Claude Code Guardian

Create a selective security guardrail system, similar to a plugin for Claude Code, designed to enhance safety when using a dangerous permission mode. The system should intercept operations before they execute, allowing most safe operations to run silently while requiring explicit user approval for high-risk actions like destructive shell commands, accessing secrets (like `.env` or SSH keys), or force pushes. Include features like auto-committing pending changes, archiving files before deletion, and implementing hard blocks for catastrophic commands. The configuration should allow users to define custom regex patterns and path restrictions.