Write Your Repo for AI, Not Humans

The Technique: One Repo, Multiple Agents, AI-First Documents

Takotyan00 realized a fundamental truth: when you're solo, there's no human team to read your docs. The primary reader is the AI. So he stopped writing for humans and started writing for Claude Code.

His claude-code-life template on GitHub is open-source and immediately actionable. It's a single Git repository that houses his journal, self-analysis, project metadata, and ideas — all structured so Claude Code can query across them seamlessly.

The Three Rules for AI-First Documents

1. One fact per file. No long prose. Each file contains a single, atomic piece of information.
2. Searchable metadata in frontmatter. Every file starts with a YAML block the AI can grep or query against.
3. Name files for AI retrieval. Ask "will the AI be able to find this later?" — not "will a human understand this filename?"

Repository Skeleton

life/
├── journal/      # dialogue-style journal (year/month/day.md)
├── persona/      # self-analysis (dimensions/, chapters/, insights/)
├── projects/     # metadata for each dev project
├── ideas/        # seeds of ideas
└── .claude/
    ├── skills/            # per-purpose agents (/journal, /weekly-review, /persona)
    ├── rules/             # behavioral rules injected into agents
    └── config/context.yaml  # context shared by every skill

Why It Works: Compounding Context

Separate apps split up the one thing that should stay continuous: you. Your investing tool doesn't know your values. Your journaling app doesn't know your cognitive quirks. By bundling everything into one repo, the AI reads across all domains. The base gets stronger the more you build.

This is the opposite of a "second brain." A second brain is meant to be read back later. Here, the notes are deliberately unreadable to you. The human-facing interface is conversation — you ask Claude Code, not open files.

How To Apply It

1. Clone the template: git clone https://github.com/takotyan00/claude-code-life
2. Set up shared context: Copy .claude/config/context.example.yaml to .claude/config/context.yaml and fill in your interests and projects.
3. Use the built-in skills: Run /journal for dialogue-style journaling, /weekly-review for weekly retrospectives, /persona for self-analysis.
4. Customize rules: Adjust .claude/rules/ for behavioral tweaks. The template includes correction rules (e.g., example-understated-expression.md) to help the AI read you correctly.

What's Included in the Template

• CLAUDE.md — shared instruction template
• .claude/skills/ — three skills: journal, weekly review, persona (generalized)
• .claude/rules/ — methodology rules + dialogue behavior rules + format templates
• .claude/rules/correction-rules/ — one generalized example for correcting how the AI reads you
• .claude/config/context.example.yaml — context template for all skills
• Directory READMEs for journal/, ideas/, projects/, persona/

Try It Now

# Clone the template
cd ~
git clone https://github.com/takotyan00/claude-code-life.git life

# Set up your context
cp life/.claude/config/context.example.yaml life/.claude/config/context.yaml

# Edit context.yaml with your own interests and projects

# Start Claude Code and try a skill
claude code
# Then type: /journal

This isn't about organizing your life for a human reader. It's about building a knowledge base that Claude Code can query across domains — and the more you add, the smarter the AI becomes about you.

Source: dev.to