Skip to main content

GitHub Copilot Best Practices

A guide to effective development patterns and techniques using GitHub Copilot.

1. Planning & Problem Solving​

1.1. Explore → Plan → Code → Commit​

  • Research and understand the problem space using @workspace to explore the codebase
  • Create a detailed implementation plan using Copilot's Plan Mode
  • Write the code following the plan
  • Commit with descriptive messages

When to use: New features, bug fixes, refactoring tasks. This is your daily driver for structured development.

💡 Use commits as checkpoints, commit frequently. Consider using a commit message prompt to standardize commit messages.

💡 Use git worktree for parallel development

1.2. Spec-Driven Development​

  • Create a Product Requirements Document (PRD) based on user input
  • Generate tasks from the PRD
  • Track and manage task progress

Consider using spec-kit for structured spec-driven development workflows.

When to use: Greenfield projects, PoC development, pet projects

💡 This methodology works best with clear requirements. Be prepared to iterate and refine - feel free to discard approaches that don't work well.

1.3. Brainstorming & Ideation​

  • Ask Copilot to brainstorm multiple approaches and iterate on ideas
  • Use generated ideas as grounding context for subsequent prompts
  • Build on previous suggestions rather than starting from scratch

When to use: Early stages of feature design, architectural decisions, or when stuck on a problem.

💡 Be very direct and explicit with the way you want to interact. E.g.: "give me 3 results and compare pros and cons for each result using markdown tables"

💡 For common tasks, you can create reusable prompts to describe your problem-solving approach. See pros-and-cons for an example.

2. Context Engineering​

2.1. Context Grounding​

  • Provide relevant files and examples using #file: references
  • Perform prior codebase/feature exploration with @workspace and ask Copilot to focus on relevant parts
  • Use web search and MCP servers to gather additional context and information

💡 Hint Copilot in the right direction. E.g.: "Hint: you might want to check relevant tests for this functionality"

💡 Use MCP servers for latest/authoritative information. E.g.: (Microsoft Docs MCP, Context7)

2.2. Project Memory​

GitHub Copilot uses instruction files for project-wide context:

Repository-level: .github/copilot-instructions.md

  • Purpose: Team-shared context and project knowledge
  • Content: Architecture decisions, design patterns, coding standards
  • Best practice: Keep it concise but comprehensive and high-level

IDE-level: .github/instructions/*.instructions.md

  • Purpose: Development environment specific guidance
  • Content: Testing conventions, coding style, editor-specific workflows
  • Examples: coding-guidelines, testing-xunit

Agent coordination: AGENTS.md

  • Purpose: Multi-agent orchestration guidelines
  • Content: Agent responsibilities, coordination patterns, task delegation

2.3. Session Management​

  • Start new chat session - Reset conversation context when switching to a different task
  • Context Control: Keep conversations focused on a single topic or feature
  • Manual context management: Use #file: to include specific files, @workspace for broader context

💡 Be aware of context limits and how they may impact response quality. It is better to keep your context short and focused.

3. Customization Overview​

GitHub Copilot can be customized through several mechanisms:

3.1. Instructions​

Instructions provide persistent context that shapes Copilot's behavior.

LocationScopePurpose
.github/copilot-instructions.mdRepositoryTeam-shared project standards
.github/instructions/*.instructions.mdIDEDevelopment environment guidance
AGENTS.mdRepositoryMulti-agent coordination

Examples in this repo:

3.2. Prompts​

Prompts are reusable templates for common tasks.

LocationScope
.github/prompts/*.prompt.mdRepository-wide prompts

Examples in this repo:

3.3. Agents​

Agents extend Copilot with specialized capabilities for complex multi-step tasks.

Location: .github/agents/*.agent.md

Examples in this repo:

  • debug - Debugging assistance
  • research - Research and exploration
  • differ - Git branch diff analysis

3.4. Skills​

Skills are reusable, composable capabilities with specialized knowledge.

Location: .github/skills/<skill-name>/SKILL.md

Examples in this repo:

Skills can define:

  • Specialized knowledge and workflows
  • Tool permissions via allowed-tools
  • Reference files for progressive disclosure

4. Tips for Success​

4.1. Communication​

  • Be specific about requirements and constraints
  • Provide context about your development environment
  • Ask for explanations when solutions aren't clear

4.2. Iteration​

  • Start with working solutions, then optimize
  • Use feedback loops to refine approaches
  • Don't hesitate to backtrack and try different methods

4.3. Learning​

  • Experiment with different prompt styles
  • Learn from Copilot's reasoning and explanations
  • Build a personal knowledge base of effective patterns

Resources​