GrantGuard is a local, read-only-by-default tool. It runs a web UI bound to 127.0.0.1, reads your Claude Code settings files, and only ever writes when you explicitly confirm a removal. It makes no network connections of its own.

We still take security seriously and welcome reports.

Please do not open a public issue for security problems.

Report privately via GitHub's "Report a vulnerability" button on the repository's Security tab (Security Advisories). If you can't use that, email security@vanta.com with "GrantGuard" in the subject.

Please include:

• A description of the issue and its impact.
• Steps to reproduce (or a proof of concept).
• The version / commit you tested, and your OS.

We'll acknowledge your report within 5 business days and keep you updated as we work on a fix. We're happy to credit reporters who want it.

GrantGuard is pre-1.0; security fixes land on the latest main and the most recent release.

• The UI server enforces loopback-only Host and same-origin Origin checks to resist DNS-rebinding and CSRF against the local write endpoint.
• GrantGuard never transmits the contents of your settings files anywhere; all analysis happens locally and secrets are redacted in the UI/CLI output.
• Removing a rule that contained a leaked credential does not rotate the secret — see the note in the README.