Matt Pocock's Engineering Skills
Paste the following prompt into your AI chat to install this skill:
Follow https://skillhub.cn/install/skillhub.md and install @user_3cdd1ed9/mattpocock-skills.
About this skill
Problem
Agent-assisted coding often fails in four concrete ways: requirements drift, verbose output with inconsistent terminology, weak feedback loops, and codebases that accumulate structural debt. Matt Pocock's Skills turns these failures into explicit workflows instead of leaving the model to improvise. It targets real repository work where clarification, validation, and review need to leave a trace.
How It Works
The skill is a router over self-contained procedures in references/. The main path starts with /grill-with-docs or /grill-me, using interview-style questioning to sharpen the goal. When a codebase exists, decisions are captured in CONTEXT.md and ADRs. If a question needs a runnable answer, the flow moves to /prototype and uses /handoff to carry context between sessions. For multi-session builds, /to-prd turns the thread into a PRD, /to-issues splits it into independently grabbable vertical slices, and each issue is implemented in a fresh session with /implement. That implementation step drives /tdd internally, working in one red-green slice at a time, then finishes with /code-review to check standards and specification.
It also includes supporting paths: /triage for incoming bugs and requests, /diagnosing-bugs for reproduce-minimize-hypothesize-instrument-fix-regression-test loops, /improve-codebase-architecture for finding deepening opportunities, and /domain-modeling plus /codebase-design for shared vocabulary around domain terms and deep modules. /research delegates reading to a background agent against primary sources, while /teach supports multi-session concept learning.
Boundaries
The skills assume an issue tracker, triage labels, and a domain-document layout, so run /setup-matt-pocock-skills before the first engineering flow. They are best suited to codebases where process matters: clarification, review, and context continuity. Avoid forcing them for one-off scripts or simple questions, and use /handoff at intentional phase breaks rather than mid-phase compaction.
Use Cases
- Take over a feature request in an existing codebase, clarify boundaries with /grill-with-docs, and record decisions in CONTEXT.md and ADRs.
- Turn a long discussion into a PRD, then split it into independently grabbable vertical slices with /to-issues for fresh implementation sessions.
- Debug an intermittent failure with /diagnosing-bugs by reproducing, minimizing, instrumenting, fixing, and adding a regression test.
- Review a branch or PR with /code-review against standards and spec, while advancing implementation slices with /tdd in red-green steps.
Best For
- Senior engineers maintaining real business codebases and needing traces for requirements, implementation, review, and architecture decisions.
- Team leads who delegate multi-turn agent coding but want it to proceed through TDD, code review, and independently grabbable issues.
- Architecture maintainers working on complex codebases who need to surface deep-module opportunities and codify domain terminology.
- Independent developers using AI agent tools and needing stable context handoff across long sessions and implementation issues.
Related Skills
A systematic code error fixing skill that follows a five-step workflow: Reproduce, Isolate, Understand, Fix, Verify, to help developers resolve code errors.
Supports inline SVG/HTML rendering and Mermaid code output, covering common technical diagram creation like architecture and flow diagrams.
Publish HTML, static assets, or build output as a public link, with account management, update, deletion, and AI safety review.
A comprehensive browser automation guide covering navigation, interaction, multi-tab management, and data collection, tailored for domestic web ecosystems and tools like Playwright/Selenium.