If you find a security issue in any hook or plugin — a bypass of a guard hook (a dangerous command or secret access that should be blocked but isn't), a way for hook input to trigger unintended command execution, or secrets leaking into logs — please report it privately rather than opening a public issue:

• GitHub: use private vulnerability reporting on this repository.

You'll get a response within a few days. Once fixed, reporters are credited in the release notes unless they prefer otherwise.

These hooks are guardrails, not a sandbox. They pattern-match tool calls Claude Code is about to make and block the obviously dangerous ones. A determined adversary (or a sufficiently creative agent) can construct commands that evade any regex — that is a known limitation, not a vulnerability by itself. Reports that meaningfully tighten a guard against realistic agent behavior are still very welcome as regular issues or PRs.

Bypasses of protect-secrets that leak credentials through a documented, blocked channel, and any case where a hook itself executes untrusted input, are always in scope.

The main branch and the latest published plugin versions in the marketplace are supported. Older plugin versions in the install cache are not patched retroactively — reinstall to get fixes.