I made this after worrying that AI coding sessions, copied logs, or quick test fixtures could leak real data into a repo.

Git Shield installs pre-commit/pre-push hooks. It uses gitleaks for secrets and OpenAI Privacy Filter for contextual PII in outgoing diff additions. Everything runs locally; code is not sent to a service.

The current tradeoff is: fast secret scan at commit time, heavier PII scan at push time. It supports Linux, macOS, and WSL, with CPU fallback for small diffs.

Repo: https://github.com/vekexasia/git-shield

The tool is made for crankers so output contains both line numbers and clear instructions for llms to follow. Most of the times the agent ask me what to do but was able to automatically ignore some false positives.