[{"data":1,"prerenderedAt":192},["ShallowReactive",2],{"en-post-71":3},{"id":4,"title":5,"body":6,"description":173,"extension":185,"meta":186,"navigation":187,"path":188,"seo":189,"stem":190,"__hash__":191},"content/post/71.md","71",{"type":7,"value":8,"toc":172},"minimark",[9,14,38,42,45,84,88,107,111,114,118,121,125,135,139,142,146,149,153,160,166],[10,11,13],"h2",{"id":12},"what-it-does","What It Does",[15,16,17,18,22,23,26,27,32,33,37],"p",{},"mattpocock/skills is a collection of AI agent skills for real engineers, drawn directly from the author's daily ",[19,20,21],"code",{},".agents"," directory rather than toy demo skills. It hosts a batch of composable, easy to edit, small, and model agnostic agent skills for real software engineering tasks, including coding, debugging, architecture design, and requirement alignment. After installing, a developer runs ",[19,24,25],{},"/setup-matt-pocock-skills"," to configure issue tracking, labels, and doc locations, then calls the matching skill at each engineering step. To run these dev tools smoothly on Windows, start with ",[28,29,31],"a",{"href":30},"/post/46/","Install and Set Up WSL2 on Windows to Run Linux"," for a Linux dev environment inside Windows, or ",[28,34,36],{"href":35},"/post/45/","Run Virtual Machines on Windows with Sandbox and Hyper-V"," for a clean isolated build environment.",[10,39,41],{"id":40},"the-problem-it-solves","The Problem It Solves",[15,43,44],{},"The README lists four typical failure modes of AI coding agents and pairs each with a skill:",[46,47,48,60,63,74],"ul",{},[49,50,51,52,55,56,59],"li",{},"The agent does not follow instructions, a communication gap, addressed by ",[19,53,54],{},"/grill-me"," and ",[19,57,58],{},"/grill-with-docs",", which interrogate and align before any code is written.",[49,61,62],{},"Output is too verbose and lacks a shared language, fixed by a GLOSSARY.md glossary so the agent uses concise domain language.",[49,64,65,66,69,70,73],{},"Generated code will not run, covered by ",[19,67,68],{},"/tdd",", the red green refactor test loop, and ",[19,71,72],{},"/diagnosing-bugs",", a staged debugging loop.",[49,75,76,77,55,80,83],{},"The codebase rots into a big ball of mud, handled by ",[19,78,79],{},"/to-spec",[19,81,82],{},"/improve-codebase-architecture",", which focus on modules and architecture.",[10,85,87],{"id":86},"language-and-architecture","Language and Architecture",[46,89,90,93,100],{},[49,91,92],{},"The body is Shell config and SKILL.md skill definitions, split into engineering and productivity.",[49,94,95,96,99],{},"It uses changesets, specifically ",[19,97,98],{},"@changesets/cli",", for version and changelog management; Codex compatible metadata lives in openai.yaml, and AGENTS.md is a symlink to CLAUDE.md.",[49,101,102,103,106],{},"It installs for Claude, Codex, Copilot, and Gemini CLI, or via skills.sh, that is ",[19,104,105],{},"npx skills@latest add",". Invocation distinguishes User-invoked, for orchestration, from Model-invoked, for reusable specs.",[10,108,110],{"id":109},"license","License",[15,112,113],{},"MIT",[10,115,117],{"id":116},"why-it-is-trending","Why It Is Trending",[15,119,120],{},"The author mattpocock is a well known TypeScript educator with roughly 60,000 newsletter subscribers. The skills hit real AI coding pain points and are model agnostic, and the agent skills concept exploded in 2026. It gained about 1,164 stars that day.",[10,122,124],{"id":123},"project-links","Project Links",[15,126,127,130,131,134],{},[19,128,129],{},"https://github.com/mattpocock/skills"," (site ",[19,132,133],{},"https://aihero.dev/skills",")",[10,136,138],{"id":137},"summary","Summary",[15,140,141],{},"The value of this collection is that it comes from a real workflow rather than a demo. It pairs each of the most common AI coding failures with a skill, which makes it a ready made discipline library for engineers who want agents to be more reliable.",[10,143,145],{"id":144},"getting-started","Getting Started",[15,147,148],{},"After installing, run /setup-matt-pocock-skills to configure issue tracking and doc locations so later skills work well. Newcomers should start with /grill-me and /tdd, one for alignment and one for testing, which fix the most common daily failures. The GLOSSARY.md is worth writing carefully, since a precise glossary keeps agent output concise. The skills are model agnostic, working under Claude, Codex, or Gemini, and a shared set across a team reduces style drift. Spend the first session wiring the setup command and reading two or three skill definitions, because understanding what each skill expects is what turns a loose prompt into a reliable, repeatable step in your real workflow.",[10,150,152],{"id":151},"faq","FAQ",[15,154,155,159],{},[156,157,158],"strong",{},"Is it Claude only?","\nNo. It explicitly supports Claude, Codex, Copilot, and Gemini CLI, and offers a generic install path. The skills are written in a model agnostic format, so anyone can call them.",[15,161,162,165],{},[156,163,164],{},"Do I need a programming background?","\nThese skills target software engineering tasks, so you should understand basic engineering steps like testing, debugging, and architecture. Absolute beginners should learn the basics first, then come back to it.",[15,167,168,171],{},[156,169,170],{},"How is it better than ad hoc AI coding in chat?","\nChat style coding drifts, rambles, and rarely lands. This collection hardens steps like alignment, testing, and architecture into callable commands, which saves detours and is easier to reuse across a team.",{"title":173,"searchDepth":174,"depth":174,"links":175},"",2,[176,177,178,179,180,181,182,183,184],{"id":12,"depth":174,"text":13},{"id":40,"depth":174,"text":41},{"id":86,"depth":174,"text":87},{"id":109,"depth":174,"text":110},{"id":116,"depth":174,"text":117},{"id":123,"depth":174,"text":124},{"id":137,"depth":174,"text":138},{"id":144,"depth":174,"text":145},{"id":151,"depth":174,"text":152},"md",{},true,"/post/71",{"description":173},"post/71","9rZCT11qC4leFMwJteekai0T9w8Ho9eUJDV3xFzP-GU",1794378304265]