This is my personalized GitHub Copilot workflow — a collection of skills, instructions, and agents I use daily to enhance my coding experience. It's organized for personal use but open for anyone to explore, adapt, or draw inspiration from.
The repo also includes a deployment script that makes it easy to install and keep everything organized in a clean folder structure while maintaining Copilot's required file format for seamless integration.
| Type | What It Is | Purpose | Example |
|---|---|---|---|
| Skill | Reusable, single-purpose function | Do one specific thing well, and use it across multiple scenarios | Code formatting, email polishing, quiz generation |
| Agent | Multi-step orchestrator | Combine multiple skills or tasks to complete one larger goal | Research agent that gathers info, searches, and synthesizes into a report |
| Instruction | Global rule set | Define conventions, constraints, and behavior that apply across all skills and agents | Coding standards, tone guidelines, tool restrictions |
- Skill: Create a skill when you have a reusable workflow that solves one problem and could be invoked across many different projects or scenarios. Don't create a skill if it's a one-off task.
- Agent: Create an agent when you need to orchestrate multiple skills or complex multi-step tasks to achieve a single larger goal. Don't create an agent if a single skill would do.
- Instruction: Create an instruction when you have a rule, convention, or constraint that should apply globally across your entire workflow. Don't create an instruction for task-specific logic.
This workflow is built on a collaborative dialogue between human thought and AI execution — think of it as a senior developer guiding a junior who also thinks critically and challenges back.
The core principle: Human intention and human verification guide AI execution at every step, with continuous feedback loops. This means you, as the user, maintain your technical knowledge and understanding of software engineering, architecture, and project design. You're not delegating thinking — you're augmenting it. The AI is a junior collaborator that executes your vision and brings critical feedback, suggestions, and identifies potential flaws or gaps in your approach.
This is a true dialogue: you clarify intent, the AI executes and offers suggestions, you verify and course-correct, the AI challenges your assumptions if needed — and critically, before anything gets implemented, you both agree on exactly what's going to happen. This mutual verification ensures you're speaking the same language and moving forward with intention.
This back-and-forth is evident in Matt Pocock's "grill me" and brilliantly explained in David Ondrej's video on the future of AI. The MD builder skill in this repo is a direct implementation of this philosophy — it's a loop of ask/write/probe/confirm that mirrors true collaboration, keeping human thought in the driver's seat while leveraging AI's ability to catch blind spots.
This project is heavily inspired by Matt Pocock and his work on skills. His approach to prompt engineering and agentic engineering — especially the principle of reducing context and giving language models just enough information to execute one task effectively — guides what I want my workflow to become better at.