Updated
agent-skills is the public home for the Claude skills we build and rely on internally — the same ones that show up in our client engagements, kept open so anyone can use, fork, or contribute back.
Skills we actually use.
Abstract
Most "AI skill" libraries are speculative — patterns someone thought might be useful. This isn't that.
Every skill in this repo earned its place by surviving real client work first. If it ships here, we use it.
What's in it
- Skills for code review, refactoring, spec authoring, and shipping
- Conventions for working safely under an
AGENTS.mdcontract - Templates for self-contained implementation briefs
- Helpers for token-efficient agent workflows
Public, MIT, yours to fork.
Abstract
The skills are tools — useful only if other people can pick them up, adapt them, and run them on their own work.
So they're public, MIT-licensed, and intended to be forked. No "request access" gate, no per-seat anything.
What that means
- Use them in your own engagements, commercial or otherwise
- Fork the repo and adapt skills to your conventions
- Send pull requests if a skill could be tighter
- Open issues with use cases that aren't yet covered
A library that compounds, slowly.
Abstract
The repo isn't a one-time release. It's the public-facing version of a library we extend every time an engagement teaches us something new.
Skills land here when they've been used enough to be worth handing to someone else — not before.
What you can expect
- New skills as new engagement patterns emerge
- Refinements to existing skills based on real friction
- Documentation that reflects how we actually use them
- Conservative adds — sparse beats noisy