Extrai findings estruturados de um QA report (tabelas + headers de finding). */
+function extractQaFindings({ body, slug, sourcePath, date, status }) {
+  const found = new Map(); // finding_id → finding (primeiro vence)
+  for (const rawLine of body.split(/\r?\n/)) {
+    const line = rawLine.trim();
+    // Linha de tabela: | ID | Sev | ... |
+    if (line.startsWith('|')) {
+      const cells = line.split('|').map((c) => c.trim()).filter((c, i, arr) => i > 0 && i < arr.length);
+      if (cells.length >= 2) {
+        const idMatch = cells[0].match(FINDING_ID_RE);
+        if (idMatch) {
+          const sevCell = cells.find((c) => KNOWN_SEVERITIES.has(c.toLowerCase()));
+          if (sevCell) {
+            const id = idMatch[1];
+            if (!found.has(id)) {
+              found.set(id, makeFinding({
+                source_type: 'qa_report', feature_slug: slug, finding_id: id,
+                severity: sevCell, title: cells[2] || cells[1] || id,
+                date, status, source_path: sourcePath
+              }));
+            }
+          }
+        }
+      }
+      continue;
+    }
+    // Header de finding: ### C-01 — Title (High)
+    const head = rawLine.match(/^#{2,5}\s+([A-Z]{1,2}-\d{1,2})\b\s*[—:-]?\s*(.*)$/);
+    if (head) {
+      const id = head[1];
+      if (!found.has(id)) {
+        const sevMatch = head[2].match(/\(([A-Za-z]+)/);
+        found.set(id, makeFinding({
+          source_type: 'qa_report', feature_slug: slug, finding_id: id,
+          severity: sevMatch ? sevMatch[1] : 'unknown',
+          title: head[2].replace(/\(.*$/, '').trim() || id,
+          date, status, source_path: sourcePath
+        }));
+      }
+    }
+  }
+  return [...found.values()];
+}
+
+function readQaReports({ rootDir, ctxDir, slug, locations }) {
+  const findings = [];
+  const warnings = [];
+  let count = 0;
+  for (const dir of locations.qaDirs) {
+    for (const ent of listFilesSafe(dir)) {
+      if (!ent.isFile()) continue;
+      if (!/^qa-report-/.test(ent.name) || !ent.name.endsWith('.md')) continue;
+      if (!ent.name.includes(slug)) continue;
+      const full = path.join(dir, ent.name);
+      const text = readTextSafe(full);
+      if (text === null) {
+        warnings.push(`qa_report ilegível: ${relPath(rootDir, full)}`);
+        continue;
+      }
+      count += 1;
+      const { data, body } = parseFrontmatter(text);
+      const date = isoDate(data.created_at || data.updated_at || data.date);
+      const status = (data.verdict || '').toUpperCase() === 'FAIL' ? 'open' : 'fixed';
+      findings.push(...extractQaFindings({ body, slug, sourcePath: relPath(rootDir, full), date, status }));
+    }
+  }
+  return { findings, warnings, count };
+}
+
+// --- 2. Corrections plans ---------------------------------------------------
+
+function mapCorrectionStatus(raw) {
+  const v = String(raw || '').toLowerCase();
+  if (v === 'resolved') return 'fixed';
+  if (v === 'open' || v === 'in_progress') return 'open';
+  return 'unknown';
+}
+
+function readCorrections({ rootDir, slug, locations }) {
+  const findings = [];
+  const warnings = [];
+  let count = 0;
+  let bytes = 0;
+  let entries = 0;
+  for (const dir of locations.planDirs) {
+    for (const ent of listFilesSafe(dir)) {
+      if (!ent.isFile()) continue;
+      if (!/^corrections-.*\.md$/.test(ent.name)) continue;
+      const full = path.join(dir, ent.name);
+      const text = readTextSafe(full);
+      if (text === null) {
+        warnings.push(`corrections ilegível: ${relPath(rootDir, full)}`);
+        continue;
+      }
+      count += 1;
+      bytes += statSizeSafe(full);
+      const { data, body } = parseFrontmatter(text);
+      const status = mapCorrectionStatus(data.status);
+      const date = isoDate(data.created || data.date);
+      const sourcePath = relPath(rootDir, full);
+      const lines = body.split(/\r?\n/);
+      for (let i = 0; i < lines.length; i += 1) {
+        const head = lines[i].match(/^#{2,5}\s+([A-Z]{1,2}-\d{1,2})\s*[—:-]\s*(.+?)\s*$/);
+        if (!head) continue;
+        entries += 1;
+        const sevMatch = head[2].match(/\(([A-Za-z]+)/);
+        // Procura linha File:/Files: logo após o header.
+        let fileRef = null;
+        for (let j = i + 1; j < Math.min(i + 4, lines.length); j += 1) {
+          const fm = lines[j].match(/^Files?:\s*(.+)$/i);
+          if (fm) { fileRef = fm[1].trim(); break; }
+        }
+        findings.push(makeFinding({
+          source_type: 'corrections', feature_slug: slug, finding_id: head[1],
+          severity: sevMatch ? sevMatch[1] : 'unknown',
+          title: head[2].replace(/\s*\(.*$/, '').trim(),
+          file_ref: fileRef, date, status, source_path: sourcePath
+        }));
+      }
+    }
+  }
+  return { findings, warnings, count, bytes, entries };
+}
+
+// --- 3. Dossier Agent Trail (verdicts + ciclos FAIL→PASS) -------------------
+
+function readDossierTrail({ rootDir, slug, locations }) {
+  const findings = [];
+  const warnings = [];
+  const cycles = [];
+  let count = 0;
+  let illegible = 0;
+
+  for (const full of locations.dossierFiles) {
+    const text = readTextSafe(full);
+    if (text === null) continue;
+    const sourcePath = relPath(rootDir, full);
+    const trailIdx = text.indexOf('## Agent Trail');
+    const region = trailIdx === -1 ? text : text.slice(trailIdx);
+    const lines = region.split(/\r?\n/);
+
+    // Acha headers de entrada e fatia o corpo entre eles.
+    const entries = [];
+    for (let i = 0; i < lines.length; i += 1) {
+      const m = lines[i].match(TRAIL_ENTRY_RE);
+      if (!m) continue;
+      const ts = m[1].trim();
+      const agent = m[2].trim().toLowerCase();
+      const section = m[3].trim();
+      if (!ISO_RE.test(ts)) { illegible += 1; continue; }
+      // Corpo: até o próximo header de entrada.
+      const bodyLines = [];
+      for (let j = i + 1; j < lines.length; j += 1) {
+        if (TRAIL_ENTRY_RE.test(lines[j])) break;
+        if (/^<!--\s*sha256:/.test(lines[j])) continue;
+        bodyLines.push(lines[j]);
+      }
+      entries.push({ ts, agent, section, body: bodyLines.join('\n') });
+    }
+
+    if (entries.length === 0) continue;
+    count += entries.length;
+
+    // Verdicts ordenados por timestamp → ciclos FAIL→PASS (D5). O trail é fonte
+    // de VERDICTS/ciclos, não de findings: extrair findings do resumo @qa do
+    // trail duplicaria o que já vem do corrections plan / QA report (fonte
+    // autoritativa). Mantemos a leitura determinística e sem dupla contagem.
+    const verdicts = [];
+    for (const e of entries) {
+      let vm = e.body.match(VERDICT_RE);
+      if (!vm && e.agent === 'qa') vm = e.body.match(VERDICT_FALLBACK_RE);
+      if (vm) verdicts.push({ ts: e.ts, verdict: vm[1].toUpperCase() });
+    }
+    verdicts.sort((a, b) => (a.ts < b.ts ? -1 : a.ts > b.ts ? 1 : 0));
+    let openFail = null;
+    for (const v of verdicts) {
+      if (v.verdict === 'FAIL') {
+        openFail = v.ts;
+      } else if (v.verdict === 'PASS' && openFail) {
+        cycles.push({ feature_slug: slug, fail_at: openFail, pass_at: v.ts, source_path: sourcePath });
+        openFail = null;
+      }
+    }
+  }
+
+  if (illegible > 0) warnings.push(`dossier_trail: ${illegible} entrada(s) ilegível(is) (sem timestamp ISO)`);
+  return { findings, warnings, count, cycles };
+}
+
+// --- 4. execution_events (aios.sqlite, readonly — D7) -----------------------
+
+function readExecutionEvents({ rootDir, targetDir, slug }) {
+  const warnings = [];
+  let count = 0;
+  let tokenAvailable = false;
+  let tokenTotal = 0;
+  const dbPath = path.join(targetDir, '.aioson', 'runtime', 'aios.sqlite');
+  if (!fs.existsSync(dbPath)) {
+    return { findings: [], warnings: [`execution_events: ${relPath(rootDir, dbPath)} ausente`], count: 0, tokenAvailable, tokenTotal };
+  }
+  let Database;
+  try {
+    Database = require('better-sqlite3');
+  } catch {
+    return { findings: [], warnings: ['execution_events: better-sqlite3 indisponível'], count: 0, tokenAvailable, tokenTotal };
+  }
+  let db = null;
+  try {
+    db = new Database(dbPath, { readonly: true, fileMustExist: true });
+    const rows = db.prepare('SELECT payload_json, token_count FROM execution_events').all();
+    for (const row of rows) {
+      if (!row.payload_json) continue;
+      let payload;
+      try {
+        payload = JSON.parse(row.payload_json);
+      } catch {
+        continue;
+      }
+      if (!payload || payload.slug !== slug) continue;
+      count += 1;
+      if (row.token_count !== null && row.token_count !== undefined) {
+        tokenAvailable = true;
+        tokenTotal += Number(row.token_count) || 0;
+      }
+    }
+  } catch (err) {
+    warnings.push(`execution_events: DB ilegível/lockado (${err.code || 'erro'})`);
+  } finally {
+    try { if (db) db.close(); } catch { /* best-effort */ }
+  }
+  return { findings: [], warnings, count, tokenAvailable, tokenTotal };
+}
+
+// --- 5. attempts/{n}/ -------------------------------------------------------
+
+function readAttempts({ rootDir, slug, locations }) {
+  const warnings = [];
+  let count = 0;
+  for (const planDir of locations.planDirs) {
+    const attemptsDir = path.join(planDir, 'attempts');
+    for (const ent of listFilesSafe(attemptsDir)) {
+      if (ent.isDirectory() && /^\d+$/.test(ent.name)) count += 1;
+    }
+  }
+  return { findings: [], warnings, count };
+}
+
+// --- 6. progress.json failure_signatures ------------------------------------
+
+function readFailureSignatures({ rootDir, slug, locations }) {
+  const findings = [];
+  const warnings = [];
+  let count = 0;
+  for (const planDir of locations.planDirs) {
+    const full = path.join(planDir, 'progress.json');
+    const text = readTextSafe(full);
+    if (text === null) continue;
+    let data;
+    try {
+      data = JSON.parse(text);
+    } catch {
+      warnings.push(`progress.json ilegível: ${relPath(rootDir, full)}`);
+      continue;
+    }
+    const sigs = Array.isArray(data.failure_signatures) ? data.failure_signatures : [];
+    const sourcePath = relPath(rootDir, full);
+    for (const entry of sigs) {
+      const signature = typeof entry === 'string' ? entry : (entry && (entry.signature || entry.sha1)) || null;
+      if (!signature) continue;
+      const occurrences = typeof entry === 'object' && Number.isInteger(entry.occurrences)
+        ? Math.max(1, entry.occurrences)
+        : (typeof entry === 'object' && Number.isInteger(entry.count) ? Math.max(1, entry.count) : 1);
+      const title = (typeof entry === 'object' && entry.title) ? entry.title : `failure signature ${String(signature).slice(0, 12)}`;
+      const severity = (typeof entry === 'object' && entry.severity) ? entry.severity : 'unknown';
+      for (let k = 0; k < occurrences; k += 1) {
+        count += 1;
+        findings.push(makeFinding({
+          source_type: 'progress', feature_slug: slug, finding_id: null,
+          severity, title, date: null, status: 'open',
+          source_path: sourcePath, signature: String(signature)
+        }));
+      }
+    }
+  }
+  return { findings, warnings, count };
+}
+
+// --- 7. Devlogs (aioson-logs/) ----------------------------------------------
+
+function readDevlogs({ rootDir, targetDir, slug }) {
+  const findings = [];
+  const warnings = [];
+  let count = 0;
+  const logsDir = path.join(targetDir, 'aioson-logs');
+  for (const ent of listFilesSafe(logsDir)) {
+    if (!ent.isFile() || !ent.name.endsWith('.md')) continue;
+    const full = path.join(logsDir, ent.name);
+    const text = readTextSafe(full);
+    if (text === null) continue;
+    const { data } = parseFrontmatter(text);
+    if (data.feature !== slug) continue;
+    count += 1;
+  }
+  return { findings, warnings, count };
+}
+
+// --- Localização de artefatos por feature (ativo + arquivado) ---------------
+
+function dirExists(p) {
+  try {
+    return fs.statSync(p).isDirectory();
+  } catch {
+    return false;
+  }
+}
+
+/**
+ * Resolve os diretórios/arquivos de uma feature em locais ativos E arquivados
+ * (`done/{slug}/`). Não inventa um segundo enumerador — espelha o layout que
+ * `feature-archive.js` produz.
+ */
+function resolveLocations(targetDir, ctxDir, slug) {
+  const doneRoot = path.join(ctxDir, 'done', slug);
+
+  const qaDirs = [ctxDir, doneRoot].filter(dirExists);
+
+  const planDirs = [
+    path.join(targetDir, '.aioson', 'plans', slug),
+    path.join(doneRoot, 'plans')
+  ].filter(dirExists);
+
+  const dossierCandidates = [
+    path.join(ctxDir, 'features', slug, 'dossier.md'),
+    path.join(doneRoot, 'features', slug, 'dossier.md'),
+    path.join(doneRoot, 'dossier.md')
+  ];
+  // SF-02: lstat (não statSync) para NÃO seguir symlink — consistente com os
+  // demais readers, que pulam symlinks via Dirent.isFile(). Um dossier.md que
+  // seja um symlink apontando para fora do workspace é ignorado, não seguido.
+  const dossierFiles = dossierCandidates.filter((p) => {
+    try { return fs.lstatSync(p).isFile(); } catch { return false; }
+  });
+
+  return { qaDirs, planDirs, dossierFiles, doneRoot };
+}
+
+/**
+ * Minera todas as fontes de UMA feature. Best-effort: nunca lança.
+ *
+ * @returns {{ slug, findings, cycles, counts, cost, minedPaths, warnings }}
+ */
+function collectFeatureSources(targetDir, slug) {
+  const ctxDir = path.join(targetDir, '.aioson', 'context');
+  const rootDir = targetDir;
+  const locations = resolveLocations(targetDir, ctxDir, slug);
+
+  const warnings = [];
+  const findings = [];
+  const cycles = [];
+
+  const qa = readQaReports({ rootDir, ctxDir, slug, locations });
+  const corr = readCorrections({ rootDir, slug, locations });
+  const trail = readDossierTrail({ rootDir, slug, locations });
+  const events = readExecutionEvents({ rootDir, targetDir, slug });
+  const attempts = readAttempts({ rootDir, slug, locations });
+  const sigs = readFailureSignatures({ rootDir, slug, locations });
+  const devlogs = readDevlogs({ rootDir, targetDir, slug });
+
+  for (const r of [qa, corr, trail, events, attempts, sigs, devlogs]) {
+    findings.push(...r.findings);
+    warnings.push(...r.warnings);
+  }
+  cycles.push(...trail.cycles);
+
+  const counts = {
+    qa_reports: qa.count,
+    corrections: corr.count,
+    dossier_trail: trail.count,
+    execution_events: events.count,
+    attempts: attempts.count,
+    failure_signatures: sigs.count,
+    devlogs: devlogs.count
+  };
+
+  const cost = {
+    execution_events: events.count,
+    corrections: corr.entries,
+    fail_pass_cycles: cycles.length,
+    corrections_bytes: corr.bytes,
+    token_count_available: events.tokenAvailable,
+    token_total: events.tokenAvailable ? events.tokenTotal : null
+  };
+
+  const minedPaths = [];
+  for (const d of locations.qaDirs) minedPaths.push(relPath(rootDir, d));
+  for (const d of locations.planDirs) minedPaths.push(relPath(rootDir, d));
+  for (const f of locations.dossierFiles) minedPaths.push(relPath(rootDir, f));
+  minedPaths.sort();
+
+  return { slug, findings, cycles, counts, cost, minedPaths, warnings };
+}
+
+/**
+ * Minera uma janela de features (1+). Soma contagens/custo e concatena
+ * findings/cycles (a chave de agrupamento inclui o slug, então misturar é seguro).
+ */
+function collectSources(targetDir, slugs) {
+  const list = Array.isArray(slugs) ? slugs : [slugs];
+  const findings = [];
+  const cycles = [];
+  const warnings = [];
+  const minedPaths = [];
+  const counts = { qa_reports: 0, corrections: 0, dossier_trail: 0, execution_events: 0, attempts: 0, failure_signatures: 0, devlogs: 0 };
+  const cost = { execution_events: 0, corrections: 0, fail_pass_cycles: 0, corrections_bytes: 0, token_count_available: false, token_total: null };
+  const costByFeature = {};
+
+  for (const slug of list) {
+    const f = collectFeatureSources(targetDir, slug);
+    findings.push(...f.findings);
+    cycles.push(...f.cycles);
+    warnings.push(...f.warnings);
+    minedPaths.push(...f.minedPaths);
+    costByFeature[slug] = f.cost;
+    for (const k of Object.keys(counts)) counts[k] += f.counts[k];
+    cost.execution_events += f.cost.execution_events;
+    cost.corrections += f.cost.corrections;
+    cost.fail_pass_cycles += f.cost.fail_pass_cycles;
+    cost.corrections_bytes += f.cost.corrections_bytes;
+    if (f.cost.token_count_available) {
+      cost.token_count_available = true;
+      cost.token_total = (cost.token_total || 0) + (f.cost.token_total || 0);
+    }
+  }
+
+  return { features_mined: list.slice(), findings, cycles, counts, cost, costByFeature, minedPaths, warnings };
+}
+
+/** Subdiretórios de `.aioson/context/done/` = features fechadas (D6). */
+function enumerateClosedFeatures(targetDir) {
+  const doneDir = path.join(targetDir, '.aioson', 'context', 'done');
+  const slugs = [];
+  for (const ent of listFilesSafe(doneDir)) {
+    if (ent.isDirectory()) slugs.push(ent.name);
+  }
+  slugs.sort();
+  return slugs;
+}
+
+/**
+ * Data de PASS de uma feature para ordenar a janela `--last=N` (D2: trail vence).
+ * Prioridade: último PASS do Agent Trail → frontmatter de QA report → null.
+ * (O caller pode cair em MANIFEST `completed` quando isto retornar null.)
+ */
+function resolveFeatureExists(targetDir, slug) {
+  const ctxDir = path.join(targetDir, '.aioson', 'context');
+  const loc = resolveLocations(targetDir, ctxDir, slug);
+  if (loc.planDirs.length > 0 || loc.dossierFiles.length > 0) return true;
+  if (dirExists(loc.doneRoot)) return true;
+  if (dirExists(path.join(ctxDir, 'features', slug))) return true;
+  // qa-report-{slug}*.md no contexto ativo
+  for (const ent of listFilesSafe(ctxDir)) {
+    if (ent.isFile() && /^qa-report-/.test(ent.name) && ent.name.includes(slug) && ent.name.endsWith('.md')) return true;
+  }
+  // listado em features.md
+  const featuresText = readTextSafe(path.join(ctxDir, 'features.md'));
+  if (featuresText && new RegExp(`\\b${slug.replace(/[-/\\^$*+?.()|[\]{}]/g, '\\$&')}\\b`).test(featuresText)) return true;
+  return false;
+}
+
+function resolvePassDate(targetDir, slug) {
+  const ctxDir = path.join(targetDir, '.aioson', 'context');
+  const locations = resolveLocations(targetDir, ctxDir, slug);
+
+  // 1. Trail: maior pass_at dos ciclos FAIL→PASS.
+  const trail = readDossierTrail({ rootDir: targetDir, slug, locations });
+  if (trail.cycles.length > 0) {
+    return trail.cycles.map((c) => c.pass_at).sort().slice(-1)[0];
+  }
+
+  // 2. QA report com verdict PASS → created_at/updated_at.
+  let best = null;
+  for (const dir of locations.qaDirs) {
+    for (const ent of listFilesSafe(dir)) {
+      if (!ent.isFile() || !/^qa-report-/.test(ent.name) || !ent.name.includes(slug) || !ent.name.endsWith('.md')) continue;
+      const text = readTextSafe(path.join(dir, ent.name));
+      if (text === null) continue;
+      const { data } = parseFrontmatter(text);
+      if ((data.verdict || '').toUpperCase() === 'FAIL') continue;
+      const d = isoDate(data.updated_at || data.created_at || data.date);
+      if (d && (!best || d > best)) best = d;
+    }
+  }
+  return best;
+}
+
+module.exports = {
+  collectSources,
+  collectFeatureSources,
+  resolveLocations,
+  resolvePassDate,
+  resolveFeatureExists,
+  enumerateClosedFeatures,
+  normalizeSeverity,
+  parseFrontmatter,
+  relPath,
+  // exportados para teste unitário dos parsers
+  _internal: { readQaReports, readCorrections, readDossierTrail, readFailureSignatures, makeFinding }
+};

package/src/parser.js

@@ -29,7 +29,7 @@ function parseArgv(argv) {
         'all', 'force', 'dry-run', 'no-interactive', 'fix', 'json',
         'help', 'version', 'no-launch', 'attach', 'tmux',
         'allow-warnings', 'install-hook', 'uninstall-hook', 'remove-hook',
-        'agent-safe',
+        'agent-safe', 'agentic',
         'selective',
         'status', 'suggest', 'apply',
         'runtime-only', 'template-only', 'inception', 'locales',