The radar · Friday, August 21
packaged as agent skills
It's Friday, August 21, and this is the radar. Today, Matt Pocock packaged an engineer's habits as skills your coding agent follows, and three more repos are climbing behind it. As always, you leave with a command you can run tonight.
The Headliner
The headliner is today's fastest riser. Matt Pocock's skills repo, almost two hundred and thirty thousand stars, and it added over three thousand today alone. These are the skills he uses every day, published straight from his own setup. Small markdown files your coding agent loads as habits. Planning, test driven development, code review, debugging. Not prompts you paste in. Procedures the agent actually follows.
The demo
# Claude Code $ claude plugins install mattpocock-skills # any other agent $ npx skills@latest add mattpocock/skills
tdd red-green loops code-review multi-axis review diagnosing-bugs structured debugging grill-with-docs aligned planning implement build with testing to-spec specs and tickets + prototype, research, domain-modeling …
Here's the whole setup. Two install paths, straight from the readme. On Claude Code it's a plugin, one line. On anything else, npx skills, add, and the repo name. And this is what lands. The engineering shelf alone has test driven development, code review, structured debugging, and a planning skill that interviews you against your own docs. One command and the roster is in your session. It works with Claude Code, Codex, and other agents.
Inside a skill
TDD is the red → green loop. This skill is the reference that makes that loop produce tests worth keeping…
A seam is the public boundary you test at. Tests live at seams, never against internals. Test only at pre-agreed seams — before writing any test, write down the seams under test and confirm them with the user.
+ tests.md · mocking.md — the anti-pattern references ship with it
Let me show you inside one, because this is where it earns the stars. This is the test driven development skill, quoted straight from the file. The rule that jumped out at me. Tests live at seams, the public boundaries, and the agent has to agree those seams with you before it writes a single test. That's why I don't think this is star bait. These files encode judgment, not boilerplate. Your agent inherits opinions about what a good test even is.
Three Fast Ones
google-timeline-visualizer
Also climbing today. OpenLogi, a native Rust replacement for Logitech's bloated driver app. Local first, no account, no telemetry, and it added almost fourteen hundred stars today. If you own a Logitech mouse, this one's for you. Google timeline visualizer draws your year in travel from your own location history export. It's two days old and already climbing. And Apache Maka, a local first agent workspace. Two thousand stars is early, but the Apache Software Foundation incubating an agent workspace is the actual story there.
The One-Liner
And the one liner, because the episode is not over until you can try it. npx skills at latest, add, mattpocock skills. Run it in any project tonight and your agent picks up better habits than most of us have. That's the radar. Back tomorrow.
