Do not open a public issue.

• Preferred: GitHub → Security → Report a vulnerability (private advisory) on this repo
• Fallback: Email security@freeai.local with subject [FreeAI Security]

Include: affected version/commit, reproduction steps, impact, and any logs. We aim to acknowledge within 72h and ship a fix with a coordinated disclosure.

• Router auth (X-API-Key / ROUTER_API_KEY), rate limiting, and provider key handling
• Agent API auth (X-API-Key / AGENT_API_KEY) for red/blue/purple team operations
• Autonomous API auth (X-API-Key / AUTONOMOUS_API_KEY) for run creation and shell operations
• SDLC sandbox (workspaces/<run_id>/ traversal guards, ENABLE_SHELL_TOOLS gating)
• Supply chain: Docker images (ghcr.io/.../freeai-*), Live ISO build (live/build-live.sh)

The default install enables UFW (22/8030/8050), fail2ban (sshd), and unattended-upgrades. For internet-exposed hosts:

1. Always set authentication keys:

   • ROUTER_API_KEY for router access
   • AGENT_API_KEY for agent API (especially red/blue/purple team operations)
   • AUTONOMOUS_API_KEY for autonomous SDLC API (required when ENABLE_SHELL_TOOLS=1)
   • DASHBOARD_AUTH_TOKEN for dashboard write operations

2. Network isolation:

   • Use ENABLE_DESKTOP_PORTS=1 only behind TailScale/Cloudflare
   • Keep router (:8010) and llama.cpp (:9001) on localhost/tailnet only
   • Expose only dashboard (:8030) and autonomous (:8050) to the internet when necessary

3. Shell tools:

   • Only enable ENABLE_SHELL_TOOLS=1 when necessary
   • Always set AUTONOMOUS_API_KEY when shell tools are enabled
   • Monitor /auto/runs endpoint for suspicious activity

See AUTONOMOUS_SECURITY.md for detailed autonomous API security configuration.

• sk-or-v1 placeholder keys in skill docs were redacted in 6004c83 (push-protection). Rotate any key you may have pasted from docs.