# Security Policy ## Supported Versions | Version | Supported | | ------- | --------- | | latest | ✅ Yes | ## Reporting a Vulnerability **Please do not report security vulnerabilities through public GitHub issues.** Email: **security@node9.ai** Include: - A description of the vulnerability and its potential impact - Steps to reproduce - Any suggested mitigation (optional but appreciated) You will receive an acknowledgment within 48 hours and a resolution timeline within 5 business days. We follow responsible disclosure — we ask that you give us time to patch before publishing details publicly. ## Threat Model node9-proxy is a **localhost-only** daemon that intercepts AI agent tool calls before they execute. The daemon binds exclusively to `127.0.0.1` and is never exposed to the network. | Component | Trust boundary | | ------------------------------------------ | ------------------------------------------------------------------------------ | | Daemon HTTP server | localhost only (`127.0.0.1:7391`) | | Browser UI | Protected by a per-session CSRF token (`x-node9-token` header) | | Internal resolve endpoint (`/resolve/:id`) | Protected by a separate per-session internal token (`x-node9-internal` header) | | Audit log | Written to `~/.node9/audit.log`; secrets are redacted before storage | | Decisions file | Written to `~/.node9/decisions.json`; stores only `allow`/`deny` per tool name | **Known limitation**: If an attacker already has local code execution on the machine running the daemon, they can bypass controls. This is by design — the daemon is a human-in-the-loop safety layer, not a sandbox. ## Scope Reports are in scope for: - Command injection or privilege escalation in the proxy or CLI - Policy bypass — a dangerous tool call that passes through without triggering HITL - Credential exposure (API keys stored or transmitted insecurely) - Dependency vulnerabilities with a clear exploit path against Node9 users ## Out of Scope - Vulnerabilities in the wrapped MCP servers themselves (report those upstream) - Theoretical attacks with no practical exploit path