GitHub Copilot offers several customization features that let you tailor its behavior to your workflow, your team's standards, and your project's needs. Use the tables below to find the right one for your use case.
Feature overview
This table shows what each customization feature is and where it lives.
| Feature | What it is | Filename and location |
|---|---|---|
| Custom instructions | Always-on context that automatically applies to every interaction within its defined scope | .github/copilot-instructions.md (repo-wide), .github/instructions/*.instructions.md (path-specific), AGENTS.md (third-party agents), or personal/org settings via UI on GitHub |
| Prompt files | Reusable, standalone prompt template with input variables | .github/prompts/*.prompt.md |
| 自定义智能体 | Specialist persona with its own instructions, tool restrictions, and context | .github/agents/AGENT-NAME.md (repo), agents/AGENT-NAME.md in .github-private repo (org/enterprise), or user profile |
| 子代理 | Separate agent spawned by the main agent to handle delegated work in an isolated context | N/A (runtime process, not a user-configured file) |
| Agent skills | Folder of instructions, scripts, and resources that Copilot loads when relevant to a task | .github/skills/<skill-name>/SKILL.md (project) or ~/.copilot/skills/<skill-name>/SKILL.md (personal) |
| MCP servers | Connection to external systems, APIs, and databases | mcp.json (path varies by IDE), repo settings on GitHub (编码代理), or mcp-servers property in 自定义智能体 configurations |
Usage comparison
This table helps you decide which customization feature to use.
| Feature | How to trigger | Best for | Example use cases |
|---|---|---|---|
| Custom instructions | Automatic | Standards, guidelines, or expectations that apply broadly across a context | Enforce coding standards, accessibility rules, review checklists |
| Prompt files | Manual: reference directly in chat or use the prompt file picker | Focused single tasks you run once with different inputs each time | Generate unit tests, run a code review checklist |
| 自定义智能体 | Manual: select from the agent dropdown in your IDE, on GitHub, or in Copilot 命令行界面（CLI） | Projects or processes with distinct stages that need specialized capabilities or strict handoffs | React reviewer agent, read-only auditing agent |
| 子代理 | Automatic, or reference a 子代理 directly in your prompt | Complex subtasks that should run in isolation from the main agent | Codebase research, running test suites |
| Agent skills | Automatic: chosen by Copilot when relevant to your prompt | Multi-step workflows with bundled assets that should be loaded as needed | GitHub Actions failure debugging, deployment procedures, release note drafting |
| MCP servers | Automatic, or ask for a specific tool by name | Tasks that require access to external tools or real-time data | Manage issues and PRs (GitHub MCP server), automate browser testing (Playwright MCP server) |
IDE and surface support
This table shows which customization features are supported in each IDE and surface. For the full Copilot feature matrix, see Copilot 特征矩阵.
GitHub recommends using the latest stable IDE, Copilot 命令行界面（CLI）, and Copilot extension versions to get the best Copilot experience.
Key:
- ✓ = supported
- ✗ = not supported
- P = under preview
| Feature | VS Code | Visual Studio | JetBrains IDEs | Eclipse | Xcode | GitHub.com | Copilot 命令行界面（CLI） |
|---|---|---|---|---|---|---|---|
| Custom instructions | ✓ | ✓ | P | P | P | ✓ | ✓ |
| Prompt files | ✓ | ✓ | P | ✗ | P | ✗ | ✓ |
| 自定义智能体 | ✓ | ✗ | P | P | P | ✓ | ✓ |
| 子代理 | ✓ | ✗ | P | P | P | ✗ | ✓ |
| Agent skills | ✓ | ✗ | P | ✗ | ✗ | ✓ | ✓ |
| MCP servers | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |
For a detailed breakdown of which types of custom instructions are supported in each IDE and surface, see 支持不同类型的自定义说明.
Further reading
- 自定义库—a curated collection of examples