Vanta releases GrantGuard for auditing Claude Code permissions
GrantGuard is a local utility designed to audit the permission list that Claude Code accumulates when users select "always allow." These entries reside in rarely checked configuration files and can include hardcoded API keys, unrestricted git push commands, or destructive operations like rm -rf.
The tool categorizes permissions by risk level: remove (inline keys, git reset *, git push *), review (npm install *, gh api *), and keep (fixed Bash commands, Read). The audit process is read-only, and deletions occur only after manual confirmation or via the --fix flag.
GrantGuard is written in standard Python 3.10+ with zero dependencies. The UI binds to 127.0.0.1 and makes no network calls. It runs on macOS, Linux, and Windows via uv run grantguard.py. The source code is available under the MIT license at VantaInc/grantguard.
Related stories
- Claude Code's deny rules don't stop grep
- The .claude directory creates an attack surface in Claude Code
- Researcher identifies hidden tracking in Claude Code
- A strict sandbox mode surfaces in Claude Code desktop
- Confessor reconstructs what Claude Code actually touched
- Claude Code 2.1.283 keeps new models out until admins say so
Comments
No comments yet. Be the first.
Join the conversation
Sign in with Google to leave a comment. Your name and avatar come from your Google profile, and the comment appears after moderation.