These are the 5 agent skills I use every single day to get production-grade code from Claude Code. Each one is a small markdown file that encodes real engineering process so the AI has a strict path to follow.
The result? Code quality from AI shot up dramatically.
Skill 01 → Grill Me
What it does: Forces the AI to interview you relentlessly before writing a single line of code. Just 3 sentences long, but it generates 15–50 questions per session.
Why it matters: Claude tends to spit out a plan too early. This skill forces a real conversation until you reach a shared understanding. No more hallucinated plans.
The skill:
"Interview me relentlessly about every aspect of this plan until we reach a shared understanding. Walk down each branch of the design tree, resolving dependencies between decisions one by one. If a question can be answered by exploring the codebase, explore the codebase instead."
npx skills@latest add mattpocock/skills/grill-me
Skill 02 → Write a PRD
What it does: Takes a grilled idea and turns it into a proper Product Requirements Document — problem statement, user stories, implementation decisions — then submits it as a GitHub issue.
Why it matters: Your destination is now documented. The PRD becomes the source of truth for everything that follows.
The workflow:
- Ask for a detailed description
- Explore the repo to verify assertions
- Interview the user (reuses grill-me)
- Sketch out major modules
- Write the PRD and submit as a GitHub issue
npx skills@latest add mattpocock/skills/write-a-prd
Skill 03 → PRD to Issues
What it does: Takes the PRD (your destination) and breaks it into a journey — a kanban board of independently grabbable vertical slices with blocking relationships.
Why it matters: Each issue is a thin vertical slice that cuts through all integration layers, not a horizontal slab of one layer. This flushes out unknown unknowns fast. Parallel agents can grab unblocked work simultaneously.
npx skills@latest add mattpocock/skills/prd-to-issues
Skill 04 → TDD
What it does: Forces the agent into a Red → Green → Refactor loop. Write one failing test, then the code to pass it, then refactor. Repeat until complete.
Why it matters: TDD with agents is the single most consistent way to improve code quality. It includes guidance on interfaces, testability, mocking, and deep modules.
The workflow:
- Confirm what interface changes are needed
- Confirm which behaviors to test
- Design interfaces for testability
- Write one failing test
- Write code to pass it
- Look for refactor candidates
- Repeat
npx skills@latest add mattpocock/skills/tdd
Skill 05 → Improve Codebase Architecture
What it does: Explores your repo, finds shallow modules, then spawns parallel sub-agents to design 3+ radically different interfaces. You pick the winner.
Why it matters: Garbage codebase = garbage AI output. This skill identifies where understanding one concept requires bouncing between too many files, where pure functions were extracted just for testability but bugs hide in how they're called, and where tightly coupled modules create integration risk.
Run it once a week. Watch code quality compound.
npx skills@latest add mattpocock/skills/improve-codebase-architecture
The Pipeline
These 5 skills form a complete engineering pipeline:
Idea → Grill Me → PRD → Issues → TDD → Architecture
The key insight: treat AI agents like engineers with no memory. Give them strict, well-defined processes and the code quality goes up dramatically.