A set of small, composable agent skills that Matt Pocock uses for his daily coding workflow: the agent first interrogates you to clarify requirements, establishes a project glossary, and then implements step-by-step according to specs, tickets, TDD, and code reviews. It addresses common agent issues such as "not doing what I wanted," "talking too much," "code not running," and "getting messier as it goes."
Stars 307489 (+876 / 24h)
Rationale
- Divided into user-invoked and model-invoked categories: user-invoked skills like grill-with-docs, to-spec, to-tickets, implement, wayfinder, and retro handle orchestration; model-invoked skills like tdd, diagnosing-bugs, code-review, prototype, and research can be loaded by the agent itself
- grill-with-docs maintains the project's GLOSSARY.md and ADRs during interrogation; the author claims that shared terminology helps agents reduce fluff and maintain consistent naming; tdd adheres to red-green-refactor, while diagnosing-bugs stages the process of reproduction, minimization, and fixing
- The productivity set is not limited to code: includes grill-me, handoff (compressing conversations into handover documents), teach, to-questionnaire, etc.
Usage
- Claude Code: claude plugin install mattpocock-skills@claude-plugins-official; for Codex and Copilot, use their respective plugin marketplace add mattpocock/skills; for other agents, use npx skills@latest add mattpocock/skills -a <agent-name>
- Run /setup-matt-pocock-skills once per repository to select the issue system (GitHub, GitLab, local files, etc.), triage labels, and documentation storage location
Risks
- You must choose between the plugin version and the skills.sh version; installing both will result in each skill appearing twice; the skills are designed around issue systems and ticket workflows, so they are largely unnecessary for small single-file script projects
- This is the author's own methodology, with no effectiveness evaluation; the repository was created in 2026 2, licensed under MIT