Overview — AtelierBID
What AtelierBID is, who it is for, and the seven-phase lifecycle it covers.
What AtelierBID is#
AtelierBID is a Power BI design and modelling toolkit that runs entirely inside your Claude Code session as a plugin. You describe what you want — a model, a measure, a report, a theme, the SQL behind a table — and it produces real, openable Power BI Project artefacts (PBIP / TMDL / PBIR / theme.json) plus review-only source text (SQL, Power Query/M).
It runs inside your Claude Code session: no Anthropic API key, no backend, no rate limits, no per-call cost. Every "agent" is a Claude Code subagent; every deterministic step is a plain Node script; every output is written to disk under output/<run-id>/.
AtelierBID is a distributable Claude Code plugin — consumed by installing it into Claude Code, not hosted or deployed.
Who it is for#
- Power BI practitioners (analysts and BI developers) who want to go from idea to a well-built project faster, without being an expert in every layer of the stack.
- Consultants and agencies producing repeatable, consistent client deliverables across many projects.
- Not for: upstream data engineers building pipelines or warehouses, or Power Platform (Canvas / Dataverse / Flow) builders — different stacks, not covered.
Why it exists#
Building a production-quality Power BI solution means crossing many disjoint, expert-only disciplines: clean star schema, relationships, correct and performant DAX, optimisation, accessible on-brand reporting, and AI-readiness. Most practitioners are strong in only one or two. AtelierBID covers the whole offline build lifecycle — from starting a project through a valid, on-brand report and into making that model Copilot/Q&A-ready — grounded so it binds to real data instead of inventing it.
The complete package#
AtelierBID ships as ten slash-command skills and 32 deterministic Node CLIs covering the full lifecycle:
| Layer | What ships |
|---|---|
Skills (/AtelierBID:*) | 10 slash-command skills covering all seven lifecycle phases |
Scripts (node scripts/*.js) | 32 Node CLIs: project scaffold, discovery, model, DAX, design, AI readiness, documentation |
Shared libraries (scripts/lib/) | Colour maths, theme builder, HTML sanitiser, TMDL parse/emit, IR→PBIR, design-system loader, etc. |
Design constants (data/design-system.json) | Audience archetypes, layout patterns, colour roles, canvas sizes, theme defaults |
The lifecycle — seven phases#
AtelierBID covers the whole build from an empty folder to a complete, AI-ready Power BI project:
| Phase | Capability | Skill |
|---|---|---|
| A · Start a project | Scaffold a folder, conventions, and an openable PBIP shell; best-practice + schema-grounding spine | CLIs |
| B · Discovery & source text | Capture KPIs and grain; author SQL + Power Query/M (review-only, never executed) | CLIs |
| C · Model | Star-schema TMDL: relationships, settings, partitions, incremental refresh, optimisation | model-star-schema |
| D · Calculation | Idiomatic DAX, time-intelligence, optimisation, EVALUATE validation | author-dax · optimise-dax |
| E · Design & report | Tokens → theme → background → mockup IR → valid PBIR | design-report · build-theme · generate-background · prototype-report |
| F · AI | Copilot/Q&A readiness (model AI-readiness) + offline Fabric data-agent config + runbook | update-metadata |
| G · Documentation | Regenerable data dictionary from the model | document-model |
A cross-cutting spine (schema-grounding contract, best-practice rule library, hybrid offline/connected lifecycle with drift guard) underpins every phase.
Key principles#
- File-first — every step reads/writes Power BI Project text (PBIP/TMDL/PBIR/
theme.json); no binary.pbixauthoring. - Grounded, never invented — generation binds to the schema contract; off-schema = a proposal, not a fabrication.
- Offline floor, optional connectivity — usable with zero auth/network/cost; live is opt-in; publish stays manual.
- Broad-audience guardrails — safe defaults, plain-English "what & why", warn-don't-fail-silently.
- Standalone + composable — each phase is its own skill and slots into one end-to-end run.
Out of scope#
- No PNG/PDF/PPTX export — outputs are file-first text and HTML previews.
- No hosted backend, API key, or cloud calls — live read/validate is an optional tier; publish/refresh stay manual.
- Not a Power Platform (Canvas / Dataverse / Flow) builder, and not an upstream ETL or pipeline tool.
Where to go next#
- Getting started — install and run your first skill and script.
- Commands — every skill and script, with usage.
- Architecture — how the package is structured and how data flows.
- Features — every capability with usage.
- Decisions — the architectural decisions on record.