agent-guardrails

Install the agent-facing enforcement layer of a Python repo: Claude Code hooks that give a coding agent instant, deterministic feedback (and hard-block the few things it must never do), the settings wiring that ships them, a concise rules file, and vetted public skills. The failure this skill fixes is advisory-only guardrails — rules the model forgets under context pressure, hooks that silently never fire, and exit 1 "blocks" that block nothing.

When NOT to use