Tutorial: Equip the Programming Agent with Matt Pocock skills and run through the first closed loop教程：给编程智能体装上 Matt Pocock 技能并跑通第一个闭环

Fastest path: Execute claude plugin install mattpocock-skills @ claude-plugins-official in Claude Code, and the plugin agreement will be automatically updated.

Manual path: Download the source code package (0.34 MB) and copy the skills folder required in skills/engineering and skills/productivity to the client skills directory (Claude Code is ~/.claude/skills/).

Other clients: Codex uses the codex plugin marketplace to add mattpocock/skills; any agent can select npx skills @ latest add mattpocock/skills interactively, remember to check setup-matt-pocock-skills.

Execute/setup-matt-pocock-skills in your project. It will ask in turn which ticket system to use (GitHub Issues, GitLab, local markdown file, or custom description), the routing rules for ticket tags.

After the answer is completed, it will write the adaptation configuration in the warehouse, and then the skills of to-tickets and triage will know where to open the work order and what label to use.

Run each warehouse once; the team warehouse submits the generated configuration to the repository, and the skill behavior of all employees is consistent.

Pick a feature you're working on and say “grill me on this design” in the conversation with the key points of the solution. The intelligent experience is like a strict senior colleague asking questions one by one: boundary conditions, failure paths, data migration, rollback schemes, until each decision branch has a clear answer.

The point to be asked is the one you haven't figured out - that's the value of a million installs of this skill: putting “take it for granted” before writing code.

Ending it will output a list of decisions that have been taken, which can be converted into a formal specification document directly with the to-spec skill.

Say "with TDD" to the specification that has been padded. Skills force a red-green-refactoring loop: first write the failed test, then write the implementation to let the test pass, and finally refactor, each step has a verifiable output.

During the implementation process, let the code-review skill review change again, and the retro skill do a brief review, and a complete closed loop will run through.

By adhering to the "grill → to-spec → TDD → code-review" chain on a daily basis, you will find that the number of reworks has decreased significantly - this is also the design intention of this skill set: to combine rigorous engineering disciplines with small tools.