mattpocock/skills: agent skills from a .agents dir
mattpocock/skills collects the agent skills the author ships from his own .agents directory. 273,938 stars, 23,000 forks, MIT licensed.
TL;DR: mattpocock/skills is a MIT licensed collection of agent skills, described only as “Skills for Real Engineers. Straight from my .agents directory.” It holds 273,938 stars against 23,000 forks, a ratio that says most people read it rather than run it.
What it is
The repository description is the entire pitch, and it is short enough to quote: “Skills for Real Engineers. Straight from my .agents directory.”
That second sentence is the useful part. It positions the repo as a mirror of a working setup rather than a packaged product. Whatever sits in there comes from an agent configuration directory the author actually works in. The project’s home page is aihero.dev/skills, which ties it to the same author’s teaching material.
GitHub reports the language as Shell. That fits a repository of small executable scripts and setup files rather than a library with a build step.
The numbers tell the story
| Field | Value |
|---|---|
| Stars | 273,938 |
| Forks | 23,000 |
| Language | Shell |
| License | MIT |
| Created | 2026-02-03 |
| Last push | 2026-09-29 |
| Size on disk | 1.8 MB |
The star-to-fork ratio sits at roughly twelve to one. For a runnable tool you would expect the inverse, where people fork to modify and then install their own copy. This ratio reads like reference material. People star it so they can find it again, and a much smaller group forks it to adapt the skills for their own agent directory.
The repository is eight months old and still had a push three days before this writing, so it is not a one-shot launch.
How to look at it
Cloning takes the usual form:
git clone https://github.com/mattpocock/skills
At 1.8 MB the whole thing is small enough to read end to end, which is the point. A skills directory is only useful if you can see exactly what your agent will be told to do, and there is no black box here. Nothing compiles, nothing resolves, nothing phones home.
That property matters more than it used to. Agent instructions are executable surface area even though they are not code. A shell script you can read in ten seconds and a skills file you cannot are very different risk profiles.
When this fits
It fits if your agent harness reads skills from a directory and you want examples pulled from someone else’s daily setup. Reading costs nothing and the repo is small enough to skim in a sitting.
It does not fit if you need a supported dependency. MIT is permissive and unambiguous, but there is no company behind it and no compatibility promise. Forking and owning the result is the safer pattern, which the fork count suggests is what actually happens.
Treat it as a reference implementation of a working directory rather than as a package you install.
Source and Accuracy Notes
- Project page: https://github.com/mattpocock/skills
- Source repository: mattpocock/skills (verified via GitHub API
html_url) - License: MIT (verified via GitHub API
license.spdx_id) - Stars at time of writing: 273,938
- Counts, language, license, and dates were read from the GitHub REST API. The README was not fetched, so the individual skills contained in the directory are not described here.
Related Posts
ai-setup
Recall – Persistent Memory for Claude Code via MCP Hooks
Recall gives Claude Code a permanent memory store that survives session restarts and context compaction. Four hooks capture and restore context automatically — with cloud SaaS or self-hosted options.
2/28/2026
dev-tools
Automotive Skills Suite for AI Engineering
Evaluate Automotive Skills Suite for APQP, ASPICE, HARA, safety-plan, and DIA workflows with setup notes, governance risks, and SME review guidance.
5/28/2026
dev-tools
awesome-agentic-ai-zh Roadmap Guide
Explore awesome-agentic-ai-zh as a Chinese agentic AI learning roadmap, with setup notes, track selection, study workflow, and evaluation guidance.
5/28/2026