Microsoft Fabric Skills

microsoft/skills-for-fabric is a Microsoft-maintained collection of AI coding assistant skills for Microsoft Fabric developers, analysts, and operators. It packages Fabric-specific instructions for authoring, consumption, operations, migration, and end-to-end architecture work across Fabric workloads.

Use this when an assistant needs current Fabric domain context instead of generic SQL, Spark, Power BI, or Azure advice. The repository includes GitHub Copilot CLI plugin bundles, root-level instructions for compatible tools, skill folders, prompt examples, and helper scripts for registering external Fabric MCP servers.

Knowledge Freshness

Fabric APIs, workload capabilities, Copilot CLI plugin behavior, and supported MCP registration patterns can change quickly. Before applying generated commands, verify the current Microsoft Fabric documentation, the installed skills-for-fabric revision, the Fabric workspace region, and the available workload endpoints for the tenant.

Prefer the upstream README, CHANGELOG, selected skill folder, and current Microsoft Fabric docs over model memory when there is a conflict.

Retrieval Sources

Use these sources to ground an agent session:

• The upstream microsoft/skills-for-fabric README for install and bundle selection.
• The selected folder under skills/ for workload-specific workflow rules.
• package.json for project identity, license, package description, and shipped directories.
• mcp-setup/README.md for the external Fabric MCP registration boundary.
• CHANGELOG.md for recent bundle and skill changes.

Core Workflow

1. Choose the narrowest bundle or workload filter for the task.
2. Confirm the target tenant, workspace, capacity, and Fabric item before querying or authoring.
3. Authenticate with Azure CLI or the workload-specific credential path required by the selected skill.
4. Ask the agent to inspect current sources and produce a plan before generating commands, notebooks, schemas, KQL, T-SQL, reports, or deployment scripts.
5. Review the generated plan and commands manually, then run them against a development workspace or read-only role first.

Capability Scope

This capability pack helps an AI assistant reason about Microsoft Fabric workloads and implementation patterns. It can guide SQL DW, Spark/Lakehouse, Power BI, Eventhouse/KQL, Eventstreams, Dataflows Gen2, search, migration, and architecture tasks when the user provides the right workspace context and credentials.

It does not provide Fabric credentials, grant permissions, host a Fabric MCP server, guarantee current API support, or replace workspace governance and data platform review.

Production Rules

• Do not run destructive Fabric, SQL, Spark, KQL, Dataflows, Eventstream, semantic model, report, or deployment changes without human review.
• Use least-privilege roles, development workspaces, and sample or obfuscated data for first-pass validation.
• Pin or record the skills-for-fabric source revision used for production runbooks.
• Keep Azure and Fabric tokens, connection strings, workspace secrets, and customer data out of prompts and committed config.
• Treat external Fabric MCP servers as separate trust boundaries with their own authentication, logging, retention, and network exposure risks.

Bundles

The recommended install path is through GitHub Copilot CLI:

/plugin marketplace add microsoft/skills-for-fabric
/plugin install fabric-skills@fabric-collection

Focused bundles are available when the agent only needs one operating mode:

/plugin install fabric-authoring@fabric-collection
/plugin install fabric-consumption@fabric-collection
/plugin install fabric-operations@fabric-collection

You can also filter the complete bundle by workload:

/plugin install fabric-skills@fabric-collection --filter "sqldw-*"
/plugin install fabric-skills@fabric-collection --filter "spark-*"
/plugin install fabric-skills@fabric-collection --filter "eventhouse-*"

Bundle	Use It For
fabric-skills	Complete Fabric skill bundle across authoring, consumption, operations, migrations, and architecture
fabric-authoring	REST APIs, CLI automation, notebooks, T-SQL, KQL, Dataflows Gen2, Eventstreams, schemas, semantic models, and deployment
fabric-consumption	Read-only exploration, query, discovery, monitoring, catalog search, and data-understanding workflows
fabric-operations	Diagnostics, performance investigation, warehouse query insights, and slow-query analysis

Workloads Covered

The skill catalog includes Fabric-focused coverage for:

• SQL data warehouse authoring, consumption, and operations.
• Spark and Lakehouse authoring, consumption, and operations.
• Power BI report planning, design, authoring, management, and semantic models.
• Eventhouse and KQL authoring and consumption.
• Eventstreams and Dataflows Gen2 authoring and consumption.
• Fabric catalog search and Fabric IQ.
• Databricks, Synapse, and HDInsight migration scenarios.
• End-to-end medallion architecture design.

Other Agent Hosts

GitHub Copilot CLI plugin installation is the primary path, but the repository also includes compatibility files for other coding agents:

Host	Source File
Claude Code	CLAUDE.md
Codex, Jules, OpenCode	AGENTS.md
Cursor	.cursorrules
Windsurf	.windsurfrules

For these hosts, clone the repository or copy the relevant instruction content into the project or agent configuration that loads local rules and skills.

Authentication

Most useful Fabric operations require Azure and Fabric authentication. Start by signing in with Azure CLI and requesting a Fabric token:

az login
az account get-access-token --resource https://api.fabric.microsoft.com

SQL, Spark, Power BI, KQL, and other workload paths may need different endpoints, token audiences, or permissions. The selected skill should guide the agent toward the workload-specific command pattern, but humans still need to verify the target tenant, workspace, item, role, and environment before running generated commands.

MCP Setup

The repository includes an mcp-setup folder for registering an external Fabric MCP server with local AI tools. That is registration support, not a hosted MCP server implementation.

For example, the Bash helper expects an existing server URL:

./register-fabric-mcp.sh --server-url "https://your-fabric-mcp-server.com" --server-name "fabric"

Use this only after your organization or vendor has provided a Fabric MCP server and its authentication model. Store bearer tokens in environment variables or secret stores, not in prompts or committed MCP config.

Use Cases

• Ask a coding agent to design a medallion architecture for Fabric Lakehouse ingestion.
• Generate or review Fabric REST API automation for item creation and deployment.
• Investigate slow SQL warehouse queries or Spark job behavior with Fabric operations context.
• Build a Power BI semantic model or report plan from existing workspace metadata.
• Explore Eventhouse/KQL data or Eventstreams configuration with domain-specific prompts.
• Plan migrations from Databricks, Synapse, or HDInsight into Fabric.
• Register an external Fabric MCP server after the endpoint and credentials are already approved.

Safety and Privacy

Fabric skills can help an assistant reason about live enterprise data platforms. Treat generated commands as proposals until a human checks workspace scope, tenant, region, capacity, billing, RBAC, data residency, and downstream report or pipeline impact.

For exploratory work, prefer development workspaces, read-only roles, short query windows, and sample datasets. For authoring or migration work, review schema changes, notebook code, ingestion paths, deployment scripts, semantic model edits, and report changes before execution. If an external Fabric MCP server is registered, verify its URL, authentication type, logging behavior, and retention policy before sending workspace data through it.

Source Review

Verified on 2026-06-18:

• The upstream README documents Fabric skill bundles, focused authoring, consumption, and operations installs, workload filters, prompt examples, Azure authentication, MCP setup links, and compatibility files for Claude, Cursor, Windsurf, and AGENTS.md-based tools.
• package.json identifies the project as @microsoft/skills-for-fabric, MIT licensed, with files under agents, skills, and mcp-setup.
• The skills directory contains Fabric workload skill folders for SQL DW, Spark, Eventhouse/KQL, Eventstreams, Dataflows Gen2, semantic models, Power BI reports, Fabric IQ, migrations, and medallion architecture.
• The MCP setup README states that its scripts register external Fabric MCP servers and do not create or run servers themselves.

Duplicate Check

Checked current content/skills/, content/mcp/, content/tools/, open pull requests, prior PR history, and repository-wide content for Microsoft Fabric, skills-for-fabric, fabric-skills, fabric-collection, Fabric MCP, Fabric Lakehouse, and Power BI semantic model skills. Existing Microsoft entries cover general Microsoft Agent Skills, Microsoft Learn MCP, AutoGen, and Microsoft MCP Gateway. No dedicated Microsoft Fabric skills entry, source URL duplicate, or open duplicate PR was found.