Agentic Setup for GitHub Copilot — Context, Skills, and Tools

The Three-Layer Model Applied to GitHub Copilot

GitHub Copilot has evolved far beyond inline code completion. With Copilot Chat, Copilot Workspace, and the expanding extension ecosystem, it now supports a genuinely agentic workflow. The same three-layer model (Context, Skills, Tools) applies, but the implementation details differ significantly.

This guide shows you how to build a fully agentic Copilot setup using the same framework.

Context in Copilot

Copilot draws context from three sources:

.github/copilot-instructions.md

• Repository-level instruction file, equivalent to CLAUDE.md
• Defines coding conventions, architectural patterns, and project-specific rules
• Automatically included in Copilot Chat context for the repository
• Checked into version control, shared across the team

Per-user VS Code settings

• Personal preferences configured in VS Code settings.json
• Language-specific instructions, formatting preferences
• Not shared with the team, purely personal context

Implicit context from open files

• Copilot uses currently open files, recent edits, and cursor position as context
• The more relevant files you have open, the better the suggestions
• This is automatic but can be guided by intentionally opening reference files

Skills in Copilot

Copilot's skill layer works differently from Claude Code's SKILL.md approach:

Copilot Extensions

• Third-party and custom extensions that add capabilities to Copilot Chat
• Invoked with @mention syntax (@extension-name)
• Can access external APIs, databases, and services
• Built as GitHub Apps with a conversational interface

Agent Skills Open Standard (SKILL.md)

• The same SKILL.md format used by Claude Code
• Portable skill definitions that work across tools adopting the standard
• Copilot can read SKILL.md files as context to understand available workflows

Built-in slash commands

• /fix — Propose a fix for problems in the selected code
• /explain — Explain how the selected code works
• /tests — Generate unit tests for the selected code
• /doc — Add documentation comments

Tools in Copilot

Copilot's tool layer combines MCP support with deep GitHub integration:

MCP Support

• Copilot Chat supports MCP servers for connecting to external data sources
• Configuration through VS Code settings
• Same MCP protocol as Claude Code, enabling shared server configurations

Native GitHub Integrations

• Issues: Reference and create issues directly from Copilot Chat
• Pull Requests: Generate PR descriptions, review diffs, suggest changes
• Actions: Trigger and monitor CI/CD workflows
• Codespaces: Full development environment with Copilot integrated

The native GitHub integration is Copilot's strongest differentiator. If your workflow is GitHub-centric, these integrations are seamless and require no additional configuration.

Key Differences from Claude Code

Conclusion

Copilot's strength is its IDE integration and native GitHub connectivity. If your team lives in VS Code and GitHub, Copilot's agentic setup requires minimal configuration for maximum impact. The three-layer model applies equally, but the tool layer benefits from GitHub's ecosystem rather than relying entirely on MCP.