| Version | Supported |
|---|---|
| 0.1.x | Yes |
If you discover a security vulnerability in GuardRail, please report it responsibly. Do not open a public issue.
- Email: security@promptandbuild.de
- Include: description, reproduction steps, impact assessment
- We will acknowledge within 48 hours
- We aim to provide a fix within 7 days for critical issues
GuardRail is a pre-execution guard system. Vulnerabilities in scope include:
- Guard bypass techniques (commands that should be blocked but are not)
- Dispatcher logic errors that skip guards
- Configuration injection (manipulating GUARDRAIL_* variables to weaken guards)
- Audit log tampering or evasion
Out of scope:
- Vulnerabilities in Claude Code, Cursor, or other AI tools themselves
- Issues that require root access (GuardRail assumes the host is trusted)
- Social engineering of the AI agent (GuardRail guards commands, not conversations)
We follow a 90-day disclosure policy. After reporting, we will:
- Confirm the vulnerability
- Develop and test a fix
- Release the fix
- Credit you in the changelog (unless you prefer anonymity)
After 90 days, you may disclose publicly regardless of fix status.