mattpocock/skills: A complete AI Coding workflow, end-to-end

Matt Pocock demonstrates the main path through his skills repository in an existing codebase: install and configure shared procedures, clarify the idea against the repository, persist a specification and tickets only when needed, implement a bounded slice, and review it from fresh context.

Key Points Covered

• Choose installation scope deliberately: Pocock uses a global setup for the solo demonstration but recommends project-scoped skills when a team needs to share and evolve the same workflow [00:00:00]-[00:04:03].
• Connect skills to repository conventions: The setup records issue storage, triage labels, and domain-documentation locations instead of assuming one universal project layout [00:04:03]-[00:06:58].
• Start with repository-aware clarification: The interview explores the existing code and asks follow-up questions until the developer and agent share a plan [00:06:58]-[00:09:36].
• Use durable artifacts only when the task crosses contexts: Bounded work can proceed directly; larger work can become a destination specification and context-sized tickets that survive fresh sessions [00:09:36]-[00:14:48].
• Separate implementation from fresh review: The demonstrated implementation runs repository checks and delegates specification and standards review to fresh sub-agents before committing [00:14:48]-[00:16:52].

The walkthrough uses one repository, task, harness, and model configuration. It illustrates the procedure but does not establish comparative delivery, defect, or maintenance improvements.

Full video: https://www.youtube.com/watch?v=M6mYodf0dJM(opens in a new tab)