| Version | Supported |
|---|---|
| Latest release | ✅ Yes |
| Older releases | ❌ No — upgrade to latest |
Please do not report security vulnerabilities through public GitHub issues.
If you discover a security issue in agentguard — including but not limited to:
- A hook bypass that allows a blocked operation to execute
- Shell injection in hook logic or the installer
- A way to exfiltrate secrets that agentguard is supposed to protect
- An unsafe file write or privilege escalation in
install.sh
Please report it privately via GitHub's private vulnerability reporting.
Include:
- A description of the vulnerability
- Steps to reproduce
- The potential impact
- Any suggested fix (optional)
- Acknowledgement within 3 business days
- Assessment (confirmed / not a bug / won't fix) within 7 days
- Fix and release as soon as possible for confirmed issues, with credit to the reporter in the release notes (unless you prefer to remain anonymous)
agentguard is a security tool — any bypass of its enforcement hooks is in scope. The following are not in scope:
- Vulnerabilities in the agents agentguard protects (Claude Code, Kiro, Cursor, Codex)
- Issues requiring physical access to the machine
- Social engineering
agentguard's hooks operate at the shell level. They are a defence-in-depth layer, not a security boundary that can guarantee isolation. A determined user with shell access can always disable them. The goal is to prevent accidental or agent-initiated dangerous operations, not to protect against a hostile local user.