Skip to content

Security: unwind-mcp/unwind

Security

SECURITY.md

Security Policy

Reporting a Vulnerability

If you discover a security vulnerability in UNWIND, please report it responsibly.

Email: security@brugai.com Alternative: Open a private security advisory on GitHub.

Please include:

  • Description of the vulnerability
  • Steps to reproduce
  • Affected component (enforcement pipeline, CRAFT, sidecar, adapter, Ghost Mode)
  • Impact assessment if known

Response Timeline

  • Acknowledge: within 48 hours
  • Triage: within 7 days
  • Fix: within 90 days (critical vulnerabilities prioritised)

Scope

In scope:

  • Enforcement pipeline bypasses
  • CRAFT chain integrity issues
  • Sidecar authentication or authorization flaws
  • Ghost Mode escape or data leakage
  • Adapter input validation issues
  • Path jail or self-protection bypasses

Out of scope:

  • Social engineering or phishing
  • Denial of service against a self-hosted Pi
  • Issues in upstream dependencies (report to those projects directly)
  • Attacks requiring physical access to the host

OpenClaw Hardening Recommendations

If you run UNWIND with OpenClaw, review these settings for defence-in-depth:

  • session.dmScope: Set to per-channel-peer (not the default main). The default shares a single session across all DM senders, which undermines UNWIND's per-session taint tracking. With per-channel-peer, each sender gets an isolated session.
  • gateway.auth.mode: Ensure a token or password is set. If the gateway binds to a non-loopback address without authentication, tool calls can bypass the UNWIND adapter entirely.
  • tools.fs.workspaceOnly: Set to true to complement UNWIND's path jail at the OpenClaw level.
  • tools.elevated.enabled: Leave as false unless you have a specific need. UNWIND blocks elevated exec on tainted sessions, but defence-in-depth is better.

Recognition

Valid reporters will be credited in this file's Hall of Fame section (unless they prefer anonymity).

Hall of Fame

No reports yet — be the first!

There aren't any published security advisories