← All lab notes

LAB / OCT 12, 2026

Giving my coding agents some context

Why I keep my coding agent skills in one repo, and what goes into them.

THE NOTE

I use a few different coding agents, and I want them to have the same basic context about my projects. Which repos belong together. Where the shared UI code lives. How I build and release things.

skillz is where I keep those instructions.

A skill is a folder with a Markdown file in it. It tells an agent when the instructions apply and what it needs to know. Some have reference files or templates alongside them.

The Tide projects are a good example of why I want this. They're separate repos, but they share UI code and conventions. An agent working in one repo might have no reason to know that the thing it's about to build already exists in another. The Tide skill gives it that context and tells it where to look.

Other skills cover terminal UI design, Go CLI releases, and the content workflow for my Astro sites. These are things specific to how I work, so it makes sense to keep them somewhere I can edit.

There's an installer and a little terminal picker, too. For Claude Code and Codex, the installer links back to the repo. I can change an instruction once and have both use the updated file. Hermes gets a copy, so that one needs another install after an edit.

I like having the instructions in ordinary files. I can read what I'm asking the agents to do, change anything that's wrong, and keep the history in Git.

The collection is personal, so some of it assumes my projects and setup. If you're curious, the repo is public. You can see what I've put in there and adapt anything useful.

NEXT

Got a product problem that needs this kind of thinking?

Tell me about it ↗