The 5-box system that organizes your whole AI setup
Hey, it's Cindy 🌱 You commented SYSTEM, so here's the one-page cheat sheet. Everything you build with Claude lives in one of five boxes, and once you know which is which, you stop losing your own work and your automations stop breaking for reasons you can't find 🗂️
the five boxes
Where everything lives 📦
1. CLAUDE.md, the rulebook The one place for your standing instructions. Claude reads it at the start of every session: who you are, how you work, your rules and your brand. Built into Claude Code.
2. Skills and agents, your specialists Specialists you summon for a kind of job: one writes in your voice, one researches, one picks b-roll. A skill is a playbook Claude loads when it's relevant. An agent is a separate worker with its own focus and tools. Built in.
3. Directives, the recipes A recipe written in plain English: "when this happens, do these steps, in this order." It's what a specialist follows when you give it a job, and you can read and edit it without touching code. This one you set up yourself.
4. Executions, your tools Single-purpose tools that each do exactly one boring, repeatable job the same way every time: pull a transcript, cut a video. Usually a small script. The recipe reaches for these to do the actual work. You set these up too.
5. Connections, the bridges How Claude plugs into your other apps: Notion, Gmail, Drive. These are live links, properly called MCP, so Claude can read and update your real apps rather than just local files. Built into Claude, you add the connectors you want.
The shape of it Boxes 1, 2 and 5 are official Claude features you switch on. Boxes 3 and 4 are a convention you add on top. That distinction matters when you're searching docs and wondering why two of these aren't in them.
the flow
How they connect 🔗
Your CLAUDE.md sets the rules. You summon a specialist, a skill or an agent. It follows a recipe, a directive. And to actually do the work it either runs a tool, an execution, or reaches through a connection into your apps.
Rules, then specialist, then recipe, then either a tool or a live app connection.
Keep specialists smart, tools simple and reliable, and let directives do the connecting. Most broken automations come from a tool trying to be clever, or a specialist trying to do a mechanical job by hand.
the sorting rule
One question sorts anything 🧭
Is it a rule? That's CLAUDE.md.
Is it a specialist you talk to? A skill or an agent.
Is it a recipe? A directive.
Is it one mechanical job? An execution.
Is it a link to another app? A connection.
Set the boxes up once, then every time you make something, drop it in the right one using that question. The value isn't tidiness for its own sake. It's that when everything has exactly one home, Claude always knows where to look, which is why the automations stop breaking.