These are the agent skills I lean on every day. They live in my ~/.claude
directory and ship as a public repo so anyone can install them, fork them, or open a PR with their own.
A skill is a focused, repeatable workflow an agent can run. It bundles a clear instruction set, the inputs it needs, and the outputs you should expect. Skills live as small markdown files that any Claude-compatible agent picks up.
The point isn't to wrap the agent in scaffolding. It's to write down the exact moves I'd make for a given problem so the agent runs them the same way I would, every time.
One command pulls the whole catalog into your global skills directory:
npx skills add mattpocock/skills -y -g
Drop -g
to install into the current project only. Once installed, any agent that reads ~/.claude/skills
(Claude Code is the obvious one) can pick a skill from the catalog and run it.
You can also install individual skills if you don't want the whole set. Browse the source for slugs: github.com/mattpocock/skills.
Each skill below is a workflow I reach for in a specific moment: a vague spec to interrogate, a domain to model, a PRD to write, a backlog to break down, a feature to test-drive. They aren't "AI productivity hacks." They're how I keep my taste and standards intact while the agent does the work.
The catalog lives at github.com/mattpocock/skills. If you've shaped a skill that's earned its place in your workflow, share it.