Skills For Real Engineers
WorkflowMatt Pocock's composable Claude Code skills for TDD, requirement grilling, diagnosis, and architecture improvement.
Install
$ npx skills@latest add mattpocock/skills
Update
$ npx skills@latest add mattpocock/skills
Re-running the installer fetches the latest version; re-select the skills/agents you want.
When to use it
- You want Claude to interview you and reach shared understanding before writing code (grill-me / grill-with-docs)
- You want to enforce a real TDD red-green-refactor loop instead of one-shot code generation (tdd)
- You need a disciplined debugging loop for a hard bug or performance regression (diagnose)
- You're fighting a 'ball of mud' codebase and want guided architectural deepening (improve-codebase-architecture)
Commands & flags
- /setup-matt-pocock-skills
- One-time per-repo setup to configure issue tracker and docs paths.
- /grill-me
- Interview you to reach shared understanding before coding.
- /tdd
- Enforce a real red-green-refactor TDD loop.
- /diagnose
- Run a disciplined debugging loop for a hard bug or perf regression.
- /improve-codebase-architecture
- Guided architectural deepening for a ball-of-mud codebase.
- /to-prd, /to-issues, /triage, /zoom-out, /prototype, /handoff
- Productivity and misc slash commands from the collection.
How to use it
Install with the skills.sh installer, select the skills and agents you want, then run the one-time setup once per repo to configure your issue tracker and docs paths.
npx skills@latest add mattpocock/skills
Then in Claude Code:
/setup-matt-pocock-skills
Invoke individual skills as slash commands: /grill-me, /grill-with-docs, /tdd, /diagnose, /improve-codebase-architecture, /to-prd, /to-issues, /triage, /zoom-out, /prototype, /handoff. The collection (19 skills) is grouped into Engineering, Productivity, and Misc — including git-guardrails-claude-code to block dangerous git commands and write-a-skill.
Why it earns its place
A widely-used collection maintained directly from Matt Pocock's own .claude directory, so it reflects real day-to-day engineering rather than a demo. Instead of one prescriptive framework, it offers small, composable, model-agnostic skills that target specific AI-coding failure modes — misalignment, verbosity, broken code, and architectural decay — turning the agent into a disciplined engineering participant.
Try it
> /grill-me then /tdd to build the new auth flow