Maintained by Arc’IT AI.
Turn a business question into a Power BI model and report you can check and maintain. The system provides skills for defining measures, building the model, designing reports and validating the files, with records of each result.
business question → definitions and data → model → report → decision → proof
Start from the project brief, resolved business context, or an approved Agentic Design System handoff. This repository owns the Power BI solution after the handoff while preserving the decision, data, security, refresh, and publish boundaries.
- For a Power BI outcome, resolve relevant gaps in workspace/PROJECT-BRIEF.md using existing context.
- Use the repository-local skill index to select a
route, beginning with
.agents/skills/agentic-powerbi-system/SKILL.mdfor System work. - Keep business definitions next to their measure, format, interpretation, and limitation.
- For model/report builds, establish the model before the report.
- Make each report page answer one named business question.
- Validate PBIP/PBIR/TMDL edits before opening or claiming completion; publishing also requires its own authority and evidence.
npm run doctor
npm run validate
npm test
npm run audit -- --scope both
npm pack --dry-run- One primary System skill for durable routing and run evidence.
- Specialist skills for Power BI, PBIP/PBIR/TMDL, DAX, reports, Fabric, and validation.
- A dependency-free PBIP/PBIR/TMDL validator and valid/invalid fixtures.
- A standard-library tracer for success, failure, recovery, and deliberate example promotion.
| Need | Route |
|---|---|
| System routing, prior runs, and evidence | .agents/skills/agentic-powerbi-system/SKILL.md |
| Business question, KPI, and decision | .agents/skills/powerbi/SKILL.md |
| PBIP/PBIR/TMDL structure and safe edits | .agents/skills/pbip/SKILL.md |
| Measures and semantic logic | .agents/skills/dax/SKILL.md |
| Report pages, visuals, layout, and filters | .agents/skills/report/SKILL.md |
| Fabric or Power BI Service work | .agents/skills/fabric/SKILL.md |
| Completion proof | .agents/skills/validation/SKILL.md |
| Periodic accumulated-state health | .agents/skills/system-audit/SKILL.md |
- Deterministic validation runs mechanical PBIP/PBIR/TMDL and repository checks against a named target.
- Power BI/domain audit applies the
powerbicontract to business meaning, model design, DAX, shaping risk, report storytelling, and ownership. - Semantic or visual review uses specialist judgment or native tools to assess behavior that deterministic checks cannot prove.
- Per-change Review is the lifecycle gate for accepting a proposed change.
- Periodic System audit reads accumulated repository/workspace health,
proves its live upstream relation, and returns
PASS,FAIL, orBLOCKEDwithout repair or durable audit artifacts.
Select the periodic scope explicitly:
npm run audit -- --scope repository
npm run audit -- --scope workspace
npm run audit -- --scope bothThe audit uses current repository and workspace truth. It references the existing domain-audit and validation contracts and does not require another System implementation.
An approved Agentic Design System handoff can supply DESIGN.md, report
composition, tokens, and preview intent. Revalidate it against actual Power BI
capabilities, data, accessibility, and tenant constraints; the handoff is
reference input, not an execution requirement.
workspace/
├── PROJECT-BRIEF.md decision, KPI, data, and proof contract
├── briefs/ active bounded briefs
├── models/ active semantic-model work
├── reports/ active report work
├── state/ local System state
├── runs/ one durable evidence directory per route
├── history/runs.jsonl append-only relation between runs
├── learning/ durable notes intentionally retained
└── engine/ optional local validators, tracer, and tests
examples/ deliberately curated standalone proof
docs/ contract and validation notes
The primary route inspects prior records before routing work. It never copies request transcripts into the ledger. A run is valid without promotion; a curated example is written only when explicitly requested and includes its own README and proof file.
Require explicit authorization for deleting pages or visuals, rebinding reports,
changing .platform IDs, publishing, overwriting service items, or changing
RLS, refresh and access policy. Honor existing exact authorization; local edits
do not authorize protected identity or service changes. Keep tenant IDs,
credentials, connection strings, private data, and
generated local Power BI state out of Git and fixtures.
npm pack --dry-run runs a read-only seed guard. It succeeds for blank
placeholders and deliberate curated examples, and refuses to bundle mutable
workspace evidence without deleting or resetting it.
Run the full local route in a disposable checkout or temporary copy:
python3 workspace/engine/tracer.py --promote-example
python3 workspace/engine/tracer.py --simulate-failure
python3 workspace/engine/tracer.py --recover --promote-exampleThe fixed demonstration clock makes the first clean run run-0001. The failed
run remains in the ledger, and the recovery run points to that predecessor.
The source repository should remain a seed with empty operational placeholders
after validation.
MIT.