tribunal-kit 9.2.3 → 9.2.5

package/.agent/scripts/inner_loop_validator.js

@@ -1,472 +1,472 @@
-#!/usr/bin/env node
-/**
- * inner_loop_validator.js — Tribunal Kit Inner-Loop Self-Healing CI
- * ==================================================================
- * Orchestrates security_scan.js and lint heuristics on code snippets
- * IN MEMORY before they are presented to the Human Gate.
- *
- * This is the "Phase 6 Auto-Correction Engine": it feeds structured
- * JSON findings back to the Maker Agent for autonomous self-healing
- * without requiring user involvement.
- *
- * Architecture:
- *   1. Receive a code snippet (via stdin or --snippet flag)
- *   2. Write to a temp file within the OS temp directory
- *   3. Run OWASP security pattern scan (using PATTERNS from security_scan.js)
- *   4. Run lightweight syntax heuristics (no external tools required)
- *   5. Emit structured JSON verdict for the Maker Agent to consume
- *   6. Clean up temp file
- *
- * Usage:
- *   node .agent/scripts/inner_loop_validator.js --snippet "const x = eval(input)"
- *   node .agent/scripts/inner_loop_validator.js --file ./output.js
- *   node .agent/scripts/inner_loop_validator.js --file ./output.js --lang ts
- *   node .agent/scripts/inner_loop_validator.js test-case
- *
- * Output (JSON to stdout):
- *   {
- *     "verdict": "APPROVED" | "WARNING" | "REJECTED",
- *     "passed": boolean,
- *     "issues": [{ "severity": "critical|high|medium|low", "category": string, "line": number, "message": string, "fix": string }],
- *     "summary": string,
- *     "self_healing_instructions": string | null   ← fed back to Maker Agent
- *   }
- */
-
-'use strict';
-
-const fs = require('fs');
-const path = require('path');
-
-// ── Resolve security_scan patterns (reuse — do not duplicate) ─────────────
-const SCRIPT_DIR = __dirname;
-let SECURITY_PATTERNS = [];
-let SEVERITY_RANK = {};
-
-try {
-  const secScan = require(path.join(SCRIPT_DIR, 'security_scan.js'));
-  SECURITY_PATTERNS = secScan.PATTERNS || [];
-  SEVERITY_RANK = secScan.SEVERITY_RANK || {
-    critical: 0,
-    high: 1,
-    medium: 2,
-    low: 3,
-  };
-} catch {
-  // Fallback: minimal critical patterns only (never fail silently on missing module)
-  SECURITY_PATTERNS = [
-    [
-      /(?:password|passwd)\s*=\s*["'][^"']+["']/i,
-      'critical',
-      'Hardcoded Secret',
-      'Hardcoded password',
-    ],
-    [/\beval\s*\(/, 'high', 'Code Injection', 'eval() is a code injection vector'],
-    [/\.innerHTML\s*=/, 'high', 'XSS', 'Direct innerHTML assignment'],
-    [/algorithms\s*:\s*\[\s*["']none["']/, 'critical', 'Auth Bypass', "JWT 'none' algorithm"],
-  ];
-  SEVERITY_RANK = { critical: 0, high: 1, medium: 2, low: 3 };
-}
-
-// ── Syntax heuristics (no external deps) ─────────────────────────────────
-// These catch structural issues in generated code before a linter runs.
-const SYNTAX_HEURISTICS = [
-  {
-    pattern: /\bconst\s+\w+\s*=\s*require\s*\(\s*["'](?!\.\/|\.\.\/|[a-zA-Z])/,
-    severity: 'medium',
-    category: 'Hallucination Risk',
-    message: 'Suspicious require() path — verify module exists in package.json',
-    fix: 'Check that this package is listed in package.json dependencies',
-  },
-  {
-    pattern: /\/\/\s*VERIFY:/,
-    severity: 'low',
-    category: 'Verification Flag',
-    message: 'Maker Agent flagged this line as uncertain — human review required',
-    fix: 'The Maker Agent marked this with // VERIFY: — confirm before approving',
-  },
-  {
-    pattern: /:\s*any\b(?!\s*=)/,
-    severity: 'low',
-    category: 'Type Safety',
-    message: 'TypeScript `any` type used without explanation comment',
-    fix: 'Replace :any with a specific type, or add // any: [reason] comment',
-  },
-  {
-    pattern: /process\.env\.\w+(?!\s*\?\?|\s*\|\|)/,
-    severity: 'low',
-    category: 'Config Safety',
-    message: 'process.env access without nullish fallback — may throw at runtime',
-    fix: 'Use: process.env.VAR ?? "default" — always guard env var access',
-  },
-  {
-    pattern: /throw\s+["'`]/,
-    severity: 'low',
-    category: 'Error Quality',
-    message: 'Throwing a string instead of an Error object — stack traces will be lost',
-    fix: 'Use: throw new Error("message") instead of throw "message"',
-  },
-  {
-    pattern: /catch\s*\(\s*\w+\s*\)\s*\{?\s*\}/,
-    severity: 'medium',
-    category: 'Error Handling',
-    message: 'Empty catch block swallows errors silently',
-    fix: 'Add at minimum: catch (err) { console.error(err); throw err; }',
-  },
-  {
-    pattern: /\.then\(\s*\)\s*\.catch\s*\(|\.catch\s*\(\s*(?:[^)]*)\s*\)\s*\{\s*\}/,
-    severity: 'medium',
-    category: 'Error Handling',
-    message: 'Empty .then() or .catch() handler — Promise errors may be silenced',
-    fix: 'Implement proper resolution and rejection handlers',
-  },
-  {
-    pattern: /window\.|document\.|navigator\./,
-    severity: 'low',
-    category: 'Environment Check',
-    message: 'Browser global access — may fail in SSR/Node environments',
-    fix: 'Guard with: typeof window !== "undefined" before accessing browser globals',
-  },
-];
-
-// ── ANSI colors (centralized via _colors.js) ─────────────────────────────
-const { GREEN, YELLOW, RED, CYAN, BOLD, DIM, RESET } = require('./_colors');
-
-// ── Core scanning ─────────────────────────────────────────────────────────
-
-/**
- * Scan a code string for security and heuristic issues.
- * Returns an array of structured finding objects.
- *
- * @param {string} code      - Raw source code string
- * @param {string} [lang]    - Language hint ('js' | 'ts' | 'py' | 'jsx' | 'tsx')
- * @returns {Array<{severity, category, line, message, fix, source}>}
- */
-function scanCode(code, _lang = 'js') {
-  const findings = [];
-  const lines = code.split('\n');
-
-  for (let i = 0; i < lines.length; i++) {
-    const stripped = lines[i].trim();
-    const lineNum = i + 1;
-
-    // Skip pure comments
-    if (stripped.startsWith('//') || stripped.startsWith('#') || stripped.startsWith('*')) {
-      continue;
-    }
-
-    // Run OWASP security patterns
-    for (const [pattern, severity, category, message] of SECURITY_PATTERNS) {
-      if (pattern.test(stripped)) {
-        findings.push({
-          severity,
-          category,
-          line: lineNum,
-          message,
-          fix: buildSecurityFix(category),
-          source: 'security_scan',
-        });
-      }
-    }
-
-    // Run structural heuristics
-    for (const h of SYNTAX_HEURISTICS) {
-      if (h.pattern.test(stripped)) {
-        findings.push({
-          severity: h.severity,
-          category: h.category,
-          line: lineNum,
-          message: h.message,
-          fix: h.fix,
-          source: 'heuristic',
-        });
-      }
-    }
-  }
-
-  return findings;
-}
-
-/**
- * Build a fix suggestion for known security categories.
- * @param {string} category
- * @returns {string}
- */
-function buildSecurityFix(category) {
-  const fixes = {
-    'Hardcoded Secret': 'Move to environment variable: process.env.SECRET_NAME',
-    'SQL Injection': 'Use parameterized queries. Never interpolate user input into SQL.',
-    XSS: 'Use textContent instead of innerHTML. Sanitize with DOMPurify if HTML is needed.',
-    'Code Injection': 'Remove eval()/new Function(). Use a safe alternative or a JSON parser.',
-    'Command Injection': 'Use execFile() with an args array instead of exec() with a shell string.',
-    'Weak Crypto': 'Use crypto.createHash("sha256") or bcrypt for password hashing.',
-    'Weak Randomness':
-      'Use crypto.randomBytes(n) or crypto.randomUUID() for security-sensitive values.',
-    'Auth Bypass': 'Enforce JWT algorithm explicitly: { algorithms: ["HS256"] }',
-    'Info Disclosure': 'Remove logging of sensitive values. Use structured logging with redaction.',
-  };
-  return fixes[category] || 'Review and remediate according to OWASP guidelines.';
-}
-
-/**
- * Determine the overall verdict from a list of findings.
- * REJECTED if any critical/high. WARNING if medium. APPROVED if low/clean.
- *
- * @param {Array} findings
- * @returns {{ verdict: string, passed: boolean }}
- */
-function computeVerdict(findings) {
-  // Filter out VERIFY flags from blocking logic — they are informational
-  const blocking = findings.filter(f => f.category !== 'Verification Flag');
-  const maxSeverityRank = blocking.reduce((min, f) => {
-    const rank = SEVERITY_RANK[f.severity] ?? 3;
-    return rank < min ? rank : min;
-  }, 4); // 4 = no findings
-
-  if (maxSeverityRank <= 1) return { verdict: 'REJECTED', passed: false }; // critical or high
-  if (maxSeverityRank === 2) return { verdict: 'WARNING', passed: true }; // medium
-  return { verdict: 'APPROVED', passed: true };
-}
-
-/**
- * Build a self-healing instruction string for the Maker Agent.
- * This is what you paste back into the AI to trigger auto-correction.
- *
- * @param {Array} findings
- * @returns {string|null}
- */
-function buildSelfHealingInstructions(findings) {
-  const blocking = findings.filter(f => {
-    const rank = SEVERITY_RANK[f.severity] ?? 3;
-    return rank <= 1; // critical + high only
-  });
-
-  if (!blocking.length) return null;
-
-  const lines = [
-    '⚠️ Inner-Loop Validator found blocking issues. Auto-correct the following before writing to disk:\n',
-  ];
-
-  for (const f of blocking) {
-    lines.push(`[${f.severity.toUpperCase()}] Line ${f.line} — ${f.category}`);
-    lines.push(`  Issue: ${f.message}`);
-    lines.push(`  Fix:   ${f.fix}`);
-    lines.push('');
-  }
-
-  lines.push('Re-generate the affected lines only. Do not change unaffected code.');
-  return lines.join('\n');
-}
-
-// ── Output ────────────────────────────────────────────────────────────────
-
-function printHumanReport(result) {
-  const { verdict, issues, summary, self_healing_instructions } = result;
-
-  const verdictColor = verdict === 'APPROVED' ? GREEN : verdict === 'WARNING' ? YELLOW : RED;
-  const verdictIcon = verdict === 'APPROVED' ? '✅' : verdict === 'WARNING' ? '⚠️' : '❌';
-
-  console.error(`\n${BOLD}${CYAN}━━━ Inner-Loop Validator ━━━━━━━━━━━━━━━━━━━━━━━━━━${RESET}`);
-  console.error(`  Verdict: ${verdictColor}${BOLD}${verdictIcon} ${verdict}${RESET}`);
-  console.error(`  Summary: ${summary}`);
-
-  if (issues.length) {
-    console.error(`\n  ${BOLD}Issues found:${RESET}`);
-    for (const iss of issues) {
-      const color =
-        iss.severity === 'critical' || iss.severity === 'high'
-          ? RED
-          : iss.severity === 'medium'
-            ? YELLOW
-            : DIM;
-      console.error(
-        `    ${color}[${iss.severity.toUpperCase()}]${RESET} Line ${iss.line} — ${iss.category}`,
-      );
-      console.error(`      ${iss.message}`);
-      console.error(`      ${DIM}Fix: ${iss.fix}${RESET}`);
-    }
-  }
-
-  if (self_healing_instructions) {
-    console.error(`\n  ${YELLOW}${BOLD}Self-Healing Instructions (for Maker Agent):${RESET}`);
-    console.error(
-      self_healing_instructions
-        .split('\n')
-        .map(l => `    ${l}`)
-        .join('\n'),
-    );
-  }
-
-  console.error(`${CYAN}━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━${RESET}\n`);
-}
-
-// ── Built-in test case ────────────────────────────────────────────────────
-
-function runTestCase() {
-  console.error(`\n${BOLD}${CYAN}━━━ Inner-Loop Validator — Self-Test ━━━━━━━━━━━━━━${RESET}`);
-
-  const badCode = `
-const password = "supersecret123";
-const result = db.query("SELECT * FROM users WHERE id = " + req.params.id);
-document.getElementById('output').innerHTML = userInput;
-const token = eval(req.body.expr);
-const rand = Math.random() * 1000;
-`;
-
-  const result = validate(badCode, 'js');
-  printHumanReport(result);
-
-  const hasCritical = result.issues.some(i => i.severity === 'critical' || i.severity === 'high');
-  if (hasCritical && result.verdict === 'REJECTED') {
-    console.error(
-      `${GREEN}✅ Self-test PASSED — validator correctly identified and blocked critical issues${RESET}\n`,
-    );
-    process.exit(0);
-  } else {
-    console.error(`${RED}❌ Self-test FAILED — expected REJECTED verdict for bad code${RESET}\n`);
-    process.exit(1);
-  }
-}
-
-// ── Public API ────────────────────────────────────────────────────────────
-
-/**
- * Validate a code string. Returns a structured result object.
- * This is the primary programmatic API — call this from other scripts.
- *
- * @param {string} code      - Source code to validate
- * @param {string} [lang]    - Language hint
- * @param {object} [opts]    - Options: { timeout: number }
- * @returns {{ verdict, passed, issues, summary, self_healing_instructions }}
- */
-function validate(code, lang = 'js', _opts = {}) {
-  if (!code || typeof code !== 'string') {
-    return {
-      verdict: 'APPROVED',
-      passed: true,
-      issues: [],
-      summary: 'No code provided — skipped.',
-      self_healing_instructions: null,
-    };
-  }
-
-  const issues = scanCode(code, lang);
-
-  // Sort by severity rank
-  issues.sort((a, b) => (SEVERITY_RANK[a.severity] ?? 3) - (SEVERITY_RANK[b.severity] ?? 3));
-
-  const { verdict, passed } = computeVerdict(issues);
-  const healingInstructions = buildSelfHealingInstructions(issues);
-
-  const critCount = issues.filter(i => i.severity === 'critical').length;
-  const highCount = issues.filter(i => i.severity === 'high').length;
-  const medCount = issues.filter(i => i.severity === 'medium').length;
-  const lowCount = issues.filter(i => i.severity === 'low').length;
-  const verifyCount = issues.filter(i => i.category === 'Verification Flag').length;
-
-  let summary = `${issues.length} issue(s) found`;
-  if (!issues.length) {
-    summary = 'No issues detected — code is clean';
-  } else {
-    const parts = [];
-    if (critCount) parts.push(`${critCount} critical`);
-    if (highCount) parts.push(`${highCount} high`);
-    if (medCount) parts.push(`${medCount} medium`);
-    if (lowCount) parts.push(`${lowCount} low`);
-    if (verifyCount) parts.push(`${verifyCount} VERIFY flag(s) need human review`);
-    summary = parts.join(', ');
-  }
-
-  return {
-    verdict,
-    passed,
-    issues,
-    summary,
-    self_healing_instructions: healingInstructions,
-    meta: {
-      lines_scanned: code.split('\n').length,
-      lang,
-      timestamp: new Date().toISOString(),
-    },
-  };
-}
-
-module.exports = {
-  validate,
-  validateSnippet: validate,
-  scanCode,
-  computeVerdict,
-  buildSelfHealingInstructions,
-};
-
-// ── CLI Entry ─────────────────────────────────────────────────────────────
-
-if (require.main === module) {
-  const argv = process.argv.slice(2);
-
-  if (!argv.length || argv.includes('--help') || argv.includes('-h')) {
-    console.log(`
-${BOLD}inner_loop_validator.js${RESET} — Tribunal Self-Healing CI
-
-${BOLD}Usage:${RESET}
-  node .agent/scripts/inner_loop_validator.js --snippet "<code>"
-  node .agent/scripts/inner_loop_validator.js --file ./output.js [--lang ts]
-  node .agent/scripts/inner_loop_validator.js test-case
-
-${BOLD}Output:${RESET}
-  JSON to stdout. Human-readable summary to stderr.
-  Use --json-only to suppress the human report.
-
-${BOLD}Verdict:${RESET}
-  APPROVED   → No critical/high issues. Safe to proceed.
-  WARNING    → Medium issues found. Human should review.
-  REJECTED   → Critical/high issues. Maker Agent must self-correct.
-`);
-    process.exit(0);
-  }
-
-  // Built-in self-test
-  if (argv[0] === 'test-case') {
-    runTestCase();
-    process.exit(0);
-  }
-
-  const jsonOnly = argv.includes('--json-only');
-  const fileFlagIdx = argv.indexOf('--file');
-  const snippetIdx = argv.indexOf('--snippet');
-  const langIdx = argv.indexOf('--lang');
-
-  const lang = langIdx !== -1 && argv[langIdx + 1] ? argv[langIdx + 1] : 'js';
-
-  let code = '';
-
-  if (fileFlagIdx !== -1 && argv[fileFlagIdx + 1]) {
-    const filePath = path.resolve(argv[fileFlagIdx + 1]);
-    if (!fs.existsSync(filePath)) {
-      console.error(`${RED}✖ File not found: ${filePath}${RESET}`);
-      process.exit(1);
-    }
-    code = fs.readFileSync(filePath, 'utf8');
-  } else if (snippetIdx !== -1 && argv[snippetIdx + 1]) {
-    code = argv[snippetIdx + 1];
-  } else if (!process.stdin.isTTY) {
-    // Read from stdin if piped (cross-platform, works on Windows)
-    code = fs.readFileSync(0, 'utf8');
-  } else {
-    console.error(`${RED}✖ Provide --snippet "<code>" or --file <path>${RESET}`);
-    process.exit(1);
-  }
-
-  const result = validate(code, lang);
-
-  // Always emit JSON to stdout (for machine consumption)
-  console.log(JSON.stringify(result, null, 2));
-
-  // Emit human report to stderr (safe to suppress with 2>/dev/null)
-  if (!jsonOnly) {
-    printHumanReport(result);
-  }
-
-  // Exit code: 0 = passed (APPROVED or WARNING), 1 = REJECTED
-  process.exit(result.passed ? 0 : 1);
-}
+#!/usr/bin/env node
+/**
+ * inner_loop_validator.js — Tribunal Kit Inner-Loop Self-Healing CI
+ * ==================================================================
+ * Orchestrates security_scan.js and lint heuristics on code snippets
+ * IN MEMORY before they are presented to the Human Gate.
+ *
+ * This is the "Phase 6 Auto-Correction Engine": it feeds structured
+ * JSON findings back to the Maker Agent for autonomous self-healing
+ * without requiring user involvement.
+ *
+ * Architecture:
+ *   1. Receive a code snippet (via stdin or --snippet flag)
+ *   2. Write to a temp file within the OS temp directory
+ *   3. Run OWASP security pattern scan (using PATTERNS from security_scan.js)
+ *   4. Run lightweight syntax heuristics (no external tools required)
+ *   5. Emit structured JSON verdict for the Maker Agent to consume
+ *   6. Clean up temp file
+ *
+ * Usage:
+ *   node .agent/scripts/inner_loop_validator.js --snippet "const x = eval(input)"
+ *   node .agent/scripts/inner_loop_validator.js --file ./output.js
+ *   node .agent/scripts/inner_loop_validator.js --file ./output.js --lang ts
+ *   node .agent/scripts/inner_loop_validator.js test-case
+ *
+ * Output (JSON to stdout):
+ *   {
+ *     "verdict": "APPROVED" | "WARNING" | "REJECTED",
+ *     "passed": boolean,
+ *     "issues": [{ "severity": "critical|high|medium|low", "category": string, "line": number, "message": string, "fix": string }],
+ *     "summary": string,
+ *     "self_healing_instructions": string | null   ← fed back to Maker Agent
+ *   }
+ */
+
+'use strict';
+
+const fs = require('fs');
+const path = require('path');
+
+// ── Resolve security_scan patterns (reuse — do not duplicate) ─────────────
+const SCRIPT_DIR = __dirname;
+let SECURITY_PATTERNS = [];
+let SEVERITY_RANK = {};
+
+try {
+  const secScan = require(path.join(SCRIPT_DIR, 'security_scan.js'));
+  SECURITY_PATTERNS = secScan.PATTERNS || [];
+  SEVERITY_RANK = secScan.SEVERITY_RANK || {
+    critical: 0,
+    high: 1,
+    medium: 2,
+    low: 3,
+  };
+} catch {
+  // Fallback: minimal critical patterns only (never fail silently on missing module)
+  SECURITY_PATTERNS = [
+    [
+      /(?:password|passwd)\s*=\s*["'][^"']+["']/i,
+      'critical',
+      'Hardcoded Secret',
+      'Hardcoded password',
+    ],
+    [/\beval\s*\(/, 'high', 'Code Injection', 'eval() is a code injection vector'],
+    [/\.innerHTML\s*=/, 'high', 'XSS', 'Direct innerHTML assignment'],
+    [/algorithms\s*:\s*\[\s*["']none["']/, 'critical', 'Auth Bypass', "JWT 'none' algorithm"],
+  ];
+  SEVERITY_RANK = { critical: 0, high: 1, medium: 2, low: 3 };
+}
+
+// ── Syntax heuristics (no external deps) ─────────────────────────────────
+// These catch structural issues in generated code before a linter runs.
+const SYNTAX_HEURISTICS = [
+  {
+    pattern: /\bconst\s+\w+\s*=\s*require\s*\(\s*["'](?!\.\/|\.\.\/|[a-zA-Z])/,
+    severity: 'medium',
+    category: 'Hallucination Risk',
+    message: 'Suspicious require() path — verify module exists in package.json',
+    fix: 'Check that this package is listed in package.json dependencies',
+  },
+  {
+    pattern: /\/\/\s*VERIFY:/,
+    severity: 'low',
+    category: 'Verification Flag',
+    message: 'Maker Agent flagged this line as uncertain — human review required',
+    fix: 'The Maker Agent marked this with // VERIFY: — confirm before approving',
+  },
+  {
+    pattern: /:\s*any\b(?!\s*=)/,
+    severity: 'low',
+    category: 'Type Safety',
+    message: 'TypeScript `any` type used without explanation comment',
+    fix: 'Replace :any with a specific type, or add // any: [reason] comment',
+  },
+  {
+    pattern: /process\.env\.\w+(?!\s*\?\?|\s*\|\|)/,
+    severity: 'low',
+    category: 'Config Safety',
+    message: 'process.env access without nullish fallback — may throw at runtime',
+    fix: 'Use: process.env.VAR ?? "default" — always guard env var access',
+  },
+  {
+    pattern: /throw\s+["'`]/,
+    severity: 'low',
+    category: 'Error Quality',
+    message: 'Throwing a string instead of an Error object — stack traces will be lost',
+    fix: 'Use: throw new Error("message") instead of throw "message"',
+  },
+  {
+    pattern: /catch\s*\(\s*\w+\s*\)\s*\{?\s*\}/,
+    severity: 'medium',
+    category: 'Error Handling',
+    message: 'Empty catch block swallows errors silently',
+    fix: 'Add at minimum: catch (err) { console.error(err); throw err; }',
+  },
+  {
+    pattern: /\.then\(\s*\)\s*\.catch\s*\(|\.catch\s*\(\s*(?:[^)]*)\s*\)\s*\{\s*\}/,
+    severity: 'medium',
+    category: 'Error Handling',
+    message: 'Empty .then() or .catch() handler — Promise errors may be silenced',
+    fix: 'Implement proper resolution and rejection handlers',
+  },
+  {
+    pattern: /window\.|document\.|navigator\./,
+    severity: 'low',
+    category: 'Environment Check',
+    message: 'Browser global access — may fail in SSR/Node environments',
+    fix: 'Guard with: typeof window !== "undefined" before accessing browser globals',
+  },
+];
+
+// ── ANSI colors (centralized via _colors.js) ─────────────────────────────
+const { GREEN, YELLOW, RED, CYAN, BOLD, DIM, RESET } = require('./_colors');
+
+// ── Core scanning ─────────────────────────────────────────────────────────
+
+/**
+ * Scan a code string for security and heuristic issues.
+ * Returns an array of structured finding objects.
+ *
+ * @param {string} code      - Raw source code string
+ * @param {string} [lang]    - Language hint ('js' | 'ts' | 'py' | 'jsx' | 'tsx')
+ * @returns {Array<{severity, category, line, message, fix, source}>}
+ */
+function scanCode(code, _lang = 'js') {
+  const findings = [];
+  const lines = code.split('\n');
+
+  for (let i = 0; i < lines.length; i++) {
+    const stripped = lines[i].trim();
+    const lineNum = i + 1;
+
+    // Skip pure comments
+    if (stripped.startsWith('//') || stripped.startsWith('#') || stripped.startsWith('*')) {
+      continue;
+    }
+
+    // Run OWASP security patterns
+    for (const [pattern, severity, category, message] of SECURITY_PATTERNS) {
+      if (pattern.test(stripped)) {
+        findings.push({
+          severity,
+          category,
+          line: lineNum,
+          message,
+          fix: buildSecurityFix(category),
+          source: 'security_scan',
+        });
+      }
+    }
+
+    // Run structural heuristics
+    for (const h of SYNTAX_HEURISTICS) {
+      if (h.pattern.test(stripped)) {
+        findings.push({
+          severity: h.severity,
+          category: h.category,
+          line: lineNum,
+          message: h.message,
+          fix: h.fix,
+          source: 'heuristic',
+        });
+      }
+    }
+  }
+
+  return findings;
+}
+
+/**
+ * Build a fix suggestion for known security categories.
+ * @param {string} category
+ * @returns {string}
+ */
+function buildSecurityFix(category) {
+  const fixes = {
+    'Hardcoded Secret': 'Move to environment variable: process.env.SECRET_NAME',
+    'SQL Injection': 'Use parameterized queries. Never interpolate user input into SQL.',
+    XSS: 'Use textContent instead of innerHTML. Sanitize with DOMPurify if HTML is needed.',
+    'Code Injection': 'Remove eval()/new Function(). Use a safe alternative or a JSON parser.',
+    'Command Injection': 'Use execFile() with an args array instead of exec() with a shell string.',
+    'Weak Crypto': 'Use crypto.createHash("sha256") or bcrypt for password hashing.',
+    'Weak Randomness':
+      'Use crypto.randomBytes(n) or crypto.randomUUID() for security-sensitive values.',
+    'Auth Bypass': 'Enforce JWT algorithm explicitly: { algorithms: ["HS256"] }',
+    'Info Disclosure': 'Remove logging of sensitive values. Use structured logging with redaction.',
+  };
+  return fixes[category] || 'Review and remediate according to OWASP guidelines.';
+}
+
+/**
+ * Determine the overall verdict from a list of findings.
+ * REJECTED if any critical/high. WARNING if medium. APPROVED if low/clean.
+ *
+ * @param {Array} findings
+ * @returns {{ verdict: string, passed: boolean }}
+ */
+function computeVerdict(findings) {
+  // Filter out VERIFY flags from blocking logic — they are informational
+  const blocking = findings.filter(f => f.category !== 'Verification Flag');
+  const maxSeverityRank = blocking.reduce((min, f) => {
+    const rank = SEVERITY_RANK[f.severity] ?? 3;
+    return rank < min ? rank : min;
+  }, 4); // 4 = no findings
+
+  if (maxSeverityRank <= 1) return { verdict: 'REJECTED', passed: false }; // critical or high
+  if (maxSeverityRank === 2) return { verdict: 'WARNING', passed: true }; // medium
+  return { verdict: 'APPROVED', passed: true };
+}
+
+/**
+ * Build a self-healing instruction string for the Maker Agent.
+ * This is what you paste back into the AI to trigger auto-correction.
+ *
+ * @param {Array} findings
+ * @returns {string|null}
+ */
+function buildSelfHealingInstructions(findings) {
+  const blocking = findings.filter(f => {
+    const rank = SEVERITY_RANK[f.severity] ?? 3;
+    return rank <= 1; // critical + high only
+  });
+
+  if (!blocking.length) return null;
+
+  const lines = [
+    '⚠️ Inner-Loop Validator found blocking issues. Auto-correct the following before writing to disk:\n',
+  ];
+
+  for (const f of blocking) {
+    lines.push(`[${f.severity.toUpperCase()}] Line ${f.line} — ${f.category}`);
+    lines.push(`  Issue: ${f.message}`);
+    lines.push(`  Fix:   ${f.fix}`);
+    lines.push('');
+  }
+
+  lines.push('Re-generate the affected lines only. Do not change unaffected code.');
+  return lines.join('\n');
+}
+
+// ── Output ────────────────────────────────────────────────────────────────
+
+function printHumanReport(result) {
+  const { verdict, issues, summary, self_healing_instructions } = result;
+
+  const verdictColor = verdict === 'APPROVED' ? GREEN : verdict === 'WARNING' ? YELLOW : RED;
+  const verdictIcon = verdict === 'APPROVED' ? '✅' : verdict === 'WARNING' ? '⚠️' : '❌';
+
+  console.error(`\n${BOLD}${CYAN}━━━ Inner-Loop Validator ━━━━━━━━━━━━━━━━━━━━━━━━━━${RESET}`);
+  console.error(`  Verdict: ${verdictColor}${BOLD}${verdictIcon} ${verdict}${RESET}`);
+  console.error(`  Summary: ${summary}`);
+
+  if (issues.length) {
+    console.error(`\n  ${BOLD}Issues found:${RESET}`);
+    for (const iss of issues) {
+      const color =
+        iss.severity === 'critical' || iss.severity === 'high'
+          ? RED
+          : iss.severity === 'medium'
+            ? YELLOW
+            : DIM;
+      console.error(
+        `    ${color}[${iss.severity.toUpperCase()}]${RESET} Line ${iss.line} — ${iss.category}`,
+      );
+      console.error(`      ${iss.message}`);
+      console.error(`      ${DIM}Fix: ${iss.fix}${RESET}`);
+    }
+  }
+
+  if (self_healing_instructions) {
+    console.error(`\n  ${YELLOW}${BOLD}Self-Healing Instructions (for Maker Agent):${RESET}`);
+    console.error(
+      self_healing_instructions
+        .split('\n')
+        .map(l => `    ${l}`)
+        .join('\n'),
+    );
+  }
+
+  console.error(`${CYAN}━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━${RESET}\n`);
+}
+
+// ── Built-in test case ────────────────────────────────────────────────────
+
+function runTestCase() {
+  console.error(`\n${BOLD}${CYAN}━━━ Inner-Loop Validator — Self-Test ━━━━━━━━━━━━━━${RESET}`);
+
+  const badCode = `
+const password = "supersecret123";
+const result = db.query("SELECT * FROM users WHERE id = " + req.params.id);
+document.getElementById('output').innerHTML = userInput;
+const token = eval(req.body.expr);
+const rand = Math.random() * 1000;
+`;
+
+  const result = validate(badCode, 'js');
+  printHumanReport(result);
+
+  const hasCritical = result.issues.some(i => i.severity === 'critical' || i.severity === 'high');
+  if (hasCritical && result.verdict === 'REJECTED') {
+    console.error(
+      `${GREEN}✅ Self-test PASSED — validator correctly identified and blocked critical issues${RESET}\n`,
+    );
+    process.exit(0);
+  } else {
+    console.error(`${RED}❌ Self-test FAILED — expected REJECTED verdict for bad code${RESET}\n`);
+    process.exit(1);
+  }
+}
+
+// ── Public API ────────────────────────────────────────────────────────────
+
+/**
+ * Validate a code string. Returns a structured result object.
+ * This is the primary programmatic API — call this from other scripts.
+ *
+ * @param {string} code      - Source code to validate
+ * @param {string} [lang]    - Language hint
+ * @param {object} [opts]    - Options: { timeout: number }
+ * @returns {{ verdict, passed, issues, summary, self_healing_instructions }}
+ */
+function validate(code, lang = 'js', _opts = {}) {
+  if (!code || typeof code !== 'string') {
+    return {
+      verdict: 'APPROVED',
+      passed: true,
+      issues: [],
+      summary: 'No code provided — skipped.',
+      self_healing_instructions: null,
+    };
+  }
+
+  const issues = scanCode(code, lang);
+
+  // Sort by severity rank
+  issues.sort((a, b) => (SEVERITY_RANK[a.severity] ?? 3) - (SEVERITY_RANK[b.severity] ?? 3));
+
+  const { verdict, passed } = computeVerdict(issues);
+  const healingInstructions = buildSelfHealingInstructions(issues);
+
+  const critCount = issues.filter(i => i.severity === 'critical').length;
+  const highCount = issues.filter(i => i.severity === 'high').length;
+  const medCount = issues.filter(i => i.severity === 'medium').length;
+  const lowCount = issues.filter(i => i.severity === 'low').length;
+  const verifyCount = issues.filter(i => i.category === 'Verification Flag').length;
+
+  let summary = `${issues.length} issue(s) found`;
+  if (!issues.length) {
+    summary = 'No issues detected — code is clean';
+  } else {
+    const parts = [];
+    if (critCount) parts.push(`${critCount} critical`);
+    if (highCount) parts.push(`${highCount} high`);
+    if (medCount) parts.push(`${medCount} medium`);
+    if (lowCount) parts.push(`${lowCount} low`);
+    if (verifyCount) parts.push(`${verifyCount} VERIFY flag(s) need human review`);
+    summary = parts.join(', ');
+  }
+
+  return {
+    verdict,
+    passed,
+    issues,
+    summary,
+    self_healing_instructions: healingInstructions,
+    meta: {
+      lines_scanned: code.split('\n').length,
+      lang,
+      timestamp: new Date().toISOString(),
+    },
+  };
+}
+
+module.exports = {
+  validate,
+  validateSnippet: validate,
+  scanCode,
+  computeVerdict,
+  buildSelfHealingInstructions,
+};
+
+// ── CLI Entry ─────────────────────────────────────────────────────────────
+
+if (require.main === module) {
+  const argv = process.argv.slice(2);
+
+  if (!argv.length || argv.includes('--help') || argv.includes('-h')) {
+    console.log(`
+${BOLD}inner_loop_validator.js${RESET} — Tribunal Self-Healing CI
+
+${BOLD}Usage:${RESET}
+  node .agent/scripts/inner_loop_validator.js --snippet "<code>"
+  node .agent/scripts/inner_loop_validator.js --file ./output.js [--lang ts]
+  node .agent/scripts/inner_loop_validator.js test-case
+
+${BOLD}Output:${RESET}
+  JSON to stdout. Human-readable summary to stderr.
+  Use --json-only to suppress the human report.
+
+${BOLD}Verdict:${RESET}
+  APPROVED   → No critical/high issues. Safe to proceed.
+  WARNING    → Medium issues found. Human should review.
+  REJECTED   → Critical/high issues. Maker Agent must self-correct.
+`);
+    process.exit(0);
+  }
+
+  // Built-in self-test
+  if (argv[0] === 'test-case') {
+    runTestCase();
+    process.exit(0);
+  }
+
+  const jsonOnly = argv.includes('--json-only');
+  const fileFlagIdx = argv.indexOf('--file');
+  const snippetIdx = argv.indexOf('--snippet');
+  const langIdx = argv.indexOf('--lang');
+
+  const lang = langIdx !== -1 && argv[langIdx + 1] ? argv[langIdx + 1] : 'js';
+
+  let code = '';
+
+  if (fileFlagIdx !== -1 && argv[fileFlagIdx + 1]) {
+    const filePath = path.resolve(argv[fileFlagIdx + 1]);
+    if (!fs.existsSync(filePath)) {
+      console.error(`${RED}✖ File not found: ${filePath}${RESET}`);
+      process.exit(1);
+    }
+    code = fs.readFileSync(filePath, 'utf8');
+  } else if (snippetIdx !== -1 && argv[snippetIdx + 1]) {
+    code = argv[snippetIdx + 1];
+  } else if (!process.stdin.isTTY) {
+    // Read from stdin if piped (cross-platform, works on Windows)
+    code = fs.readFileSync(0, 'utf8');
+  } else {
+    console.error(`${RED}✖ Provide --snippet "<code>" or --file <path>${RESET}`);
+    process.exit(1);
+  }
+
+  const result = validate(code, lang);
+
+  // Always emit JSON to stdout (for machine consumption)
+  console.log(JSON.stringify(result, null, 2));
+
+  // Emit human report to stderr (safe to suppress with 2>/dev/null)
+  if (!jsonOnly) {
+    printHumanReport(result);
+  }
+
+  // Exit code: 0 = passed (APPROVED or WARNING), 1 = REJECTED
+  process.exit(result.passed ? 0 : 1);
+}