25 agent skills to improve your workflow in GitHub Copilot | Matt Pocock | GitHub Copilot Day

Matt Pocock presents a GitHub Copilot Day session focused on improving reliability and usefulness when working with AI coding agents by using a curated set of reusable “Agent Skills”.

Overview

The session walks through a set of agent-oriented workflows (published as an open-source skills collection) that aim to:

• Align the agent with documentation and requirements before implementation
• Produce pull requests that are easier for humans to review
• Catch standards, specification, and architecture issues earlier
• Turn findings into actionable specs and tickets

Skills and workflows covered

Intro: the skills repo

• The presenter introduces the open-source “Agent Skills” repository and frames it as a collection of repeatable patterns for working with AI coding agents.

Grill with Docs: aligning with the agent

• “Grill with Docs” is presented as a way to align the agent with existing documentation before writing code.
• The presenter demonstrates using this approach in a real project context (a course video manager) and then “grilling” a specific issue live.
• Goal: reach shared understanding faster and reduce rework.

From alignment to a pull request

• The workflow results in a pull request that reflects the earlier alignment work.
• Emphasis is placed on making PRs “worth a human’s time” by improving clarity and usefulness.

“Show Me” skill: visual pull requests

• The presenter covers Dexter Horthy’s “Show Me” skill.
• Focus: generating visual pull requests to help reviewers understand changes more quickly.

Code Review skill in action

• The “Code Review” skill is demonstrated as part of the implementation flow.
• It is used to surface:
  • Standards issues
  • Spec check findings
• The presenter positions code review as something to run during implementation, not only at the end.

Improve Codebase Architecture skill

• The session includes an “Improve Codebase Architecture” skill aimed at identifying architecture issues early.

Plain-English output with “potato”

• The presenter shows a technique for forcing plain-English output using the word “potato”.

From report to spec and tickets

• The workflow moves from an agent-generated report into a clearer specification and then into tickets.

Implement Spec with parallel worktrees

• The presenter covers implementing a spec using parallel worktrees.

New skill: Retro

• A new “Retro” skill is introduced near the end as a wrap-up improvement loop.

Links shared

• Matt Pocock’s “AI Skills for Real Engineers”: https://github.com/mattpocock/skills
• Copilot app: https://gh.io/ghcpdaycopilotapp
• Copilot app resources: https://gh.io/ghcpdaycopilotappresources