Learn OSS codebases commit by commit — build a knowledge base with AI-powered explanations

A Claude Code skill for systematic deep code reading. Analyze commits chronologically, understand how open source projects evolved, and automatically generate rich Markdown documentation.

Deep Code Reader doesn't just explain code — it builds a searchable knowledge base. Every commit analysis is saved as a Markdown file with Mermaid diagrams, design pattern breakdowns, and your own Q&A. Over time, you accumulate a structured library of architectural knowledge that you can browse, search, and annotate.

graph LR
    CC["Claude Code<br/>(analyze & explain)"]
    OD["Output Directory<br/>(browse, search, & note)"]
    CC -->|"generates Markdown"| OD

    subgraph "Output Structure"
        IDX["_index.md<br/>Project overview"]
        C1["0001-abc1234.md"]
        C2["0002-def5678.md"]
        C3["..."]
    end

    OD --- IDX
    OD --- C1
    OD --- C2
    OD --- C3

/deep-code-reader

📚 Deep Code Reader
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

What would you like to read?

> I want to read express

✅ Set up expressjs/express
📊 Total commits: 5,432

Start from the first commit?

> Yes

[Detailed analysis is displayed and saved]

> Why is this pattern used here?

[Answers your question, saves Q&A to the Markdown file]

> Next

[Next commit analysis...]

• Conversational interface — No commands to memorize; just talk naturally
• Rich analysis output — Mermaid diagrams, design patterns, Before/After comparisons
• PR context included — Discussion threads and review comments, translated and explained
• Q&A accumulation — Your questions and answers are auto-saved to each commit's file
• Flexible output — Defaults to Obsidian vault, configurable to any directory
• Session hooks — Auto-detects active projects on session start

All operations require explicit command invocation:

Every commit generates a Markdown file with:

• Basic info — PR number, author, change size
• Change summary — Core concept, Before/After
• Architecture — Visualized with Mermaid diagrams
• Code walkthrough — With design rationale
• PR discussions — Original English + Japanese translation + context
• Design patterns — Patterns and principles applied
• Before/After — Comparison of improvements
• Learning points — Key takeaways from this commit

{output_path}/
├── _index.md              # Project overview
└── commits/
    ├── 0001-abc1234.md    # Each commit analysis
    ├── 0002-def5678.md
    └── ...

Default output path: ~/obsidian-vault/deep-code-reading/{owner}-{repo}/

Custom output path:

/deep-code-reader expressjs/express --output ~/my-notes/express/

• Claude Code CLI
• Git
• GitHub MCP server (for PR information)
• Obsidian (recommended) or any Markdown editor

1. Open Manage Plugins (Claude Code settings)
2. Go to the Marketplaces tab
3. Enter Kuroakira/deep-code-reader and click Add
4. Switch to the Plugins tab and install deep-code-reader

To update or uninstall, use the same Plugins tab.

git clone https://github.com/Kuroakira/deep-code-reader.git
cd deep-code-reader
mkdir -p ~/.claude/skills
ln -s $(pwd) ~/.claude/skills/deep-code-reader

# To uninstall:
rm ~/.claude/skills/deep-code-reader

Default paths:

• Repos: ~/.claude/deep-code-reader/repos/
• Projects: ~/.claude/deep-code-reader/projects/
• Output: ~/obsidian-vault/deep-code-reading/ (configurable)

MIT