- Use the right built-in mode: Know when to reach for Ask (sparring partner), Plan (review before execution), or Agent (direct execution), and why the distinction matters
- Blueprint before building: Use AI to produce BLUEPRINT.md and developer_todo.md before writing a single line of implementation code
- Author specialist agents: Create focused agents with clearly defined responsibilities, minimal tool sets, and explicit domain boundaries
- Replace guesswork with skills: Build reusable skills that give agents deterministic capabilities — consistent backlog item structure, correct test execution, reliable project state
- Implement end-to-end with TDD: Drive one backlog item from data layer through to a tested UI component using the strict test-first loop
- If you have not used GitHub Copilot much you can start with this repository.
- Select only one language, more instructions in the exercises folder
- Agent mode can help to install the missing parts of your development environment
- Try inline Copilot, chat, agent-mode
- Guided exercises for testing Copilot instruction files, prompt files, and MCP
- Exercises for agents, hooks, and skills. Select based on the language, the exercise is same in all repositories.
- Prompting guide
- Awesome Copilot
- GitHub list of MCP
- GitHub Copilot plans and their features
- Atlassian MCP Server repository
- Atlassian getting started with MCP
- Spark https://github.com/spark (needs GitHub Copilot Enterprise license)
- Speckit https://github.com/github/spec-kit