| name | fabric-data-agent |
| description | This skill should be used when grounding agents in Microsoft Fabric: Fabric Data Agent, semantic models, lakehouses, warehouses, Power BI aligned metrics, governed analytical data, and natural-language data access. |
Fabric Data Agent
Fabric Data Agent is the Microsoft-native binding for analytical data grounding. Use it when the agent needs governed answers from semantic models, lakehouses, warehouses, or Power BI-aligned business definitions rather than document chunks.
When to Activate
Activate this skill when:
- The agent answers questions over Fabric semantic models, lakehouses, warehouses, or Power BI datasets.
- Business metric definitions must match governed analytical models.
- Retrieval needs structured data, measures, dimensions, filters, and aggregations.
- Users expect natural-language data analysis over enterprise analytics assets.
- Data access must inherit Fabric, Foundry IQ, and Microsoft 365 governance.
Do not activate this skill for adjacent work owned by other skills:
- Reusable enterprise knowledge bases:
foundry-iq-knowledge-layer.
- Direct unstructured document retrieval and custom indexing:
azure-agentic-retrieval.
- General memory architecture: core
memory-systems.
- Tool schema design around Fabric APIs:
foundry-tool-governance.
- Model-judge evaluation of data answers: core
advanced-evaluation.
Core Concepts
Analytical grounding is not the same as document retrieval. A document chunk can explain what a metric means, but a semantic model computes the metric. When users ask business questions, prefer the governed analytical layer that owns definitions and permissions.
Use Fabric when the agent needs:
| Need | Why Fabric fits |
|---|
| Consistent business metrics | Semantic model definitions prevent prompt-level metric drift |
| Structured aggregations | Lakehouse and warehouse data support filtered computation |
| Power BI alignment | Answers can match reports and dashboards |
| Enterprise governance | Permissions and lineage remain close to the data estate |
Practical Guidance
- Identify whether the question is analytical, documentary, or mixed.
- Route analytical questions to Fabric Data Agent or a governed Fabric-backed tool.
- Route policy or explanatory document questions to Azure AI Search or SharePoint grounding.
- Return measure names, filters, time periods, and source artifacts with the answer.
- Evaluate numerical answers with deterministic checks where possible.
- Keep data definitions in Fabric, not duplicated in prompts.
Examples
Example: mixed grounding
Question: Why did support cost rise last month?
Fabric: compute support cost by month, region, and product line.
Azure AI Search: retrieve operational notes explaining staffing changes.
Final answer: separate computed metrics from explanatory evidence.
Example: metric ownership
Avoid: prompt says "gross margin means revenue minus cost" for every agent.
Prefer: query the semantic model where gross margin is defined and governed.
Guidelines
- Use Fabric for governed analytical facts, not free-text document search.
- Preserve metric names, filters, and time windows in the response.
- Do not let the model invent business definitions that already exist in semantic models.
- Keep permission checks aligned with Fabric workspace and item governance.
- Combine Fabric with document retrieval only when the answer needs both computation and explanation.
Gotchas
- Natural language hides query ambiguity: Users may ask "revenue" without specifying gross, net, booked, or recognized. Ask or default to a governed semantic measure.
- Documents and metrics can disagree: Treat computed Fabric data as the quantitative source of truth and documents as explanatory evidence unless the business owner says otherwise.
- Preview surfaces can move: Keep Fabric-specific setup details in references or deployment docs so product changes do not rot the skill body.
Integration
- foundry-iq-knowledge-layer - Complements Fabric with reusable enterprise knowledge bases.
- azure-agentic-retrieval - Complements Fabric with direct unstructured document grounding and custom indexes.
- azure-memory-state - Stores session state and analytical context outside prompts.
- foundry-tool-governance - Packages Fabric access through curated tools or toolbox entries.
- core evaluation - Adds deterministic checks for computed outputs.
References
Skill Metadata
Created: 2026-05-24
Last Updated: 2026-05-24
Author: Agent Skills for Context Engineering Contributors
Version: azure-0.1.0
