Matt Pocock Skills is an MIT-licensed open-source collection of Claude Code agent skills published straight from Matt Pocock's personal .claude directory. The repository has rocketed past 82,600 GitHub stars and 7,100 forks since its February 2026 release, becoming one of the fastest-growing single-author skill libraries on the platform. Pocock, well known to TypeScript and developer-education communities, has distilled years of professional engineering practice into 18 composable skills that give Claude Code real engineering discipline: from issue triage to test-driven development to architectural recovery on legacy codebases. ## Why Matt Pocock Skills Matters Matt Pocock's core thesis is that the rate of feedback is the speed limit of software engineering, and that AI agents fail most often when they widen the feedback loop rather than tighten it. The skills in this library are engineered to do the opposite: they break large tasks into small verifiable steps, demand documentation of the project's domain language before coding, push the agent toward test-driven workflows, and surface architectural decay before it becomes terminal. The result is an agent that behaves less like an over-eager intern and more like a senior engineer pairing with the developer, which is exactly the failure mode most teams are trying to fix. ## Engineering Skills The engineering category includes ten skills: diagnose for systematic problem analysis, grill-with-docs that forces the agent to consult documentation before answering, triage for converting issue backlogs into actionable work, improve-codebase-architecture for refactoring with clear goals, setup-matt-pocock-skills for one-command configuration, tdd for strict red-green-refactor cycles, to-issues for turning conversations into tracked issues, to-prd for generating product requirement documents, zoom-out for stepping back when the agent gets lost in details, and prototype for fast disposable experimentation. ## Productivity Skills Four productivity skills round out the day-to-day: caveman, a deliberately stripped-down mode for noise reduction, grill-me, which has the agent quiz the developer rather than the reverse, handoff, which produces context-rich notes for the next session or teammate, and write-a-skill, a meta-skill that uses Pocock's own structure to generate new skills consistently. ## Miscellaneous and Setup Skills The miscellaneous category adds git-guardrails-claude-code for safer git operations under agent control, migrate-to-shoehorn for adopting Pocock's Shoehorn TypeScript helpers, scaffold-exercises for educational content authoring, and setup-pre-commit for installing pre-commit hook discipline into a repository. Each skill follows the same compact format: a clear purpose, a precise procedure, and the minimum required guardrails to keep the agent on rails. ## One-Line Install and Issue-Tracker Setup Installation uses the same cross-platform skills installer that powers the broader ecosystem: npx skills@latest add mattpocock/skills. After installation, running /setup-matt-pocock-skills inside Claude Code walks the developer through configuring an issue tracker, with GitHub, Linear, and local-file options, choosing triage labels, and pointing the agent at the project's documentation. From that point on the agent uses Pocock's workflows automatically when relevant. ## Core Philosophy in Practice The library's design philosophy is opinionated and visible in every skill. Small iterative steps are preferred over big-bang commits. Domain-specific language is documented before code is written. Tests are written first whenever feasible. Architectural drift is detected with zoom-out passes and corrected with improve-codebase-architecture. Documentation is treated as primary source material, not afterthought. These are the practices that experienced engineers internalize over decades, and the library makes them executable by an agent. ## Limitations The library is opinionated, which is its strength and its constraint. Teams whose conventions differ sharply from Pocock's may need to fork and adapt rather than adopt directly. Some skills depend on issue trackers or documentation locations that need configuration up front. The library targets Claude Code primarily, though the markdown skill format is portable enough for other agents to adapt. Finally, like any skill library, it raises the floor on agent output but cannot substitute for human review on consequential design and security decisions.