Claude Code has six permission modes, and yes, deny rules apply to every one of them, even bypass. Here's what each mode auto-approves and how to block commands like rm for good.
There are millions of public agent skills now, and a benchmark of 47,150 of them found most barely move the needle. Here is how to spot the junk and write a SKILL.md that measurably improves your agent.
Got a spare or broken-screen MacBook Pro? Here is how I turned mine into an always-on, closed-lid homelab server for running self-hosted AI agents like OpenClaw and Hermes 24/7.
AI coding agents are fast, but an unsupervised one can delete files, leak secrets, or push junk to main. Here are the practical guardrails I use to let agents run hard without wrecking the repo.
How do you test an AI agent when it never gives the same answer twice? Here is a beginner's guide to evals, with practical examples you can copy.
If your AI coding agent keeps spinning up git worktrees and leaving your work somewhere that isn't main, here's what a worktree actually is, why agents love them, and how to stop fighting it.