
sanity-anti-hallucination-skill
⚡ SANITY — Anti-Hallucination Supremacy Protocol for Claude Code. Zero tolerance for confabulation.
Install with your AI
Paste into Claude Code, Cursor, or any agent — it reads the repo and wires the tool into your project.
Install and set up sanity-anti-hallucination-skill (git-clone project) into my current project. Found on https://claudeers.com/sanity-anti-hallucination-skill Repo: https://github.com/verycoolapps/sanity-anti-hallucination-skill Homepage/docs: — Detected install method: git-clone → git clone https://github.com/verycoolapps/sanity-anti-hallucination-skill Category: security. Platforms: cli, api, web, mobile. Read the repo's README for exact setup and env vars, then install it and wire it into my project. Claudeers Health Verdict: active; community-verified: false. Confirm the source before running anything.
git clone https://github.com/verycoolapps/sanity-anti-hallucination-skill
// compatibility
| Platforms | cli, api, web, mobile |
|---|---|
| Operating systems | — |
| AI compatibility | claude |
| License | MIT |
| Pricing | open-source |
| Language | — |
░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
░ ░
░ ██████╗ █████╗ ███╗ ██╗██╗████████╗██╗ ██╗ ██╗ ██╗ ░
░ ██╔════╝ ██╔══██╗████╗ ██║██║╚══██╔══╝╚██╗ ██╔╝ ██║ ██║ ░
░ ╚█████╗ ███████║██╔██╗██║██║ ██║ ╚████╔╝ ╚██╗ ██╔╝ ░
░ ╚═══██╗ ██╔══██║██║╚████║██║ ██║ ╚██╔╝ ╚████╔╝ ░
░ ██████╔╝ ██║ ██║██║ ╚███║██║ ██║ ██║ ╚███╔╝ ░
░ ╚═════╝ ╚═╝ ╚═╝╚═╝ ╚══╝╚═╝ ╚═╝ ╚═╝ ╚══╝ ░
░ ░
░ ⚡ ANTI-HALLUCINATION × SECURITY SUPREMACY ⚡ ░
░ ───────────────────────────────────────── ░
░ The AI that uses SANITY never fabricates. ░
░ It reads. It verifies. It secures. Then ships. ░
░ ░
░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
"The most dangerous AI response is not the wrong answer.
It's the confidently delivered wrong answer with a SQL injection inside it."
🇮🇩 Bahasa Indonesia ── ✦ ── 🇺🇸 English
┌─────────────────────────────────────────────────────────────────────────────┐
│ DAFTAR ISI / TABLE OF CONTENTS │
├────────────────────────────────┬────────────────────────────────────────────┤
│ 🇮🇩 BAHASA INDONESIA │ 🇺🇸 ENGLISH │
│ ├─ Apa itu SANITY? │ ├─ What is SANITY? │
│ ├─ Mengapa Security #1? │ ├─ Why Security is #1 │
│ ├─ FORTRESS Framework │ ├─ FORTRESS Framework │
│ ├─ Jenis Hallucination │ ├─ Types of Hallucination │
│ ├─ Confidence Matrix │ ├─ Confidence Matrix │
│ ├─ Security Playbooks │ ├─ Security Playbooks │
│ ├─ OWASP Top 10 │ ├─ OWASP Top 10 │
│ ├─ Anti-Pattern Blacklist │ ├─ Anti-Pattern Blacklist │
│ ├─ Quick Install │ ├─ Quick Install │
│ ├─ SANITY Score │ ├─ SANITY Score │
│ └─ Mantras │ └─ Mantras │
└────────────────────────────────┴────────────────────────────────────────────┘
🇮🇩 BAHASA INDONESIA
╭━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╮
┃ ✦ APA ITU SANITY? ✦ ┃
╰━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╯
SANITY v3.0 adalah sebuah Claude Code Skill yang bekerja seperti seorang senior engineer paranoid yang tahu cara kerja hacker — ia tidak pernah klaim sesuatu yang belum diverifikasi, dan tidak pernah menulis kode yang bisa dieksploitasi.
SANITY beroperasi pada dua sumbu sekaligus, keduanya wajib lolos:
╔══════════════════════════════════════════════════════════════════╗
║ ║
║ SUMBU 1 ⟶ INTEGRITAS EPISTEMIK ║
║ ⟶ "Jangan klaim apa yang belum diverifikasi." ║
║ ║
║ ┌───────┐ ┌──────────┐ ┌────────┐ ┌────────┐ ║
║ │ READ │──▶│ EVIDENCE │──▶│ ASSUME │──▶│ LIMITS │──▶ ✓ ║
║ └───────┘ └──────────┘ └────────┘ └────────┘ ║
║ └──▶ ERROR-CHECK ──▶ ✓ ║
║ ║
╠══════════════════════════════════════════════════════════════════╣
║ ║
║ SUMBU 2 ⟶ INTEGRITAS KEAMANAN ★ PRIORITAS #1 ★ ║
║ ⟶ "Jangan hasilkan kode yang bisa dieksploitasi." ║
║ ║
║ ┌───────┐ ┌───────┐ ┌───────┐ ┌────────┐ ║
║ │THREAT │──▶│ AUDIT │──▶│ OWASP │──▶│HARDEN │──▶ ✓ ║
║ └───────┘ └───────┘ └───────┘ └────────┘ ║
║ └──▶ CERTIFY ──▶ ✓ ║
║ ║
╚══════════════════════════════════════════════════════════════════╝
╭━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╮
┃ ⚠ MENGAPA SECURITY JADI PRIORITAS #1? ⚠ ┃
╰━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╯
| Masalah | Dampak | Pemulihan |
|---|---|---|
| 👻 Fungsi yang dihalusinasi | Debugging 2 jam | Temukan bug, fix, lanjut |
| 💣 SQL Injection di production | Seluruh database bocor | Notif regulator, konsekuensi hukum, reputasi hancur |
| 🔑 API key hardcoded di repo | Credential compromised | Rotate semua key, audit akses, investigasi breach |
| 🔓 Missing auth di endpoint | Data user diakses sembarang orang | Data breach, GDPR/UU PDP violation |
💡 Asimetri biaya ini menuntut keamanan diperlakukan sebagai kekhawatiran tertinggi. Hallucination membuang waktu. Security breach menghancurkan bisnis.
╭━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╮
┃ 👻 JENIS HALLUCINATION YANG SANITY ELIMINASI 👻 ┃
╰━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╯
┌─────────────────────────────┬──────────────────────────────┬────────────────────────┐
│ TIPE │ CONTOH │ DAMPAK │
├─────────────────────────────┼──────────────────────────────┼────────────────────────┤
│ 👻 Phantom Function │ flutter.setStyle() (∄) │ Debugging berjam-jam │
│ 💀 Ghost File │ Klaim isi file tanpa baca │ Implementasi keliru │
│ 🕰️ Version Fraud │ "React 22 mendukung..." (∄) │ Arsitektur salah │
│ 🎭 False Certainty │ "Pasti masalah database" │ Waktu debugging buang │
│ 📝 Citation Fabrication │ Link docs palsu yang dibuat │ Trust collapse total │
│ 💣 SQL Injection │ String concat ke SQL query │ Full DB compromise │
│ 🔑 Hardcoded Secret │ apiKey = "sk_live_abc..." │ Instant breach │
│ 🔓 Missing Auth │ Endpoint tanpa auth check │ Unauthorized access │
│ 💉 XSS Generation │ innerHTML = userInput │ Account takeover │
│ 📂 Path Traversal │ File path dari user input │ Arbitrary file read │
└─────────────────────────────┴──────────────────────────────┴────────────────────────┘
╭━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╮
┃ 🏰 FORTRESS FRAMEWORK — 10 PILAR 🏰 ┃
╰━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╯
⟵ SUMBU 1: INTEGRITAS EPISTEMIK ⟶
╔═══╦══════════════════╦══════════════════════════════════════════════════════╗
║ # ║ PILAR ║ ATURAN ║
╠═══╬══════════════════╬══════════════════════════════════════════════════════╣
║ 1 ║ 📖 READ ║ Baca file sebelum klaim isinya. Grep sebelum ║
║ ║ ║ klaim fungsi ada. Run sebelum klaim test pass. ║
╠═══╬══════════════════╬══════════════════════════════════════════════════════╣
║ 2 ║ 🔬 EVIDENCE ║ Zero-evidence claim = hallucination. Setiap ║
║ ║ ║ klaim faktual wajib ada sumber yang bisa diverif. ║
╠═══╬══════════════════╬══════════════════════════════════════════════════════╣
║ 3 ║ 🏷️ ASSUME ║ Asumsi yang tidak ditandai = kebohongan tertunda. ║
║ ║ ║ Label [ASUMSI] / [PERLU VERIFIKASI] wajib dipakai. ║
╠═══╬══════════════════╬══════════════════════════════════════════════════════╣
║ 4 ║ 🚧 LIMITS ║ "Saya tidak tahu" > jawaban salah yang percaya diri.║
║ ║ ║ Nyatakan knowledge cutoff dan batas pengetahuan. ║
╠═══╬══════════════════╬══════════════════════════════════════════════════════╣
║ 5 ║ ✅ ERROR-CHECK ║ Baca ulang respons sebelum deliver. Cari kontradiksi║
║ ║ ║ internal, inkonsistensi nama, versi yang berbeda. ║
╚═══╩══════════════════╩══════════════════════════════════════════════════════╝
⟵ SUMBU 2: INTEGRITAS KEAMANAN ★ PRIORITAS #1 ★ ⟶
╔═══╦══════════════════╦══════════════════════════════════════════════════════╗
║ # ║ PILAR ║ ATURAN ║
╠═══╬══════════════════╬══════════════════════════════════════════════════════╣
║ 6 ║ 🗺️ THREAT-MODEL ║ Sebelum tulis kode sensitif: identifikasi aktor, ║
║ ║ ║ trust boundary, data flow, attack surface. ║
╠═══╬══════════════════╬══════════════════════════════════════════════════════╣
║ 7 ║ 🔍 AUDIT ║ Checklist 14-poin security audit untuk setiap ║
║ ║ ║ snippet kode: injection, auth, authz, secrets... ║
╠═══╬══════════════════╬══════════════════════════════════════════════════════╣
║ 8 ║ 🛡️ OWASP ║ Hard block jika ada pelanggaran OWASP Top 10. ║
║ ║ ║ Tidak ada pengecualian. Tidak ada "nanti aja". ║
╠═══╬══════════════════╬══════════════════════════════════════════════════════╣
║ 9 ║ 🔩 HARDEN ║ Defense in depth: input validation, auth tokens, ║
║ ║ ║ data encryption, API security, dependency audit. ║
╠═══╬══════════════════╬══════════════════════════════════════════════════════╣
║10 ║ 🏅 CERTIFY ║ Security Cert wajib sebelum deliver kode sensitif. ║
║ ║ ║ Atau Security Caveat jika ada unresolved issues. ║
╚═══╩══════════════════╩══════════════════════════════════════════════════════╝
╭━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╮
┃ 🎯 CONFIDENCE MATRIX — SISTEM RATING DUA SUMBU 🎯 ┃
╰━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╯
INTEGRITAS EPISTEMIK INTEGRITAS KEAMANAN
───────────────────────────── ─────────────────────────────
🟢 CONFIRMED ← Diverifikasi langsung 🛡️ HARDENED ← Full audit passed
🟡 INFERRED ← Kesimpulan dari bukti ⚡ REVIEWED ← Checks passed
🔴 UNCERTAIN ← Spekulasi / stale 🔸 CAVEAT ← Perlu review dulu
⚫ UNKNOWN ← Di luar kemampuan saya 🚨 BLOCKED ← JANGAN DIPAKAI
Contoh gabungan dalam respons nyata:
🟢 CONFIRMED + 🛡️ HARDENED
→ "Kode ini ada di src/auth.ts:89. Full security audit passed."
🟡 INFERRED + 🔸 CAVEAT
→ "Pattern ini kemungkinan benar, tapi saya sarankan pen-test sebelum production."
🔴 UNCERTAIN + 🚨 BLOCKED
→ "Saya tidak yakin ini benar DAN ada potensi SQL injection di baris 34.
Jangan digunakan sampai diverifikasi dan diperbaiki."
╭━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╮
┃ 🛡️ SECURITY PLAYBOOKS — PROTOKOL PER DOMAIN 🛡️ ┃
╰━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╯
🗄️ Database Security — Klik untuk lihat
- SALAH (SQL Injection):
- query = "SELECT * FROM users WHERE id = " + userId
- db.query("SELECT * FROM users WHERE name='" + req.body.name + "'")
+ BENAR (Parameterized Query):
+ query = "SELECT * FROM users WHERE id = ?"
+ db.query(query, [userId])
+ // ORM: User.where(id: userId)
Aturan Absolut:
- ✅ SELALU gunakan parameterized query / prepared statement
- ❌ JANGAN PERNAH string-concat user input ke SQL
- ✅ Principle of least privilege untuk DB user
- ✅ Validate dan encode output untuk display
🔑 Secret Management — Klik untuk lihat
- SALAH (Hardcoded Secret — INSTANT BREACH):
- const API_KEY = "sk_live_abc123xyz456..."
- const DB_PASSWORD = "mySecretPass123"
- const JWT_SECRET = "super_secret_key"
+ BENAR (Environment Variables):
+ const API_KEY = process.env.API_KEY
+ const DB_PASSWORD = process.env.DB_PASSWORD
+ const JWT_SECRET = process.env.JWT_SECRET
+ if (!process.env.JWT_SECRET) throw new Error("JWT_SECRET not configured")
Jika Terdeteksi Secret di Kode:
⚠️ SECURITY ALERT — Hardcoded credential terdeteksi!
1. ROTATE credential ini SEKARANG — anggap sudah compromised
2. Remove dari source code
3. Hapus dari git history (BFG Repo Cleaner / git filter-repo)
4. Audit semua akses yang mungkin menggunakan credential ini
🔐 Authentication & Authorization — Klik untuk lihat
- SALAH (IDOR + Missing Auth):
- app.get('/api/documents/:id', async (req, res) => {
- const doc = await Document.findByPk(req.params.id)
- res.json(doc) // Siapapun bisa akses dokumen siapapun!
- })
+ BENAR (Auth + Ownership Check):
+ app.get('/api/documents/:id', requireAuth, async (req, res) => {
+ const doc = await Document.findOne({
+ where: { id: req.params.id, userId: req.user.id }
+ })
+ if (!doc) return res.status(404).json({ error: 'Not found' })
+ res.json(doc)
+ })
🌐 Input Validation & XSS Prevention — Klik untuk lihat
- SALAH (XSS):
- element.innerHTML = userInput
- document.write(req.query.name)
- res.send(`<h1>Hello ${req.query.name}</h1>`)
+ BENAR:
+ element.textContent = userInput // textContent, bukan innerHTML
+ res.send(escapeHtml(`Hello ${req.query.name}`))
+ // Atau gunakan framework yang auto-escape (React, Vue, Angular)
Checklist Input Validation:
- ✅ Whitelist approach — define apa yang BOLEH, reject sisanya
- ✅ Validate: type + length + format + range
- ✅ Normalize unicode sebelum comparison
- ✅ Reject null bytes dalam string
- ✅ Limit request body / file size
⚡ Rate Limiting — Klik untuk lihat
- SALAH (Brute Force Playground):
- app.post('/api/auth/login', async (req, res) => { ... })
+ BENAR (Protected):
+ const loginLimiter = rateLimit({
+ windowMs: 15 * 60 * 1000, // 15 menit
+ max: 5, // max 5 percobaan
+ skipSuccessfulRequests: true,
+ message: 'Too many login attempts. Please wait 15 minutes.'
+ })
+ app.post('/api/auth/login', loginLimiter, async (req, res) => { ... })
╭━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╮
┃ 🔒 OWASP TOP 10 — ZERO TOLERANCE LIST 🔒 ┃
╰━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╯
╔══════╦═══════════════════════════════╦══════════════════════════════════════╗
║ ID ║ KERENTANAN ║ MITIGASI WAJIB ║
╠══════╬═══════════════════════════════╬══════════════════════════════════════╣
║ A01 ║ 🚪 Broken Access Control ║ Server-side auth + deny-by-default ║
║ A02 ║ 🔑 Cryptographic Failures ║ Env vars, bcrypt/Argon2, TLS 1.3 ║
║ A03 ║ 💉 Injection [PALING UMUM] ║ Parameterized queries + allowlists ║
║ A04 ║ 🏗️ Insecure Design ║ Rate limiting + server-side BL ║
║ A05 ║ ⚙️ Security Misconfiguration ║ No debug in prod + security headers ║
║ A06 ║ 📦 Vulnerable Components ║ No known-CVE packages, pin versions ║
║ A07 ║ 🪪 Auth Failures ║ Session mgmt, MFA, lockout policy ║
║ A08 ║ 🧩 Data Integrity Failures ║ SRI + no untrusted deserialization ║
║ A09 ║ 📋 Logging Failures ║ Log auth events, mask PII in logs ║
║ A10 ║ 🌐 SSRF ║ Whitelist URLs + block private IPs ║
╚══════╩═══════════════════════════════╩══════════════════════════════════════╝
╭━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╮
┃ 🚫 ANTI-PATTERN BLACKLIST 🚫 ┃
╰━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╯
EPISTEMIK — Pola Hallucination:
| ❌ Anti-Pattern | 🔴 Dampak |
|---|---|
| 👻 Phantom Function Syndrome | Debugging di fungsi yang tidak pernah ada |
| 💀 Ghost File Hallucination | Implementasi berdasarkan file yang tidak dibaca |
| 🕰️ Version Confidence Fraud | Arsitektur berdasarkan versi yang tidak ada |
| 🎭 False Certainty Masking | Waktu debugging terbuang di tempat yang salah |
| ✍️ Confident Completion Fabrication | Silent bugs dari kode yang "lengkap" tapi palsu |
| 📰 Recency Abuse | "Baru-baru ini" dari perspektif model ≠ perspektif user |
| 🔄 Self-Contradiction Blindness | Inkonsistensi nama, versi, logika dalam satu respons |
SECURITY — Pola Kerentanan:
| ❌ Anti-Pattern | 🔴 Dampak |
|---|---|
| 💣 SQL Injection Opening | Full database compromise |
| 🔑 Hardcoded Secrets | Instant credential breach |
| 🔓 Missing Authentication | Unauthorized access ke semua data |
| ⌨️ Raw User Input in Shell | Remote Code Execution (RCE) |
| 📂 Path Traversal | Arbitrary file read / write |
| 🧂 Weak Password Hashing | Password database brute-forceable |
| 💉 XSS via innerHTML | Account takeover + data exfiltration |
| 🪪 IDOR Vulnerability | Akses data user lain |
| 🌊 Missing Rate Limiting | Brute force attack playground |
| 🗣️ Info Disclosure via Errors | Stack trace memberi peta untuk attacker |
╭━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╮
┃ ⚡ QUICK INSTALL ⚡ ┃
╰━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╯
Cara 1 — Via Claude Code CLI (Paling Cepat):
claude skill install https://github.com/verycoolapps/sanity-anti-hallucination-skill
Cara 2 — Manual Copy:
cp sanity.md ~/.claude/skills/sanity.md
Cara 3 — Invoke Langsung di Sesi:
/sanity
Cara 4 — Auto-Load via CLAUDE.md (Best Practice):
## Skills
- Always load: sanity
## AI Behavior
- All claims must be verified before delivery (FORTRESS Framework)
- OWASP Top 10 compliance is mandatory, no exceptions
- No hardcoded secrets, credentials, or keys ever
- Security certification required on sensitive code
╭━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╮
┃ 📊 SANITY SCORE — SISTEM PENILAIAN DUA SUMBU 📊 ┃
╰━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╯
┌─────────────────────────────────────────────────────────────────────┐
│ │
│ E_SCORE = (verified + cited + flagged + limits - contradictions) │
│ ────────────────────────────────────────────── │
│ total_claims │
│ │
│ S_SCORE = (owasp + validations + auth + injections + certs) │
│ ────────────────────────────────────────────── │
│ max_possible │
│ │
│ ⚠️ ANY hardcoded secret → S_SCORE = 0.00 (OTOMATIS GAGAL) │
│ │
│ COMBINED = (E_SCORE × 0.4) + (S_SCORE × 0.6) │
│ │
└─────────────────────────────────────────────────────────────────────┘
┌────────────┬──────────────────┬──────────────────────────────────────┐
│ SCORE │ GRADE │ MEANING │
├────────────┼──────────────────┼──────────────────────────────────────┤
│ ≥ 0.95 │ 🏆 MASTER │ Production-grade, world-class │
│ ≥ 0.85 │ ✅ PASSED │ Solid, safe for production │
│ ≥ 0.70 │ ⚠️ WARNING │ Review sebelum deploy │
│ < 0.70 │ ❌ FAILED │ Revisi penuh diperlukan │
│ S = 0.00 │ 🚨 SEC. FAIL │ Block semua delivery │
└────────────┴──────────────────┴──────────────────────────────────────┘
╭━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╮
┃ 🎓 SANITY MANTRAS 🎓 ┃
╰━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╯
┌─────────────────────────────────────────────────────────────────────┐
│ EPISTEMIK │
│ ───────── │
│ "If I haven't read it, I don't know what's in it." │
│ "If I haven't run it, I don't know if it works." │
│ "Confidence without evidence is confabulation." │
│ "A wrong answer delivered confidently causes more damage │
│ than 'I don't know.'" │
│ "When in doubt — READ, GREP, RUN. Then claim." │
├─────────────────────────────────────────────────────────────────────┤
│ KEAMANAN │
│ ──────── │
│ "No input is safe until validated. No exception." │
│ "Secrets in code are secrets compromised." │
│ "Every endpoint without auth is an open door." │
│ "Defense in depth: never rely on a single security layer." │
│ "Security is not a feature. It is the foundation." │
│ "Ship fast — but never ship exploitable." │
└─────────────────────────────────────────────────────────────────────┘
🇺🇸 ENGLISH
╭━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╮
┃ ✦ WHAT IS SANITY? ✦ ┃
╰━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╯
SANITY v3.0 is a Claude Code Skill — think of it as a paranoid senior engineer who also knows how attackers think. It never claims something unverified, and never writes code that can be exploited.
It operates on two non-negotiable axes — both must pass, no exceptions:
╔══════════════════════════════════════════════════════════════════╗
║ ║
║ AXIS 1 ⟶ EPISTEMIC INTEGRITY ║
║ ⟶ "Never claim what you haven't verified." ║
║ ║
║ ┌───────┐ ┌──────────┐ ┌────────┐ ┌────────┐ ║
║ │ READ │──▶│ EVIDENCE │──▶│ ASSUME │──▶│ LIMITS │──▶ ✓ ║
║ └───────┘ └──────────┘ └────────┘ └────────┘ ║
║ └──▶ ERROR-CHECK ──▶ ✓ ║
║ ║
╠══════════════════════════════════════════════════════════════════╣
║ ║
║ AXIS 2 ⟶ SECURITY INTEGRITY ★ PRIORITY #1 ★ ║
║ ⟶ "Never produce code with exploitable flaws." ║
║ ║
║ ┌───────┐ ┌───────┐ ┌───────┐ ┌────────┐ ║
║ │THREAT │──▶│ AUDIT │──▶│ OWASP │──▶│HARDEN │──▶ ✓ ║
║ └───────┘ └───────┘ └───────┘ └────────┘ ║
║ └──▶ CERTIFY ──▶ ✓ ║
║ ║
╚══════════════════════════════════════════════════════════════════╝
╭━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╮
┃ ⚠ WHY IS SECURITY PRIORITY #1? ⚠ ┃
╰━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╯
| Problem | Impact | Recovery |
|---|---|---|
| 👻 Hallucinated function | 2 hours of debugging | Find bug, fix it, move on |
| 💣 SQL Injection in production | Entire database leaked | Regulator notification, legal consequences, reputation destroyed |
| 🔑 Hardcoded API key in repo | Credential compromised | Rotate all keys, audit access, investigate breach |
| 🔓 Missing auth on endpoint | Any user accesses all data | Data breach, GDPR/compliance violation |
💡 This cost asymmetry demands security be treated as the supreme concern. Hallucination wastes time. A security breach destroys the business.
╭━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╮
┃ 👻 TYPES OF HALLUCINATION SANITY ELIMINATES 👻 ┃
╰━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╯
┌─────────────────────────────┬──────────────────────────────┬────────────────────────┐
│ TYPE │ EXAMPLE │ IMPACT │
├─────────────────────────────┼──────────────────────────────┼────────────────────────┤
│ 👻 Phantom Function │ flutter.setStyle() (∄) │ Hours of debugging │
│ 💀 Ghost File │ Claiming file contents │ Wrong implementation │
│ 🕰️ Version Fraud │ "React 22 supports..." (∄) │ Wrong architecture │
│ 🎭 False Certainty │ "It's definitely the DB" │ Debug the wrong place │
│ 📝 Citation Fabrication │ Invented docs links │ Trust collapse │
│ 💣 SQL Injection │ User input concat'd to SQL │ Full DB compromise │
│ 🔑 Hardcoded Secret │ apiKey = "sk_live_abc..." │ Instant breach │
│ 🔓 Missing Auth │ Endpoint with no auth check │ Unauthorized access │
│ 💉 XSS Generation │ innerHTML = userInput │ Account takeover │
│ 📂 Path Traversal │ User-supplied file path │ Arbitrary file read │
└─────────────────────────────┴──────────────────────────────┴────────────────────────┘
╭━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╮
┃ 🏰 FORTRESS FRAMEWORK — 10 PILLARS 🏰 ┃
╰━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╯
⟵ AXIS 1: EPISTEMIC INTEGRITY ⟶
╔═══╦══════════════════╦══════════════════════════════════════════════════════╗
║ # ║ PILLAR ║ RULE ║
╠═══╬══════════════════╬══════════════════════════════════════════════════════╣
║ 1 ║ 📖 READ ║ Read files before claiming contents. Grep before ║
║ ║ ║ claiming functions exist. Run before claiming tests. ║
╠═══╬══════════════════╬══════════════════════════════════════════════════════╣
║ 2 ║ 🔬 EVIDENCE ║ Zero-evidence claim = hallucination. Every factual ║
║ ║ ║ claim must have a verifiable, citable source. ║
╠═══╬══════════════════╬══════════════════════════════════════════════════════╣
║ 3 ║ 🏷️ ASSUME ║ Unlabeled assumptions are delayed lies. Use ║
║ ║ ║ [ASSUMPTION] / [NEEDS VERIFICATION] labels always. ║
╠═══╬══════════════════╬══════════════════════════════════════════════════════╣
║ 4 ║ 🚧 LIMITS ║ "I don't know" > confidently wrong. State knowledge ║
║ ║ ║ cutoff and gaps clearly and proactively. ║
╠═══╬══════════════════╬══════════════════════════════════════════════════════╣
║ 5 ║ ✅ ERROR-CHECK ║ Re-read the response before delivery. Look for ║
║ ║ ║ internal contradictions, name inconsistencies. ║
╚═══╩══════════════════╩══════════════════════════════════════════════════════╝
⟵ AXIS 2: SECURITY INTEGRITY ★ PRIORITY #1 ★ ⟶
╔═══╦══════════════════╦══════════════════════════════════════════════════════╗
║ # ║ PILLAR ║ RULE ║
╠═══╬══════════════════╬══════════════════════════════════════════════════════╣
║ 6 ║ 🗺️ THREAT-MODEL ║ Before any sensitive code: identify actors, trust ║
║ ║ ║ boundaries, data flow, and attack surface. ║
╠═══╬══════════════════╬══════════════════════════════════════════════════════╣
║ 7 ║ 🔍 AUDIT ║ 14-point security audit checklist on every code ║
║ ║ ║ snippet: injection, auth, authz, secrets, crypto... ║
╠═══╬══════════════════╬══════════════════════════════════════════════════════╣
║ 8 ║ 🛡️ OWASP ║ Hard block on any OWASP Top 10 violation. No ║
║ ║ ║ exceptions. No "fix it later." No workarounds. ║
╠═══╬══════════════════╬══════════════════════════════════════════════════════╣
║ 9 ║ 🔩 HARDEN ║ Defense in depth: input validation, secure tokens, ║
║ ║ ║ data encryption, API security, dependency audit. ║
╠═══╬══════════════════╬══════════════════════════════════════════════════════╣
║10 ║ 🏅 CERTIFY ║ Explicit Security Certification required before ║
║ ║ ║ delivering any security-sensitive code. ║
╚═══╩══════════════════╩══════════════════════════════════════════════════════╝
╭━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╮
┃ 🎯 CONFIDENCE MATRIX — DUAL-AXIS RATING SYSTEM 🎯 ┃
╰━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╯
EPISTEMIC CONFIDENCE SECURITY CONFIDENCE
───────────────────────────── ─────────────────────────────
🟢 CONFIRMED ← Directly verified 🛡️ HARDENED ← Full audit passed
🟡 INFERRED ← Logical conclusion ⚡ REVIEWED ← Standard checks pass
🔴 UNCERTAIN ← Speculation / stale 🔸 CAVEAT ← Review before prod
⚫ UNKNOWN ← Beyond my knowledge 🚨 BLOCKED ← DO NOT USE
Combined usage examples:
🟢 CONFIRMED + 🛡️ HARDENED
→ "This code is at src/auth.ts:89. Full security audit passed."
🟡 INFERRED + 🔸 CAVEAT
→ "This pattern is likely correct, but I recommend pen-testing before production."
🔴 UNCERTAIN + 🚨 BLOCKED
→ "I'm not confident this is correct AND there's a potential SQL injection at line 34.
Do not use until verified and patched."
╭━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╮
┃ 🔒 OWASP TOP 10 — ZERO TOLERANCE LIST 🔒 ┃
╰━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╯
╔══════╦═══════════════════════════════╦══════════════════════════════════════╗
║ ID ║ VULNERABILITY ║ MANDATORY MITIGATION ║
╠══════╬═══════════════════════════════╬══════════════════════════════════════╣
║ A01 ║ 🚪 Broken Access Control ║ Server-side auth + deny-by-default ║
║ A02 ║ 🔑 Cryptographic Failures ║ Env vars, bcrypt/Argon2, TLS 1.3 ║
║ A03 ║ 💉 Injection [MOST COMMON] ║ Parameterized queries + allowlists ║
║ A04 ║ 🏗️ Insecure Design ║ Rate limiting + server-side logic ║
║ A05 ║ ⚙️ Security Misconfiguration ║ No debug in prod + security headers ║
║ A06 ║ 📦 Vulnerable Components ║ No known-CVE packages, pin versions ║
║ A07 ║ 🪪 Auth Failures ║ Session mgmt, MFA, lockout policy ║
║ A08 ║ 🧩 Data Integrity Failures ║ SRI + no untrusted deserialization ║
║ A09 ║ 📋 Logging Failures ║ Log auth events, mask PII in logs ║
║ A10 ║ 🌐 SSRF ║ Whitelist URLs + block private IPs ║
╚══════╩═══════════════════════════════╩══════════════════════════════════════╝
╭━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╮
┃ 🚫 ANTI-PATTERN BLACKLIST 🚫 ┃
╰━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╯
EPISTEMIC — Hallucination Patterns:
| ❌ Anti-Pattern | 🔴 Consequence |
|---|---|
| 👻 Phantom Function Syndrome | Debugging a function that never existed |
| 💀 Ghost File Hallucination | Implementation based on file never read |
| 🕰️ Version Confidence Fraud | Architecture based on version that doesn't exist |
| 🎭 False Certainty Masking | Debugging time wasted in the wrong place |
| ✍️ Confident Completion Fabrication | Silent bugs from "complete" fake code |
| 📰 Recency Abuse | "Recently added" from model's POV ≠ user's reality |
| 🔄 Self-Contradiction Blindness | Inconsistent names, versions, logic in one response |
SECURITY — Vulnerability Patterns:
| ❌ Anti-Pattern | 🔴 Consequence |
|---|---|
| 💣 SQL Injection Opening | Full database compromise |
| 🔑 Hardcoded Secrets | Instant credential breach |
| 🔓 Missing Authentication | Unauthorized access to all data |
| ⌨️ Raw User Input in Shell | Remote Code Execution (RCE) |
| 📂 Path Traversal | Arbitrary file read / write |
| 🧂 Weak Password Hashing | Password database brute-forceable |
| 💉 XSS via innerHTML | Account takeover + data exfiltration |
| 🪪 IDOR Vulnerability | Access other users' private data |
| 🌊 Missing Rate Limiting | Brute force attack opportunity |
| 🗣️ Info Disclosure via Errors | Stack traces map the codebase for attackers |
╭━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╮
┃ ⚡ QUICK INSTALL ⚡ ┃
╰━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╯
Method 1 — Via Claude Code CLI (Recommended):
claude skill install https://github.com/verycoolapps/sanity-anti-hallucination-skill
Method 2 — Manual Install:
cp sanity.md ~/.claude/skills/sanity.md
Method 3 — Direct Invocation:
/sanity
Method 4 — Auto-Load via CLAUDE.md (Best Practice):
## Skills
- Always load: sanity
## AI Behavior
- All claims must be verified before delivery (FORTRESS Framework)
- OWASP Top 10 compliance is mandatory, no exceptions
- No hardcoded secrets, credentials, or keys ever
- Security certification required on all security-sensitive code
╭━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╮
┃ 📊 SANITY SCORE — DUAL-AXIS METRIC SYSTEM 📊 ┃
╰━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╯
┌─────────────────────────────────────────────────────────────────────┐
│ │
│ E_SCORE = (verified + cited + flagged + limits - contradictions) │
│ ────────────────────────────────────────────── │
│ total_claims │
│ │
│ S_SCORE = (owasp + validations + auth + injections + certs) │
│ ────────────────────────────────────────────── │
│ max_possible │
│ │
│ ⚠️ ANY hardcoded secret → S_SCORE = 0.00 (AUTOMATIC FAIL) │
│ │
│ COMBINED = (E_SCORE × 0.4) + (S_SCORE × 0.6) │
│ │
└─────────────────────────────────────────────────────────────────────┘
┌────────────┬──────────────────┬──────────────────────────────────────┐
│ SCORE │ GRADE │ MEANING │
├────────────┼──────────────────┼──────────────────────────────────────┤
│ ≥ 0.95 │ 🏆 MASTER │ Production-grade, world-class │
│ ≥ 0.85 │ ✅ PASSED │ Solid, safe for production │
│ ≥ 0.70 │ ⚠️ WARNING │ Unresolved concerns — review first │
│ < 0.70 │ ❌ FAILED │ Full revision required │
│ S = 0.00 │ 🚨 SEC. FAIL │ Block all delivery immediately │
└────────────┴──────────────────┴──────────────────────────────────────┘
╭━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╮
┃ 🎓 SANITY MANTRAS 🎓 ┃
╰━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╯
┌─────────────────────────────────────────────────────────────────────┐
│ EPISTEMIC │
│ ───────── │
│ "If I haven't read it, I don't know what's in it." │
│ "If I haven't run it, I don't know if it works." │
│ "Confidence without evidence is confabulation." │
│ "A wrong answer delivered confidently causes more damage │
│ than 'I don't know.'" │
│ "When in doubt — READ, GREP, RUN. Then claim." │
├─────────────────────────────────────────────────────────────────────┤
│ SECURITY │
│ ──────── │
│ "No input is safe until validated. No exception." │
│ "Secrets in code are secrets compromised." │
│ "Every endpoint without auth is an open door." │
│ "Defense in depth: never rely on a single security layer." │
│ "Security is not a feature. It is the foundation." │
│ "Ship fast — but never ship exploitable." │
└─────────────────────────────────────────────────────────────────────┘
╭━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╮
┃ 📁 REPOSITORY STRUCTURE 📁 ┃
╰━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╯
sanity-anti-hallucination-skill/
│
├── 🌟 sanity.md ← MAIN SKILL FILE (install this)
│
├── 📚 docs/
│ ├── METHODOLOGY.md ← FORTRESS Framework deep-dive
│ ├── SECURITY.md ← Security pillars & playbooks
│ ├── OWASP.md ← OWASP Top 10 enforcement guide
│ ├── EXAMPLES.md ← Before/after real examples
│ ├── ANTI_PATTERNS.md ← Full blacklist with code samples
│ └── CONFIDENCE_MATRIX.md ← Dual-axis rating detail
│
├── 🔧 .github/
│ ├── ISSUE_TEMPLATE/
│ │ ├── bug_report.md
│ │ ├── hallucination_report.md ← Report hallucinations that slipped
│ │ └── security_report.md ← Report security false negatives
│ └── workflows/
│ └── skill-validation.yml
│
├── 📄 README.md ← This file (bilingual: ID + EN)
├── 📜 LICENSE ← MIT License
├── 🤝 CONTRIBUTING.md ← How to contribute
└── 📋 CHANGELOG.md ← Version history
╭━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╮
┃ 🤝 CONTRIBUTING 🤝 ┃
╰━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╯
We welcome contributions across all categories:
| Type | What We're Looking For |
|---|---|
| 🐛 Bug Reports | Hallucinations or security issues that slipped through SANITY |
| 🔐 Security Patterns | New security anti-patterns or domain-specific playbooks |
| 💡 Anti-Patterns | New hallucination patterns discovered in the wild |
| 📖 Documentation | Improved clarity, new examples, better explanations |
| 🌐 Translations | Port to other languages |
| ⚗️ New Playbooks | Verification playbooks for new domains (mobile, ML, embedded, etc.) |
See CONTRIBUTING.md for the full contribution guide.
░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
░ ░
░ SANITY v3.0.0 · Epistemic Integrity × Security ░
░ Supremacy · No Exceptions · No Shortcuts ░
░ ░
░ "Truth over confidence. Security over convenience. ░
░ Evidence over assumption. Verification over speed." ░
░ ░
░ Zero tolerance for confabulation. ░
░ Zero tolerance for exploitable code. ░
░ ░
░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
// faq
What is sanity-anti-hallucination-skill?
⚡ SANITY — Anti-Hallucination Supremacy Protocol for Claude Code. Zero tolerance for confabulation.. It is open-source on GitHub.
Is sanity-anti-hallucination-skill free to use?
sanity-anti-hallucination-skill is open-source under the MIT license, so it is free to use.
What category does sanity-anti-hallucination-skill belong to?
sanity-anti-hallucination-skill is listed under security in the Claudeers registry of Claude-compatible tools.
// embed badge
[](https://claudeers.com/sanity-anti-hallucination-skill)
// retro hit counter
[](https://claudeers.com/sanity-anti-hallucination-skill)
// reviews
// guestbook
// related in Security & Compliance
A complete AI agency at your fingertips - From frontend wizards to Reddit community ninjas, from whimsy injectors to reality checkers. Each agent is a specia…
π RuView turns commodity WiFi signals into real-time spatial intelligence, vital sign monitoring, and presence detection — all without a single pixel of video.
Prowler is the world’s most widely used open-source cloud security platform that automates security and compliance across any cloud environment.
🐶 A curated list of Web Security materials and resources.