Repo Analyzer
An AI coding agent skill for deep architectural analysis of open-source projects. Generates professional-grade architecture reports with design insights, trade-off analysis, and Mermaid diagrams.
Compatible with Claude Code, Codex, OpenClaw, and any AI coding agent that supports the skills format.
中文文档
Related Articles
Quick Install
npx (Recommended)
npx skills add yzddmr6/repo-analyzer
Plugin Marketplace
/plugin marketplace add yzddmr6/repo-analyzer
/plugin install repo-analyzer@repo-analyzer
Manual (Git Clone)
# macOS / Linux
git clone https://github.com/yzddmr6/repo-analyzer.git ~/.claude/skills/repo-analyzer
# Windows
git clone https://github.com/yzddmr6/repo-analyzer.git %USERPROFILE%\.claude\skills\repo-analyzer
Features
- Architecture-level analysis — Focuses on why things are designed the way they are, not just what the code does
- Adaptive report structure — No fixed template; chapters are designed based on each project's unique characteristics
- Parallel subagent analysis — Spawns multiple agents to analyze core modules concurrently for large codebases
- Competitive positioning — Compares design philosophy and technical trade-offs against similar projects
- Mermaid diagrams — Architecture overviews, data flows, and per-module sequence diagrams throughout the report
- Interactive workflow — Asks targeted questions based on project traits before diving into analysis
Usage
Simply ask Claude Code to analyze a project:
分析项目 https://github.com/astral-sh/ruff
分析一下 ollama/ollama 这个仓库的架构
对比分析 express 和 fastify
The skill also accepts owner/repo shorthand, full GitHub/GitLab/Gitee URLs, or local paths.
Trigger Keywords
The skill activates automatically when you mention:
分析项目 分析仓库 分析 GitHub 项目分析 源码分析 架构分析 代码分析 学习这个项目 研究这个框架 看看这个库怎么实现的 对比两个项目 项目评测 框架评测
Note: The skill outputs reports in Chinese by default. If you ask in another language, it follows your language.
Analysis Modes
After scanning the codebase, the skill asks you to choose a depth level:
| Mode | Core Module Coverage | Secondary Coverage | Best For |
|---|
| Quick | >= 30% | >= 10% | Getting a high-level overview |
| Standard (default) | >= 60% | >= 30% | Regular architecture analysis |
| Deep | >= 90% | >= 60% | Studying every design decision |
How It Works
- Clone & Scan — Clones the repo (or uses a local path), counts effective lines of code by module
- External Research — Web searches for reviews, comparisons, and architecture discussions; crawls the official website
- Adaptive Q&A — Generates targeted questions based on project characteristics, not a fixed checklist
- Dynamic Report Structure — Designs chapter layout based on your answers and project traits
- Parallel Deep Analysis — Spawns subagents for each core module, analyzing them concurrently
- Cross-Validation — Verifies conclusions across modules, checks code read coverage
- Multi-Source Fusion — Merges research, module analyses, and insights into a cohesive narrative
- Final Report — Outputs a single Markdown file with Mermaid diagrams
Report Output
The final report is saved as a single Markdown file at:
~/repo-analyses/{project-name}-{date}/ANALYSIS_REPORT.md
Every report includes (adapted per project):
- Problem Context — What problem does this solve? Why do existing solutions fall short?
- Competitive Positioning — Design philosophy differences vs. alternatives (not feature checklists)
- Project Overview — Architecture at a glance
- Deep Module Analysis — Why > What analysis with trade-offs and industry comparisons
- Evaluation & Takeaways — Honest assessment of strengths, weaknesses, and lessons learned
- Architecture Diagrams — Mermaid charts for system overview and per-module flows
Optional Dependencies
The skill works with Claude Code's built-in tools out of the box. For enhanced research capabilities, these MCP servers are optional:
File Structure