Matt Pocock's Agent Skills Repo Reaches 280.7k Stars With 27 Skills, 11 of Them Invoked by the Model on Its Own

Matt Pocock's mattpocock/skills repository on GitHub shows 280.7k stars, 23.5k forks and 27 agent skills under an MIT licence. Eleven of those skills are marked as ones the model invokes on its own, and that number decides how much of every session they occupy.

The counts come from the repository page as read on October 8. The README splits the 27 into 20 engineering skills and 7 productivity skills, and it tags each as user-invoked or model-invoked.

What the 27 skills in mattpocock/skills contain

Group	Skills	User-invoked	Model-invoked
Engineering	20	11	9
Productivity	7	5	2
Total	27	16	11

The README describes them as "My agent skills that I use every day to do real engineering - not vibe coding." The account is mattpocock, linked to aihero.dev, which lists /grill-me, /domain-model, /to-prd, /to-issues, /tdd and /triage among the featured commands.

Install paths differ by agent. For Claude Code the README gives claude plugin install mattpocock-skills@claude-plugins-official, and for other agents npx skills@latest add mattpocock/skills -a <agent>. A one-time /setup-matt-pocock-skills follows in each repository.

Plugin or copied files: Claude Code plugin against skills.sh

The README is direct about the trade: "A plugin updates itself. skills.sh copies editable files into your project, and you update them by hand." The aihero.dev page repeats it, calling the Claude Code plugin a managed, read-only install and the other route a set of editable files written into the project.

Someone who edits /tdd to match a team's test layout needs the copied files. Someone who wants fixes as they ship needs the plugin, and loses the ability to edit.

What 11 model-invoked skills cost in context

Anthropic's Claude Code documentation says a skill's description is always in context in a regular session, so Claude knows what exists, and the body loads only when the skill is used. The description and when_to_use text is truncated at 1,536 characters in the skill listing.

That sets a ceiling. Eleven descriptions at the 1,536-character cap come to 16,896 characters, which is the most the model-invoked skills could add to the listing. Real descriptions may be shorter, and nothing in the sources measures them.

One gap stays open. The docs tie a skill's absence from context to the disable-model-invocation: true field. The README's "user-invoked" label sounds like the same thing, but the page I read does not say the repository sets that field on the 16.

Stars are not adoption

The aihero.dev changelog says version 1 brought "a 63% token reduction". That is the author's figure, it gives no baseline on the page, and nobody independent is named as having measured it. The v1.2 entry adds a "Wait What" skill aimed at Opus verbosity.

On GitHub the repository carries 150 open issues, 13 open pull requests and 543 commits. Neither number counts installs. The npx route copies files into a project without a fork, and the plugin route leaves no trace on the repository page at all.

The page shows no last-commit date, so how actively the repository is maintained is not something it answers.

For other agent tooling reported here, see the docker agent runner, whose README says nothing on sandboxing, and the REA reverse-engineering toolkit.