Dcg

What it does

The Dcg skill is a Claude Code hook designed to prevent destructive commands from being executed by AI coding agents. It intercepts potentially harmful Git commands, such as those that could lead to data loss or accidental code deletion (e.g., git reset --hard, rm -rf ./src). Dcg operates with sub-millisecond latency thanks to Rust and SIMD acceleration, ensuring minimal disruption to the development workflow while providing a crucial safety net.

When to use it

• When using AI coding agents that automate Git operations.
• To prevent accidental data loss due to typos or incorrect commands generated by an agent.
• In environments where uncommitted changes are valuable and need protection from destructive actions.

Key capabilities

• Destructive Command Blocking: Intercepts and blocks specific, known dangerous Git commands.
• Whitelist-First Architecture: Prioritizes allowing explicitly safe commands while blocking potentially harmful ones.
• Default-Allow Behavior: Unrecognized commands are allowed by default to avoid disrupting legitimate workflows.
• Zero False Negatives Philosophy: Prioritizes preventing dangerous commands over avoiding false positives (occasional manual confirmation is acceptable).

Example prompts

• "Clean up the build artifacts." (Will be blocked if it translates to rm -rf ./src)
• "Reset to the last commit." (Will be blocked if it translates to git reset --hard)
• "Fix the merge conflict." (May be blocked depending on how the agent attempts to resolve it)

Tips & gotchas

• The skill operates as a Git hook, so ensure your environment supports and allows for Git hooks.
• Dcg prioritizes preventing data loss; you may occasionally need to manually confirm commands that are flagged.
• New or unusual Git commands will be allowed by default until explicitly categorized as safe or destructive.