mattpocock/skills: Learn the Whole Flow, End-to-End

Introduction & Overview 0:00

First proper tutorial for the skills repo — covering the main flow that handles most of your work.

• The mattpocock/skills repo has reached 162K+ stars and 7.5M downloads with 38 skills total
• This is the first proper tutorial — focused on the main flow, not advanced or new features
• Works on both brownfield (existing) and greenfield (new) codebases
• Demo uses the AI Hero CLI repo as a real-world example

Installing the Skills Repo 0:54

One-command install that works with any coding agent and any project structure.

• Requires Node.js installed on your machine
• Skills installer built by Vercel
• 38 skills in two groups:
  • Official — Matt's blessed, public-facing skills
  • Experimental — may be deleted at any time
• Select all official skills during install
• Choose your coding agent: Claude Code, Cursor, Codex, etc.
• Installation scope:
  • Project — committed to repo, shared with team
  • Global — for solo devs, applies everywhere
• Symlink recommended over copy — stays up to date automatically

Running Setup & Configuration 4:34

One-time project setup that configures your issue tracker, labels, and domain documentation.

The setup configures three things:

1. Issue tracker — works with anything: GitHub, local markdown, Jira, Linear — just tell the agent what you use
2. Triage labels — standardized labels for ticket communication between you and the agent
3. Domain documentation — context.md + ADR (Architecture Decision Record) files:
   • Single context — for 99% of users, one context file
   • Multi context — for monorepos with multiple packages

Creates links in claude.md to issue tracker docs, triage labels, and domain docs.

Ask Matt Skill 6:52

Matt Pocock as a skill — an AI persona that knows everything about the skills repo.

• Knows everything about the skills repo — structure, philosophy, recommended usage
• Recommends starting with grill-with-docs as the entry point
• Emphasizes being context-window conscious
• Introduces the "smart zone" concept — the range of context where AI agents perform best

Grill With Docs Interview 8:31

The starting point of the main flow — an interactive interview that builds shared understanding.

• Agent explores the codebase first, then asks targeted follow-up questions
• Typically 6–20 questions depending on complexity
• Uses auto mode, not plan mode
• Goal: reach shared understanding between you and the agent
• Produces a plan as output

Creating Specs & Tickets 10:08

Compress the entire discussion into a spec, then break it into implementable tickets.

Step 1: /to-spec

• Compresses all discussion (46.1K tokens in the demo) into a detailed spec document
• Saved to your configured issue tracker
• Spec = destination — what it looks like when done
• Contains: problem statement, solution, user stories, implementation decisions, testing decisions

Step 2: /to-tickets

• Turns spec into an implementation plan of discrete tickets
• Each ticket = one context window / one smart zone
• Real example: 11 sub-issues generated from one spec
• Tickets are short — acceptance criteria lives in the main spec

Implementation & Code Review 13:51

Implement tickets one at a time with automatic code review via sub-agents.

• Clear context between each ticket — essential for staying in the smart zone
• Command: @tickets implement this for each ticket
• Don't do all tickets at once — one ticket per session

Code review runs automatically as part of /implement, checking two axes:

1. Compare work against original spec — catches forgotten items
2. Check against repo's coding standards — or Martin Fowler defaults if none defined

Workflow Recap 16:22

The complete end-to-end flow distilled into three phases.

• This main flow handles most work — everything else is experimental improvements
• For small tasks: skip steps 2 — go directly from grill-with-docs to implement
• Sign up for the newsletter for day-of updates on new skills