Claude Fable as Your AI Operating System: Second Brain Setup with the Four C's

The Four C's Framework

Nate organizes his AI OS into four layers, built in order:

1. Context — static knowledge: who you are, your business, your background, meeting transcripts, yearly goals. Lives in files the agent can read.
2. Connections — live data: ClickUp tasks, email inbox, financial data (QuickBooks P&L). Dynamic info that changes constantly, accessed via MCP tools.
3. Capabilities — automation: skills, custom subagents, and pipelines built on top of the first two layers.
4. Cadence — scheduled runs: capabilities that execute automatically on a schedule without manual triggering.

The first two C's form the Second Brain. The last two form the AI Operating System. You must build them in this order — capabilities without context produce hallucination; cadence without capabilities produces nothing.

Step-by-Step Breakdown

1. Make the mindset shift: default to Claude Code for everything

   Close the other AI tabs (ChatGPT, custom GPTs, standalone tools). Open Claude Code — desktop app or VS Code — and make it the default interface for all work. Context and memory accumulate session over session only if you stay in one place.

2. Write your CLAUDE.md as a router, not just instructions

   Rather than only telling the agent who you are, use CLAUDE.md to point it to all your knowledge files. Include explicit file paths to your wiki, hot cache, master index, skills folder, and agents folder. The agent reads this first on every session — make it a map, not an essay.

3. Build your Context layer (static knowledge files)

   Create files covering your background, business model, YouTube channel info, client roster, recurring processes, and 2025/2026 goals. Store them at predictable paths your CLAUDE.md references. These are the files that make the second brain "know you."

4. Add Connections (live data via MCP)

   Connect real-time data sources via MCP tools: ClickUp for team tasks and commitments, Gmail or Outlook for email triage, and accounting tools for financial snapshots. These are "connections" because the data is live and changes daily — you can't bake them into static files.

5. Build Capabilities: skills and subagents

   Write skills in .claude/skills/ and custom agents in .claude/agents/ for recurring tasks — YouTube video research, client reporting, content generation. This is where the system graduates from a smart assistant to an autonomous AI OS.

6. Add Cadence: schedule skills to run automatically

   Configure skills to run on a schedule (via Claude Code's scheduling or cron) so they execute while you sleep. The goal is a system that doesn't require you sitting there pressing run — it operates independently and reports back.

7. Consolidate multi-repo work with "other worlds"

   Create an /other-worlds/ subfolder inside your main AI OS project. Move other Claude Code repos you work in frequently (book project, website, client work) into it. Now a single git push syncs everything to GitHub, and your main OS has cross-project context — it can reference your book while editing your website.

CLAUDE.md Router Template (from video)

Template shown in the video. Adapt to your project — do not copy verbatim without reviewing each line. See the OpenClaw configuration guide for more CLAUDE.md patterns.

You are [Your Name]'s executive assistant and AI operating system.
Your job is to help [him/her/them] spend less time on [operations/admin/etc.]
so [they] can focus on [core work: content creation, client work, etc.].

# Knowledge Base
- Wiki: /path/to/wiki/
- Hot cache (frequently needed context): /path/to/hot-cache/
- Master index (start here for navigation): /path/to/master-index.md

# Tools & Integrations
- API keys: /path/to/.env (never log these)
- Skills: /.claude/skills/ — [brief list of available skills]
- Agents: /.claude/agents/ — [brief list of available agents]
- Active MCP connections: [ClickUp, Gmail, QuickBooks, etc.]

# Instructions
- When looking for context, start with the master index
- When asked about live/real-time data, use the relevant MCP connection
- When asked to automate a task, route to the appropriate skill or agent
- If a file takes more than 30 seconds to find, ask me where it is
  and update the index

Folder Structure

/your-ai-os-project/
├── CLAUDE.md               ← router: points to everything below
├── .claude/
│   ├── skills/             ← recurring automated tasks
│   └── agents/             ← custom subagents with YAML front matter
├── wiki/                   ← static knowledge base
├── hot-cache/              ← frequently accessed context snippets
├── master-index.md         ← navigation map for the agent
└── other-worlds/           ← consolidated external repos
    ├── book-project/
    ├── client-site/
    └── side-project/

Gotchas & Caveats

• Pulse check for bloat: if your agent takes more than a few seconds to find a file you know instantly, the architecture needs restructuring. The master index and CLAUDE.md router should always point directly to what the agent needs.
• Architecture engineering is an evolving art — there's no single right structure. Start simple; refactor when navigation slows down.
• "Other worlds" creates a monorepo tradeoff: easier GitHub sync and cross-project context, but a larger project for the agent to index. Works best when the sub-projects are related to your main work, not just random repos.
• Build C's in order: capabilities built without solid context and connections tend to hallucinate or produce generic output. Don't skip to automation before the knowledge base is solid.

Key Takeaways

• CLAUDE.md is most powerful as a router pointing to all knowledge and tool locations — not just a system prompt.
• The Four C's (Context → Connections → Capabilities → Cadence) is a practical sequence: each layer builds on the one before it.
• The "other worlds" folder consolidates multiple repos under one project for single-push GitHub sync and unified cross-project context.
• An AI OS doesn't start with architecture — it starts with defaulting to Claude Code for all your daily work, then accumulating context over time.
• Skills and subagents in .claude/ transform the second brain from a smart assistant into a system that can run tasks autonomously on a schedule.