If you discover a security vulnerability in MxClaw, please report it responsibly:

1. DO NOT open a public GitHub Issue
2. Email: security@mxclaw.ai (or use GitHub Security Advisories)
3. Include:
   • Description of the vulnerability
   • Steps to reproduce
   • Impact assessment
   • Suggested fix (optional)

We will acknowledge receipt within 48 hours and provide a fix timeline within 7 days.

• HTTP API: Bearer token authentication (gateway.apiToken in config)
• WebSocket: Device token authentication with rotation on each connection
• Webhooks: Per-platform signature verification (HMAC-SHA256, Ed25519)

• Channel allowlists: Per-channel sender filtering
• Mention gating: Only respond when @mentioned in group chats
• Tool approval modes: always-require-approval, owner-only, yolo
• File path restrictions: Tools can only access configured paths

• Config redaction: API responses never expose tokens/keys
• Token hashing: Device tokens stored as SHA-256 hashes
• Pairing codes: 8-char hex, 5-minute expiry, single-use
• Rate limiting: Per-IP HTTP + per-client WebSocket

• Docker: --network none isolation for bash commands
• SSH: Remote execution on dedicated sandbox hosts
• File system: Workspace path enforcement

1. Always set gateway.apiToken in production
2. Use allowlists on all public-facing channels
3. Enable always-require-approval for bash and file_write tools
4. Run behind a reverse proxy (nginx/Caddy) with TLS
5. Use Docker sandboxing for bash tool execution
6. Rotate device tokens regularly
7. Monitor logs for failed auth attempts

• Voice session tokens are ephemeral (not persisted)
• Config file is not encrypted at rest
• WebSocket connections are not end-to-end encrypted (use TLS termination)