Claude Code skill for project archiving. Scans an entire project and generates a comprehensive summary document (
over.md) with architecture, features, problems solved, and a rebuild prompt.
When you finish a project, invoke /over and it will:
- Scan the project directory structure, dependencies, configs, and source code
- Analyze API routes, data models, and core business logic
- Extract development history from git commits
- Generate a structured
over.mdcontaining:- Project overview and tech stack
- Directory structure with annotated key files
- Core feature descriptions
- Development problems and solutions (from git history and code analysis)
- API endpoints and data models
- Setup and usage instructions
- Rebuild prompt — a self-contained prompt that can recreate the entire project from scratch in a new Claude Code environment
git clone https://github.com/YOUR_USERNAME/over.git ~/.claude/skills/overgit clone https://github.com/YOUR_USERNAME/over.git ~/skills/over
# Then add to your Claude Code settings or symlink into ~/.claude/skills/Copy the SKILL.md file and scripts/ directory into your Claude Code skills directory.
At the end of any project session, say any of:
/over
项目结束了
帮我总结这个项目
生成项目文档
归档项目
The skill will ask for authorization before scanning, then generate over.md in your project root.
User: /over
|
v
Authorization Request ──→ User confirms
|
v
Project Scan
├── Directory tree (3 levels deep)
├── Dependency detection (Python/Node/Go/Rust/Java)
├── Config file discovery
├── Source code analysis (imports, routes, models)
└── Git history (recent commits, total count)
|
v
Document Generation
├── Project overview
├── Tech stack table
├── Annotated directory structure
├── Core features
├── Problems & solutions
├── API endpoints
├── Data models
├── Usage instructions
└── Rebuild prompt
|
v
Save to over.md
The generated over.md follows this template:
| Section | Content |
|---|---|
| Project Overview | 2-3 sentence summary |
| Tech Stack | Component-to-technology table |
| Project Structure | Annotated tree view |
| Core Features | Feature-by-feature descriptions |
| Problems & Solutions | Phenomenon → Cause → Fix format |
| API Endpoints | Method / Path / Description table |
| Data Models | Schema descriptions |
| Usage | Environment, install, start, config |
| Dependencies | Full dependency list with purposes |
| Rebuild Prompt | Self-contained recreation prompt |
The most important section. It contains enough detail for a fresh Claude Code instance to recreate the entire project:
- Complete feature requirements
- Technology choices and rationale
- Project structure
- Key implementation details
- API interface definitions
- Configuration requirements
over/
├── SKILL.md # Skill definition (trigger, instructions, template)
├── README.md # This file
└── scripts/
└── scan_project.py # Project scanning helper (directory, deps, git, source)
- No guessing — only documents what's actually found in the project
- Privacy-safe — lists env variable names, never outputs actual secrets
- Language-matched — output language follows the project's primary language
- Authentic problems — extracts real issues from git history and code, never fabricates
- Claude Code environment with skill support
- Python 3.x (for the scanning helper script)
- Git (optional, for commit history analysis)
MIT