In late April 2026, TypeScript educator and Total TypeScript founder Matt Pocock published his daily‑used AI Agent skills catalog to GitHub, repository at mattpocock/skills. The project quickly exploded, gaining over 130k stars within weeks and becoming one of the most popular open‑source skill collections for AI coding tools like Claude Code, Cursor, and Antigravity.
In the README, Matt Pocock states clearly: “These are the Agent skills I use every day for real engineering, not Vibe Coding.” The repository's core value is turning software engineering principles (such as TDD, domain‑driven design, architectural practices) into structured “Skills” that can be called directly by AI coding agents, helping developers significantly improve AI‑assisted development efficiency and code quality.
Core philosophy
Matt Pocock argues that the biggest problems with current AI coding agents are process loss of control and insufficient feedback. Many developers let agents “freestyle,” which often produces large amounts of low‑quality, hard‑to‑maintain code. The skills project was created to solve these issues. Its design principles are:
- Small and focused: each Skill has a single responsibility and is easy to compose.
- Adaptable: suitable for mainstream AI agents like Claude Code, Cursor, Antigravity.
- Engineering‑driven: based on practical experience from classics such as The Pragmatic Programmer and Domain‑Driven Design.
- User‑led: emphasizes that developers retain control rather than letting AI take over the process.
The project divides Skills into User‑invoked (invoked manually by users) and Model‑invoked (invoked automatically by agents); the former is used for orchestration, the latter for execution.
Main skills and usage
Installation (30‑second quick start)：
Bash
npx skills@latest add mattpocock/skills
After installation, run /setup‑matt‑pocock‑skills in your AI agent to initialize configuration, including selecting an issue tracker (GitHub/Linear/local), triage labels, and document storage location.
Recommended core skills：
- /grill‑me and /grill‑with‑docs（most popular）
- Function: have the agent conduct an in‑depth “grilling” to ensure alignment on requirements before coding begins.
- Use case: essential before any new feature or refactor.
- Benefits: builds the project's domain model and generates CONTEXT.md and ADR documents.
- /tdd
- Function: enforces the strict Red‑Green‑Refactor cycle.
- Usage: after entering requirements, the agent writes a failing test, implements code, then refactors.
- /improve‑codebase‑architecture
- Function: scans the codebase, generates an HTML architecture report, identifies “shallow modules” and suggests deeper refactoring.
- Recommended frequency: run every few days to prevent the codebase from turning into a “mudball”.
- /triage
- Function: manages issue state machines and helps developers systematically handle to‑dos.
- /to‑prd and /to‑issues
- /to‑prd: converts conversations into PRDs and pushes them to an issue tracker.
- /to‑issues: breaks plans into vertically sliced, independently deliverable issues.
- /diagnosing‑bugs
- Function: structured bug diagnosis workflow (reproduce → minimize → hypothesize → detect → fix → regression test).
Other skills include prototyping (/prototype), code handoff (/handoff), teach mode (/teach), and more.
Tips：
- Prefer /ask‑matt as a router to help the agent choose the most appropriate Skill.
- Combine the agent with the project's CONTEXT.md to greatly improve its understanding of the business context.
- Run /improve‑codebase‑architecture regularly to maintain code health.
Project status and community response
As of late June 2026 the repository has accumulated over 130k stars and an active Discussion in Issues and PRs. Matt Pocock regularly updates the skills and shares progress via a newsletter with over 60k subscribers.
Community feedback generally holds that these Skills significantly reduce AI coding “hallucinations” and low‑quality outputs, helping developers move from “letting AI write code” to “engineering with AI.”
Conclusion and takeaways
The mattpocock/skills project provides a practical, structured methodology for AI‑assisted engineering. It demonstrates that good AI usage is not permissive but guided by carefully designed processes and discipline. For developers looking to improve AI coding efficiency, this repository is among the most recommended open‑source tools to install.
The project is fully open source (MIT license) and welcomes contributions and customizations to fit specific projects.
Repository: https://github.com/mattpocock/skills
▶️ Enjoy this story? Creating takes a lot of work!
Follow our YouTube channel:
AI Rosetta
Your likes, subscriptions, and shares are our greatest support! 🙏
More AI news at: https://www.airosetta.com