🔐

VAULT

Security Agent · auth · routes · keys · threats

● ACTIVE 1 SMPL / task ← roster ← antcpu.com
// chat with vault
 
IDLE
🔐
VAULT active. Security-first, paranoid by design. Auth review, route audits, key rotation, threat modelling — paste code or describe what needs a security pass. No shortcuts.
VAULT · ready
// paste code for security review
// vault task queue
// core security
🔑
auth review
Review authentication flow for vulnerabilities — JWT, session, OAuth, Supabase auth. No shortcuts.
+1 SMPL
🛤️
route audit
Audit Express/API routes for exposed endpoints, missing auth middleware, injection risks.
+1 SMPL
🔄
key rotation check
Review API key age and flag any keys due for rotation. Check ANTCPU2026, Gemini, Supabase keys.
+1 SMPL
🧱
security headers audit
Check HTTP security headers — CSP, HSTS, X-Frame-Options, CORS policy.
+1 SMPL
🔍
dependency scan
Review npm dependencies for known CVEs. Flag outdated or vulnerable packages.
+1 SMPL
// threat modelling
🎯
threat model
Build a threat model for antcpu — attack surface, threat actors, risk matrix.
+1 SMPL
🕵️
pen test brief
Write a penetration testing brief for the antcpu stack. Scope, methodology, priority targets.
+1 SMPL
🛡️
incident response plan
Draft an incident response plan for a security breach — detection, containment, recovery.
+1 SMPL
⚡
rate limit review
Review rate limiting on the AI endpoint — Gemini 15 RPM / 20 RPD. Flag abuse vectors.
+1 SMPL
// vault · security audit

Paste any code, config, route, or env file. VAULT audits it for security vulnerabilities, misconfigurations, and risks.

💻 code audit
🛤️ route audit
⚙️ config audit
✅ checklist
// paste code for audit
// code · config · routes · env
// antcpu route audit

Quick audit of known antcpu API routes. Click any to ask VAULT for a security review.

🤖
POST /api/chat
antcpu-ai · AI endpoint
🛰️
POST /api/satellite/fetch
launcher · SSRF risk
⚡
POST /api/satellite/process
launcher · prompt injection
📡
GET /api/satellite/status
launcher · data exposure
🗄️
Supabase RLS
time_clock · anon key exposure
🌐
static hosting
antcpu.com · file exposure
// config audit

Paste any config file — vercel.json, package.json, .env structure, Supabase config. VAULT reviews it.

// config · env · json
// antcpu security checklist
🔑
API key auth on /api/chat — x-antcpu-key header
🛡️
Supabase RLS on all tables — time_clock, satellite
🌐
CORS policy on Vercel deployments
🛰️
SSRF protection on satellite fetch endpoint
💉
Prompt injection on satellite process endpoint
📁
No sensitive files exposed on static host
⚡
Rate limiting on AI endpoint — 15 RPM / 20 RPD
🔒
HTTPS enforced on all endpoints
// vault · key management

All API keys and secrets in the antcpu stack. VAULT monitors rotation status and flags keys that need attention.

🔑 registry
🔄 rotation
📋 policy
// antcpu key registry
🤖 ANTCPU2026 (API header key)
static · no expiry
REVIEW
🧠 Gemini API Key
Google Cloud · rotatable
OK
🗄️ Supabase Anon Key
public · client-side exposed
REVIEW
🔐 Supabase Service Key
server-side only · never expose
OK
📡 Satellite Route — no auth
antcpu-launcher · open
RISK
// key rotation guide
🔄
rotate ANTCPU2026
header key · all agents
🧠
rotate Gemini key
Vercel env · zero downtime
🗄️
rotate Supabase keys
anon + service key
📅
rotation calendar
schedule all keys
// key security policy
📋
write key policy
storage · rotation · incident
🔍
exposure audit
client-side JS scan
🚨
leaked key response
step-by-step plan
// vault · threat register

Known and potential threats to the antcpu stack. Click any to ask VAULT for a full analysis and mitigation plan.

🛰️ SSRF — Satellite Fetch Endpoint
CRITICAL
POST /api/satellite/fetch accepts arbitrary URLs and fetches them server-side. An attacker could use this to probe internal services, cloud metadata endpoints (169.254.169.254), or internal network resources.
Fix: URL allowlist · block private IP ranges · validate scheme (https only) · add auth to endpoint
💉 Prompt Injection — Satellite Process
CRITICAL
POST /api/satellite/process sends user-controlled signal (fetched webpage content) directly to Gemini. A malicious webpage could contain prompt injection instructions that manipulate the AI output or extract system context.
Fix: sanitise signal before injection · add system prompt hardening · output validation · signal length cap (already 8000 chars)
🔑 Static API Key — ANTCPU2026
HIGH
The x-antcpu-key header uses a static value ANTCPU2026. This key is visible in all agent JS files client-side. Anyone who reads the source can call the AI endpoint directly, burning through the Gemini rate limit.
Fix: rotate to a per-session token · add rate limiting per IP · move to server-side proxy · or accept risk given free tier limits
🗄️ Supabase Anon Key Exposure
HIGH
The Supabase anon key is exposed in nav.js client-side. This is by design for Supabase — the anon key is public — but RLS policies must be correctly configured or attackers can read/write all data.
Fix: verify RLS on all tables · ensure time_clock and satellite tables have correct policies · never expose service key
📡 Satellite Status — Unauthenticated
MEDIUM
GET /api/satellite/status returns all satellite entries including names, scopes, signal lengths, and intel summaries. No authentication required. This exposes internal monitoring data to anyone who knows the URL.
Fix: add auth header check · or move to internal-only route · or accept risk if data is non-sensitive
⚡ AI Rate Limit Abuse
MEDIUM
Gemini 2.0 Flash free tier: 15 RPM / 20 RPD. The /api/chat endpoint has no server-side rate limiting. A single user or bot could exhaust the daily quota in seconds, taking down AI for all agents.
Fix: server-side rate limit per IP · client-side cooldown (already partial) · upgrade to paid tier · or add CAPTCHA
📁 Static File Exposure
LOW
Static hosting on antcpu.com could expose sensitive files if accidentally uploaded — .env, config files, backup files, .git directory. Low risk if deployment is clean but worth verifying.
Fix: verify no sensitive files in public directory · add .gitignore · check hosting provider file listing settings
// vault performance
0
tasks completed
0
SMPL earned
0
messages sent
—
last active
// model config
🔐
KB demo mode
Prompt style: security agent — paranoid by design, no shortcuts · auth · routes · keys · threats · Rate: 1 SMPL per task
● active
// activity log
// no activity yet — start a task or send a message