Player not loading? Watch on YouTube
This video reviews Matt Pocock's skills for Claude Code and compares their modular approach with Superpowers and other development workflows. The speaker favors skills that users can invoke independently or revisit when requirements change, rather than a fixed sequence of dependent steps.
The walkthrough starts with /grill-me, which asks questions until the user and AI agent share an understanding. It follows one decision branch at a time, waits for answers, recommends options and requires confirmation before acting. /to-spec records those decisions without code blocks, so the agent consults the current codebase during implementation. /to-tickets divides the work into complete features that can be tested individually, such as login, instead of separate database, API and UI tasks.
The implementation discussion covers /implement and /tdd, with tests written before code. The speaker describes /code-review as using a fresh context window and 12 refactoring terms drawn from Martin Fowler's book. Later sections cover concise instructions, deep modules with a single entry point, and deletion tests for potentially redundant functions.
The speaker concludes that modular skills suit stronger models, while more prescriptive workflows may help weaker ones. That is his assessment, not a demonstrated benchmark.
