Key points
- Developers can define repository-wide and path-specific custom instructions using markdown files.
- Agent skills allow Copilot to load specific procedures for tasks like testing or linting.
- Model Context Protocol servers enable connections to external tools such as Playwright.
The short version
- GitHub published guidance on tailoring Copilot behavior through custom instructions, agents and MCP servers.
- These features allow AI to adhere to internal architecture rules and naming conventions automatically.
- External data sources and tools can be integrated via the Model Context Protocol.
The details
GitHub released a blog post detailing how teams can configure the Copilot app to follow specific codebase standards. The company states that context is essential for AI-assisted development, noting that every repository has unique house rules. These rules include component structures, script execution paths and framework-specific conventions. Without explicit guidance, Copilot may not align with established team practices. GitHub explains that much of this knowledge can reside within the repository itself.
This approach allows the AI to pick up standards automatically. Developers no longer need to explain workflows manually for every new task. The post outlines four primary methods for tailoring Copilot behavior. Custom instructions communicate baseline conventions for the AI to follow during code changes. Custom agents provide specialized guidance for larger, more complex tasks. Model Context Protocol servers connect the AI to external tools and data sources. These mechanisms work together to expand what Copilot can do within a project.
The context
Repository custom instructions provide the baseline context for Copilot’s work. Teams can add a .github/copilot-instructions.md file to define guidance that applies across the entire repository. This file can specify preferred architecture, naming conventions and expectations for testing. GitHub notes that path-specific instruction files can also be created inside .github/instructions. These files use an applyTo field to indicate when the guidance is relevant. When Copilot works on a matching file, it uses both repository-wide and specific guidance.
This structure keeps broad rules centralized while allowing flexibility for different file types. Agent skills package repeatable procedures for specific operations. A skill is a folder containing instructions and optional scripts. Project-level skills live in .github/skills, with each skill in its own subfolder. For example, a quality-checks folder could contain a SKILL.md file explaining which scripts run tests and linting. The description helps Copilot determine when the skill applies. After updating code, Copilot can load the skill and follow the established process.
What to watch
Custom agents handle broader, specialized work that spans multiple files. An accessibility review might require inspecting components, templates and styles across the codebase. A repository-level agent profile lives in .github/agents and defines the agent’s purpose and tools. GitHub recommends using custom agents when recurring work benefits from a dedicated role.
MCP servers expand Copilot’s access beyond the repository. They connect the AI to external tools, data sources and services. A documentation server can provide framework guidance, while a server like Playwright allows browser interaction. Users can browse available MCP servers in the Copilot app or add custom ones. This integration allows Copilot to perform actions outside the immediate code files.
What to do and how to stay safe: GitHub
- Review repository structure to identify where custom instruction files should be placed for maximum coverage.
- Audit existing testing and linting scripts to ensure they are documented clearly for agent skills.
- Evaluate which external tools are necessary for development workflows and check for compatible MCP servers.
- Monitor vendor updates for changes to how Copilot interprets custom instructions and agent profiles.
Step-by-step guide: Shadow AI Mistakes That Put Corporate Data at Risk
General security guidance from the TechSec Headlines newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
How do custom instructions differ from agent skills in GitHub Copilot?
Custom instructions define baseline conventions for the entire repository or specific paths, while agent skills package repeatable procedures for specific tasks like testing.
What is the purpose of Model Context Protocol servers in this update?
MCP servers connect Copilot to external tools and data sources, expanding its capabilities beyond the repository to include services like browser automation.
Where should repository-level custom instruction files be placed?
Repository-wide instructions should be placed in a .github/copilot-instructions.md file, while path-specific files go in .github/instructions.
