agent-guard

agent-guard is a set of safety hooks that intercepts shell commands and file writes before a coding agent executes them, refusing destructive operations (like force pushes to main or rm -rf outside safe directories) and rewriting risky patterns (like pkill -f to pkill -A or git push --force to --force-with-lease). It runs without network dependencies using only Python 3.10+ and the standard library, silently passing through benign commands while guiding agents toward safe alternatives when dangerous operations are detected.