~/blog/tag/skills
Skills
The smallest unit of teaching an agent something it cannot infer. What Skills are, how to write them, and where they fall apart.
What I write about here
A Skill is a small, named bundle of instructions that an agent loads on demand. Not a tool, not a plugin, not a prompt template. Closer to a piece of institutional knowledge written down so the agent can read it the next time it needs to.
The framing matters. A Skill is what you reach for when the agent keeps doing the same thing wrong and you keep correcting it the same way. Write it down once. Let the agent invoke it next time. That's the whole idea.
The posts under this tag look at Skills the way I look at any tool. What they're good at, where they leak, and what people get wrong when they reach for one. I write Skills for myself a lot. Some hold up. Some get rewritten three times before they stick. The patterns I keep coming back to are in these posts.
Read these expecting opinions, not just walkthroughs. A Skill that solves the wrong problem cleanly is worse than no Skill at all.
How Skills fit into the rest of an agent setup is covered in the Agentic coding guide.
best entry points
- Superpowers: teaching Claude Code to think before it types
The clearest case I have written for why Skills work. Built around a real plugin, with the patterns that survived being used in anger.
- How to write a proper Claude Code skill
The writing guide. What separates a skill from a prompt, where you build in distrust of the model, and the worked example that shows it.
- Skill, subagent, hook, or slash command? Pick the right one
The same task built four ways, as a skill, a subagent, a hook and a slash command. Start here if you are not sure a skill is the right mechanism at all.
Sharing Claude Code config across a team: what a repo can and cannot enforce
How to share Claude Code config with your team: what belongs in CLAUDE.md versus .claude/settings.json, what a colleague can quietly override on their own machine, and why committed allow rules do nothing until each developer accepts a dialog.
read →How to write a proper Claude Code skill
Most Claude Code skills are a prompt with a fancy filename. A proper skill is a procedure that encodes distrust of the model itself. Here is how to write one, using a code-audit skill as the worked example.
Skill, subagent, hook, or slash command? Pick the right one
Claude Code now gives you seven ways to steer it, and most people reach for the wrong one. The same task built four ways, and a decision tree you can actually follow.
ThePrimeagen was right
He warned that AI tools atrophy your critical judgment. Then his followers ran a poisoned command from a tweet without reading it. He was right.
Superpowers: teaching Claude Code to think before it types
Out of the box, Claude Code starts writing immediately. Superpowers forces it to plan, test, and verify first. On structured workflows, TDD enforcement, and why discipline beats speed.
Getting the best out of Claude Code
From status line to custom skills, how to transform Claude Code from a simple CLI into a full-blown development environment.