the guardrails i actually use with ai agents

the concrete guardrails i put around ai coding agents, including read-only and ask modes, command allowlists, scoped database credentials, branch protection, and human-in-the-loop gates for high-risk actions

Raccontata dadev.to