AgentSkillsCN

specter-review

借助代码库知识,为代码评审提供丰富的上下文信息。

SKILL.md
--- frontmatter
name: specter-review
description: Get rich context for code review using codebase knowledge

Specter Review

Use the knowledge graph to get context for reviewing code changes.

When to Use

  • Before reviewing a pull request
  • Understanding the impact of changes
  • Finding related code that might need checking
  • Investigating why a file exists or what depends on it

How It Works

Specter Review uses the knowledge graph to answer questions like:

  • What files depend on the changed code?
  • What's the history of this file?
  • How complex are the changed functions?
  • Is there related code I should also review?

Example Workflow

1. Identify Changed Files

bash
git diff --name-only main

2. Ask Specter About Each File

Use the specter agent or MCP tools directly:

Check dependencies: "What imports src/auth/login.ts? What would break if I change its exports?"

Check history: "Who wrote src/auth/login.ts? How often has it changed?"

Check complexity: "What's the complexity of the functions in src/auth/login.ts?"

Find related files: "How does src/auth/login.ts connect to src/api/routes.ts?"

3. Get Impact Assessment

Ask the specter agent: "I'm changing the User interface in src/types/user.ts. What files will be affected?"

Key Questions for Review

For New Code

  • Does this fit the existing patterns in the codebase?
  • Are the complexity levels reasonable?
  • Are the dependencies appropriate?

For Modified Code

  • What depends on this code?
  • Has this file been problematic before (high churn)?
  • Are the changes increasing complexity?

For Deleted Code

  • Is this actually unused (check dead code)?
  • What was importing it?

Using MCP Tools Directly

If you prefer direct tool access:

code
get_file_relationships: Check what imports/exports a file
get_file_history: See modification history
get_complexity_hotspots: Find complex code
get_call_chain: Trace dependencies between files
search_symbols: Find functions/classes by name

Tips

  1. Start with high-level impact: "What depends on this file?"
  2. Check historical context: Files with high churn deserve extra scrutiny
  3. Look for complexity increases: Is the change making things harder to understand?
  4. Verify dead code removal: Use get_dead_code to confirm exports are unused