From problem to production in minutes. Less guessing. More building.
task-flows-video.mp4
It starts the same way every time — a messy whiteboard and a simple question: "What can we build?"
But what if you didn't need to know Microsoft Fabric? What if you only needed to know what your business needs to solve?
With the @fabric-advisor agent, you describe the problem. After a few prompts, the whole solution is mapped, everything deployed, and everyone can see how the pieces connect — so you can start building, and stop guessing.
- Teams with a problem to solve — who don't have time to become platform experts first
- Architects who want pre-validated patterns instead of starting from scratch
- Anyone new to Fabric who needs opinionated guidance without the guesswork
The agent walks you through seven pipeline phases that chain together automatically, with two moments where you stay in the driver's seat:
| Phase | What happens |
|---|---|
| Discover | You describe your problem — the agent asks questions and recommends an architecture |
| Design | A detailed architecture, with decision records explaining every trade-off |
| Test | A test plan validates against your acceptance criteria |
| Sign-Off | 🛑 You review the architecture + test plan and either approve or request revisions (max 3 cycles) |
| Deploy | The agent generates dependency-ordered scripts, CI/CD-ready — you pick live deployment or artifacts-only |
| Validate | Post-deployment checks against the test plan, with a remediation loop for any findings |
| Document | Everything synthesizes into a human-readable brief |
Each project produces two deliverables — one doc, one deployment:
_projects/your-project/
├── docs/project-brief.md ← ONE human-readable doc (problem, architecture, decisions, validation)
└── deploy/ ← CI/CD-ready deployment scripts
No sprawl — one project brief and the scripts to make it real.
Prerequisites: Python 3.11+, GitHub Copilot with agent mode
python _shared/scripts/run-pipeline.py start "My Project" --problem "describe your business problem"The pipeline runner generates agent prompts — paste each into Copilot chat. Use advance and next to progress through phases. See _shared/workflow-guide.md for the full pipeline reference.
This repository includes optional GitHub Copilot hooks in .github/hooks/ to enforce pipeline guardrails and capture runtime audit events.
01-policy.json—preToolUsepolicy checks (for example, blocks directpipeline-state.jsonedits and destructive commands)02-observability.json—sessionStart,userPromptSubmitted,postToolUse,errorOccurred,sessionEndlogging- Hook scripts live in
.github/hooks/scripts/ - Runtime logs are written to
.github/hooks/logs/(gitignored)
These hooks complement, but do not replace, CI checks in .github/workflows/ci.yml.
| Resource | Description |
|---|---|
| 13 Task Flows | Pre-defined architectures for common scenarios — batch, streaming, hybrid, ML, API, governance, and more |
| 7 Decision Guides | Opinionated guidance on the choices that matter |
| Deployment Diagrams | Visual maps showing how pieces connect and deploy in order |
| Item Registry | 45 Fabric item types with API paths, CI/CD strategies, and deployment order |
| Templates | Definition files for every deployable Fabric item type |
.github/agents/ → The @fabric-advisor orchestrator
.github/skills/ → Composable skills for each phase
decisions/ → Decision guides (the "why" behind each choice)
diagrams/ → Deployment diagrams (how pieces connect)
_shared/registry/ → Canonical registries the agent uses
_shared/templates/ → Starting point templates
_shared/lib/ → Shared Python modules
_shared/scripts/ → Pipeline CLI tools
_shared/tests/ → Test suite
_projects/ → Your project workspaces (gitignored)
We welcome contributions — especially to the Item Type Registry. See CONTRIBUTING.md for guidelines on adding task flows, decision guides, skills, and more.