AI Agent Guardrails

Add mechanical guardrails that stop AI agents from bypassing your rules. Enforces policies with git hooks, secret detection, deployment verification, and import registries — born from real production incidents.

This skill enforces agent safety mechanically rather than by prompt. It installs git hooks to block disallowed changes, adds secret detection to prevent token leaks, verifies deployments before they ship, and maintains import registries so agents can't silently rewrite or bypass rules — modeled on real production failures like crashes and credential leaks.

When to use

Use when you need hard, mechanical enforcement of rules on an AI agent or AI-assisted workflow — secret detection, git hooks, and deployment checks rather than prompt-only guardrails.

Examples

Block secret leaks

Pre-commit secret detection

Set up guardrails that scan agent-generated commits for API keys and secrets and block the commit if any are found

Enforce deploy checks

Verify before shipping

Add a deployment verification guardrail so an agent can't mark a deploy complete until health checks actually pass