git-guardrails-claude-code

You can block dangerous git commands in Claude Code by installing a PreToolUse hook that detects risky operations and returns a BLOCKED response. This prevents destructive actions like git push --force, git reset --hard, and git clean before execution.

Destructive git operations that should be blocked include git push --force, git reset --hard, git clean, and git branch -D. Blocking these commands prevents accidental data loss and workflow disruption in automated environments.

To set up a PreToolUse hook, copy the bundled script to your target hook location (.claude/hooks for per-project or ~/.claude/hooks for global) and make it executable. The hook will intercept and block specified risky git actions automatically.

Yes, the git guardrails hook requires jq to function properly. The jq dependency is necessary for the script to parse tool inputs and detect dangerous git commands within the Claude Code environment.

You can configure git command blocking either per-project or globally. Install the hook into .claude/hooks for project-specific scope, or into ~/.claude/hooks for global scope across all Claude Code environments.

Claude Code indicates an intercepted git command by printing a BLOCKED message to stderr and exiting with code 2. This feedback mechanism confirms that the PreToolUse hook successfully detected and stopped the dangerous operation.