AgentSkillsCN

mp-review-branch

对当前分支的变更进行多代理代码审查（仅限只读，不上传至GitHub）。

SKILL.md
--- frontmatter
name: mp-review-branch
description: Multi-agent code review of current branch changes (read-only, no GitHub posting)
disable-model-invocation: true
allowed-tools: Read, Glob, Grep, Bash(git status *), Bash(git log *), Bash(git diff *), Bash(git show *), Bash(git branch --list *), Bash(git branch --show-current *), Bash(git branch -r *), Bash(git branch -a *), Bash(git branch -v *), Bash(git remote -v *), Bash(git remote show *), Bash(git describe *), Bash(git blame *), Bash(git ls-files *), Bash(git ls-tree *), Bash(git rev-parse *), Bash(git rev-list *), Bash(git merge-base *), Bash(git shortlog *), Bash(git tag --list *), Bash(git tag -l *), Bash(git config --get *), Bash(git config --list *), Bash(git stash list *), Bash(git for-each-ref *), Bash(git cat-file *), Bash(git name-rev *)

Branch Review

Perform a thorough multi-agent code review of all changes on the current branch. $ARGUMENTS

This review is READ-ONLY. It does NOT:

  • •Post to GitHub
  • •Modify any files
  • •Make commits

Arguments

  • •No argument: Auto-detect base branch (asks user if multiple candidates found)
  • •Branch name: Compare against specific branch (e.g., /mp-review-branch dev)

Phase 1: Discovery (Haiku Agent)

Launch a Haiku agent to gather context:

1a. Determine Base Branch

If $ARGUMENTS contains a branch name, use it as base. Otherwise:

  1. •Get all remote branches: git branch -r --list
  2. •Get current branch: git branch --show-current
  3. •Identify candidate base branches:
    • •Check for main, master, dev, develop branches
    • •Check parent branch: git merge-base --fork-point origin/main HEAD
    • •Check tracking branch: git rev-parse --abbrev-ref @{upstream}
  4. •If only ONE candidate found → use it
  5. •If MULTIPLE candidates found → ask user which branch to compare against
  6. •Store the selected base branch for the review

1b. Gather Changes

  1. •Get all changed files: git diff [base]...HEAD --name-only
  2. •Get diff summary: git diff [base]...HEAD --stat
  3. •Identify the tech stack by examining:
    • •package.json (Node/TypeScript/React)
    • •Cargo.toml (Rust)
    • •go.mod (Go)
    • •requirements.txt/pyproject.toml (Python)
    • •*.csproj (C#/.NET)
    • •File extensions in changes
  4. •Check for CLAUDE.md files in the repo
  5. •Return: base branch, changed files list, detected tech stack, CLAUDE.md locations

Phase 2: Parallel Review (5 Sonnet Agents)

Launch 5 parallel Sonnet agents. Each receives the changed files and tech stack info.

Agent 1: Tech Stack Best Practices

Based on detected tech stack, check for:

TypeScript/JavaScript: Strict mode, proper typing, no any abuse, modern ES features, async/await patterns

React: Hooks rules, key props, effect cleanup, memoization, side effect abuse (useEffect for derived state)

Solid.js: Signal usage, createMemo vs createEffect, Show/For/Switch components, onCleanup

Svelte: Reactive declarations, store subscriptions, lifecycle, unnecessary reactivity

Node.js: Async patterns, stream handling, error propagation, env handling

Python: Type hints, PEP 8, context managers, exception patterns

Go: Error handling, goroutine leaks, interface usage

Rust: Ownership, Result/Option, unsafe usage

Agent 2: Security Review (OWASP Focus)

  • •SQL/NoSQL injection
  • •XSS (Cross-Site Scripting)
  • •Command injection
  • •Path traversal
  • •CSRF vulnerabilities
  • •Auth/authz issues
  • •Secrets exposure
  • •Sensitive data in logs
  • •Input validation gaps

Agent 3: Performance Analysis

  • •N+1 query patterns
  • •Unnecessary re-renders
  • •Memory leaks
  • •Bundle size impact
  • •Expensive operations in hot paths
  • •Inefficient algorithms

Agent 4: Error Handling & Reliability

  • •Try/catch usage
  • •Error boundary patterns
  • •Graceful degradation
  • •User-facing error messages
  • •Retry patterns
  • •Timeout handling
  • •Race conditions

Agent 5: Code Quality & CLAUDE.md

  • •DRY violations
  • •Cyclomatic complexity
  • •Naming conventions
  • •Type safety
  • •Edge cases
  • •Test coverage gaps
  • •CLAUDE.md compliance
  • •Overcomplicated code - over-abstraction, premature optimization, convoluted logic, excessive indirection, feature creep beyond requirements

Phase 3: Confidence Scoring (Haiku Agents)

For each issue found, score confidence 0-100:

  • •0-39 (Low): Might be false positive, stylistic preference
  • •40-65 (Medium): Probably real, worth reviewing
  • •66-80 (High): Verified real issue, should address
  • •81-100 (Top): Definitely real, must fix before merge

Phase 4: Output

Output Format:

Branch Review: [branch-name]

Tech Stack: [detected technologies] Files Changed: N | Base Branch: [base]

Issues Summary Table

#IssueCategoryConfidenceLocation
1Issue nameCategoryScorefile:line

Top Priority (score > 80)

...

High Priority (score 66-80)

...

Medium Priority (score 40-65)

...

Low Priority (score < 40)

...

Summary

CategoryTopHighMediumLow
SecurityNNNN
PerformanceNNNN
Best PracticesNNNN
Error HandlingNNNN
Code QualityNNNN

Note: This review is read-only. To fix issues, copy the suggested changes and apply them manually, or ask Claude to fix specific issues in a separate conversation.