A collection of self-contained folder-as-app blueprints for working with various data / BI tools. Each blueprint follows the 3-Layer Folder Architecture: a top-level map, rooms that load only when needed, and atomic step files that load only when their workflow runs.
- Point Claude (or any LLM that can read folders) at this directory.
- Tell it to read CLAUDE.md — the blueprint directory.
- Pick the blueprint that matches your task. Claude follows that blueprint's own
CLAUDE.mdfrom there.
Starting a new piece of work? See briefs/ — a shared hub with brief templates (linked to each blueprint) and filled example briefs to learn from.
| Blueprint | Tool | Status |
|---|---|---|
| power-bi/ | Power BI Desktop (PBIP format) | Complete — ~493 atomic files across 4 rooms (with nested sub-rooms) |
| synthetic-data/ | Synthetic / dummy data creation (Python-first) | Skeleton — 5 rooms + Power BI hand-off; rooms grow per job |
Each blueprint:
- Loads only the atomic step files needed for the current task (typical: 3–10 KB of context, not 30+ KB)
- Is portable — copy the folder into any project to work alongside it
- Uses no external state — the file system is the database, file names are the index
Create a top-level kebab-case folder. Follow the same pattern: CLAUDE.md, README.md, numbered rooms, projects/, outputs/, _examples/. Add a one-line entry to the root CLAUDE.md's "Available blueprints" list.
For the full walkthrough with workflow diagrams (intent routing, brief-first execution, 3-tier live-model preference, theme creation, hook subsystems) and an atomic-vs-plugin cost comparison, see HOW-IT-WORKS.md.
The 3-Layer Folder Architecture itself is documented in power-bi/_examples/AI_Folder_Architecture.pdf. Watch the original video: Stop Building AI Agents. Use This Folder System Instead.