Matt Pocock Skills Deep Dive: Small Agent Skills for Real Engineering Work

The repo's posture is deliberately small and composable. It does not try to replace the engineering process with a giant agent framework. It gives agents better habits for the parts of software work that routinely fail: ambiguity, feedback loops, domain language, architecture drift, testing, and handoff quality.

1. skills in One Sentence

Matt Pocock's `skills` repo is an MIT-licensed set of small, adaptable AI-agent workflow skills for real software engineering tasks such as TDD, diagnosis, product clarification, documentation, architecture improvement, issue triage, and handoffs.

2. Why It Matters

The repo matters because many AI coding failures are process failures. The agent misunderstands the request, changes too much at once, lacks feedback, ignores domain language, or produces code that is hard to maintain.

Instead of one huge methodology, these skills are narrow tools. You can run `/grill-with-docs` before a feature, `/tdd` during implementation, `/diagnose` on a bug, `/improve-codebase-architecture` after entropy builds, and `/handoff` when context must move.

The engineering stance is also notable: the README rejects vague vibe coding and emphasizes fundamentals such as shared language, red-green-refactor, small steps, feedback, deep modules, and design stewardship.

3. Architecture and Mental Model

The repo is organized by skill category: engineering, productivity, misc, personal, in-progress, and deprecated. A small setup skill records per-repo configuration such as issue tracker, triage labels, and docs location so other skills can operate consistently.

4. Smallest End-to-End Setup

The commands below are copied from the repository documentation and checked against the current research snapshot. Treat them as a starting point, then read the linked README before installing into a production environment.

# Install through skills.sh
npx skills@latest add mattpocock/skills

# In your agent, select the skills you want.
# Include setup-matt-pocock-skills.

# Then run the setup skill once per repository:
/setup-matt-pocock-skills

A small first task should prove the integration before you attach it to critical data or large workspaces.

# Clarify a feature and update docs/ADRs
/grill-with-docs

# Build the change with a red-green-refactor loop
/tdd

# Diagnose a hard bug
/diagnose

# Break a plan into independently grabbable issues
/to-issues

# Hand off a session to another agent or future context
/handoff

5. Technical Deep Dive

before coding:
  ask hard questions
  define domain terms
  record decisions
  narrow scope
  then implement

6. Real-World Wrong vs Right Patterns

7. Common Mistakes and Current Issues

The issue tracker matters because these are young, fast-moving repos. The article uses issues as risk signals, not as proof that a project is unusable.

8. Performance, Scaling, and Cost Notes

These skills are mostly text and process, so runtime cost depends on the agent and repo size rather than a heavy local build. The biggest token cost comes from long grilling sessions, architecture analysis, issue decomposition, and repeated codebase reads.

The practical way to control cost is to choose narrow skills. Use `/diagnose` for one bug, `/tdd` for one slice, `/to-issues` for one plan, and `/handoff` when context should be compressed.

The repo's value is compounding: shared language, ADRs, issue slices, and handoffs reduce future ambiguity, which can save more tokens than the initial session costs.

9. Who It Is For

10. Community Signal

The strongest community signal is the maintainer's existing TypeScript and AI-engineering audience. The README routes people through skills.sh and a large AI Hero newsletter audience, which gives the repo immediate distribution.

The issue tracker is productively opinionated. Many issues are policy or workflow questions, not only bugs: setup verification, question limits, issue tracker scope, parent issue behavior, docs paths, and command packaging.

The repo's framing is refreshingly grounded: it argues that fundamentals matter more with agents, not less.

11. The Verdict: Is It Worth Using?

12. The Bigger Picture

This repo represents a mature direction for agent tooling: not bigger agents, but better rituals around agents. The skill is the process wrapper that keeps a model from turning ambiguity into code too early.

As AI coding gets faster, the bottleneck shifts to judgment, constraints, feedback, and maintainability. Small skills like these are one way to encode those engineering habits directly into the agent workflow.

13. Frequently Asked Questions

14. Glossary

15. All Sources and Links

Internal Links

• Mastering Agent Skills
• Academic Research Skills Guide
• Claude Code Best Practice Guide

16. Source Attribution Table