How I'm Scaling My CLAUDE.md Setup

I started with a single CLAUDE.md, but as my work grew, it turned into a nested structure, three separate root maps, each routing to its own set of workspaces underneath it, rather than one flat list of projects.

My setup:

Personal root map — Routes across everything in my own life and ventures: personal admin, side projects, a festival film concept, everything that's mine, not the company's.
Company capabilities registry — Routes across every AI-driven pipeline the company runs: report builders, a document builder, a video production pipeline, each one its own workspace with its own state.
Company corporate file structure — Routes across the actual document filing system for the whole organization, broken into numbered departments (executive/strategy, finance, standards and templates, and so on), each department getting its own scoped context.

What each context includes:

Personal root map — Where each life area or venture actually lives, and a firm rule that company work never gets stored here, it only passes through on its way to the company structure.
Company capabilities registry — Current state of every capability (what's live, what's dormant, what's still pending a decision), plus a routing table so a request for "the video pipeline" or "the cost report" goes straight to the right workspace without guessing.
Company corporate file structure — One CONTEXT.md per department, describing only what belongs in that department and nothing else, plus a generated register acting as the master index across all of them.

I'm structuring it this way so each level only has to know about the level directly beneath it, not everything in the whole system at once, which keeps every session focused and fast.

This follows the principles of:

One fact, one location — Department-level rules live only in that department's CONTEXT.md, never duplicated into the root map or into another department. If a rule changes, it changes in exactly one file.

Clean sessions — Because each level has its own scoped map, opening a session in the corporate file structure never pulls in personal context, and opening a session in a capability workspace never pulls in the whole department filing structure. Nothing bleeds across levels that don't need each other.

Team consistency — Since the department-level structure lives in files rather than in my head, anyone on the team who opens a department folder gets the same behavior from Claude that I would, without needing me to brief them first.

The first thing I'm testing is whether the root registries stay accurate as new workspaces and departments get added, since this whole nested system only holds together if every new shelf gets a routing row the moment it's created. That's the one habit that has to stick, or the higher levels stop being trustworthy maps.

10 comments

How I'm Scaling My CLAUDE.md Setup