Founder OS — The full system breakdown
Founder OS
How to build your operating system for work
How the whole system connects
Every layer builds on the one below. Without structure, Claude can only do what you tell it in the moment. With structure, it already knows.
The action layer
Without MCPs, Claude can only read and write local files. With MCPs, it can take real actions on real tools. Click any card to see what it unlocks.
25+ reusable agents
A skill is a markdown file with: what triggers it, what it does step-by-step, which tools it uses. Write it once. Invoke it forever.
Automated guardrails
Shell scripts that run before or after every tool use. Set them once and forget them. Claude can't break things it shouldn't touch.
git commit, rm -rf, force push, and reset --hard before they run. Claude hits a wall instead of destroying work..env and credentials files. Warns before touching CLAUDE.md, MEMORY.md, or settings.json.~/.claude/backups/ before Claude edits it. Keeps last 10 versions. Every edit is reversible.~/.claude/changes-YYYY-MM-DD.log. Full audit trail of what Claude touched.The folder structure
Claude can navigate any client, any context, any question, because every file follows the same structure. The structure comes first. The AI leverage comes after.
- Quick capture — process weekly
- Meeting notes before filing
- Ideas, links, screenshots
- /inbox-processor to clear to zero
- Clients/ — Active, Pause, Pending, Completed, Lost, Prospects
- Each client: README + Technical + Communication-Log + Progress
- Internal/ — AY Automate HQ, EY Talent, Humanoidz
- Sales/ — ICP, Proposals, Pipeline, 30-Day Strategy
- Marketing/ — Content Strategy, Brand, LinkedIn
- Operations/ — SOPs, Team, Metrics, Finance
- Technical/ — Internal automations, Stack
- AI Agents/ — architectures, patterns, tools
- n8n Workflows/ — templates, research
- Thought-Leaders/ — Hormozi, Lakajev, Lemlist
- Learning/ — courses, frameworks
- Completed projects
- Inactive areas
- Historical reference
- PDFs, images, exports
- Client documents
- Processed and organized
What Claude knows every session
Two files that load at the start of every session. One defines the rules. One holds the state. Together they replace the 30-40 minutes you'd otherwise spend re-orienting yourself.
- Loaded at the start of every Claude Code session
- Defines who you are, how you think, how Claude should work with you
- Includes: business model, ICP, client structure, tools, preferences
- Per-project. Each project can have its own CLAUDE.md
- Keep it under 200 lines. Anything longer stops being read consistently
- Best structure: 3 sections. What the project is, domain context, how to validate changes
- Claude doesn't remember last session by default. This file does
- Holds: client status, key decisions, blockers, preferences that evolved
- Updated every session as context changes
- Current entries: client status, MCP setup, tool rules, agent limitations
- Lives at project level. Scoped to this second brain
- The thing most people skip that makes everything else 10x more useful