Use GitHub Copilot code review across the pull request lifecycle

Introduction

Copilot 代码评审 is most effective when you use it as part of a pull request workflow. This tutorial shows how to choose between manual and automatic reviews, request reviews early on draft pull requests, and decide when to trigger re-review before merge.

You will also learn when to use customization, skills, MCP servers, and runner options to tailor Copilot's reviews to your needs.

Work through the pull request lifecycle

Choose a review model and set up early feedback for draft and active pull requests.

Choose the right review model for your team

Copilot 代码评审 can run manually, when you trigger it yourself, or automatically, at one or more points in a pull request's lifecycle. Manual reviews give you control over exactly when Copilot 代码评审 weighs in. Automatic reviews ensure consistent coverage, and you can choose whether they trigger on open, on draft, or on every new push.

Review model	Best for	Why
Manual review request	High-context changes or selective use	You can trigger Copilot only when you want another perspective.
Automatic review on open pull requests	Teams that want broad, consistent coverage	Every pull request gets baseline feedback without relying on individual habits.
Automatic review on draft pull requests	Teams that want earlier feedback loops	Authors can fix issues before human review starts.
Automatic review on new pushes	Fast-moving pull requests with multiple revisions	Copilot reevaluates substantial updates as the pull request evolves.

For setup steps, see 使用 GitHub Copilot 代码审查 and 通过GitHub Copilot配置自动代码评审.

Use Copilot 代码评审 early on draft pull requests

Reviewing draft pull requests with Copilot 代码评审 gives you an early, dependable check on your code before you request a review from your team.

A useful draft workflow is:

1. Open a pull request as draft.
2. Request a review from Copilot 代码评审 under Reviewers, or enable automatic review for draft pull requests.
3. Address high-confidence feedback first, such as correctness, security, and clear maintainability issues.
4. Push updates and confirm that major comments have been resolved.
5. Mark the pull request as ready for human review.

This approach helps your team spend more human review time on design tradeoffs and product impact, instead of obvious issues that could have been caught earlier.

Tune review quality

Improve review quality by deciding when to request re-review and where customization, external context, and runners add value.

Use re-review before merge

After substantial changes, request re-review to check whether new risks were introduced while addressing earlier feedback.

Request re-review when you:

• Update multiple files across service or package boundaries.
• Change security-sensitive or data-sensitive behavior.
• Apply a large batch of suggested changes and want a final pass.

For teams with frequent updates to active pull requests, enabling automatic reviews on new pushes can reduce manual re-review requests while keeping feedback current.

Match customization to your workflow

Customization is most useful when each file type supports a review decision point in your workflow:

• Use .github/copilot-instructions.md for repository-wide review expectations that should apply to most pull requests.
• Use .github/instructions/**/*.instructions.md for agent-specific and path-specific guidance when different parts of the codebase have different standards.
• Use AGENTS.md to provide repository context that improves relevance, such as intentional patterns, architecture boundaries, or review priorities.

Keep these files focused on what reviewers need to decide during pull requests. Avoid turning them into full engineering handbooks.

For detailed customization mechanics and examples, see 使用自定义指令来激发Copilot代码评审的强大功能.

Use agent skills and MCP servers for specialized review goals

Agent skills and MCP servers are most valuable when your team needs reviews that depend on organization-specific context.

Use agent skills when you want reusable review routines, for example:

• Applying a standard checklist to migration pull requests.
• Running language- or framework-specific review passes in a monorepo.
• Prioritizing certain risk areas, such as authorization or billing logic.

Use MCP servers when reviews should reference context outside the pull request itself, for example:

• Linking pull request changes to issue, incident, or service ownership data.
• Checking expected rollout or operational context from internal systems.
• Validating implementation details against external documentation sources.

For setup details, see 为 GitHub Copilot 添加代理技能 and 为存储库配置 MCP 服务器.

Decide whether you need self-hosted runners

In many cases, standard GitHub-hosted runners are sufficient for Copilot 代码评审 and require no setup; they are configured for you by default.

Consider self-hosted or larger runners when you need:

• Better performance for heavier review workloads.
• Network access to internal systems required by your organization.

Runners are required because Copilot 代码评审 uses GitHub Actions for agentic operations, such as deeper context gathering and tool-calling. If GitHub-hosted runners are unavailable and no supported self-hosted option is configured correctly, reviews will fall back to a more limited mode.

For planning and setup details, see 为 GitHub Copilot 代码评审配置运行器.

Recommended workflows by scenario

Team scenario	Recommended workflow
Small team with frequent pull requests	Enable automatic reviews for all pull requests, review drafts for early feedback, and use manual re-review before merge when changes are substantial.
Large monorepo with mixed stacks	Enable automatic reviews and review new pushes, add path-specific instructions for each major area, and use skills for repeated review patterns.
Security-sensitive or regulated codebase	Use draft pull request reviews plus required re-review before merge, keep repository-wide security expectations in .github/copilot-instructions.md, and use MCP servers for incident or compliance context stored outside of GitHub.
Team with internal platform dependencies	Start with automatic review, then add self-hosted runners and MCP context only when internal-system access is required for useful review feedback.

Next steps

• For broader concepts, including review effort and agentic capabilities, see 关于GitHub Copilot 代码评审.
• To tune automated review quality for specific repositories, see 使用 Copilot 生成优化的评审过程.
• For excluded file types and other limits, see 被排除在 GitHub Copilot代码评审 的文件.