Copilot Claude Skills for Microsoft Fabric

A curated collection of AI Agent Skills that bring Microsoft Fabric expertise directly into your development workflow. Each skill teaches Claude structured troubleshooting, performance remediation, and operational guidance for specific Fabric workloads — so you spend less time searching documentation and more time solving problems.

How It Works

These skills follow the open Agent Skills specification and use a progressive-loading architecture. When you ask Claude a question about Fabric, only the relevant skill loads into context — keeping conversations fast and focused.

You: "My MLV refresh keeps falling back to full."

Claude: [loads fabric-lakehouse-views-perf-remediate]
        -> Checks CDF status on source tables
        -> Reviews query for unsupported expressions
        -> Walks you through enabling optimal refresh

Each skill bundles:

• SKILL.md — main instructions with diagnostic checklists, step-by-step workflows, and troubleshooting matrices
• references/ — deep-dive guides, failure-pattern catalogs, and architecture notes
• scripts/ — ready-to-run SQL, PowerShell, or Python diagnostics
• templates/ — fill-in-the-blank notebooks and runbooks

Quick Start

Install All Skills (User-Wide)

git clone https://github.com/<org>/copilot-claude-skills-fabric.git
cp -r copilot-claude-skills-fabric/skills/* \
      ~/.claude/skills/

Install a Single Skill

cp -r \
  copilot-claude-skills-fabric/skills/fabric-lakehouse-perf-remediate \
  ~/.claude/skills/

Install Into a Project

cp -r \
  copilot-claude-skills-fabric/skills/fabric-spark-perf-remediate \
  .claude/skills/

After copying, skills are available immediately. Claude discovers them automatically based on your prompts — no configuration required.

Skills Catalog

The repository contains 23 skills organized across nine Fabric workload areas.

Data Agent

Configure, troubleshoot, and optimize Fabric Data Agent capabilities.

Skill	Purpose
fabric-data-agent	Core Data Agent configuration, setup, and operational workflows
fabric-data-agent-perf-remediate	Response latency, throughput optimization, and performance diagnostics
fabric-data-agent-remediate	Error resolution, connectivity failures, and configuration fixes

Data Factory

Pipeline performance and remediation for Fabric Data Factory.

Skill	Purpose
fabric-data-factory-perf-remediate	Pipeline tuning, activity duration, copy-activity diagnostics

Lakehouse

Lakehouse access control, materialized views, and query optimization.

Skill	Purpose
fabric-lakehouse-access-control	OneLake RBAC, workspace roles, table-level security, sharing
fabric-lakehouse-perf-remediate	Query performance, table maintenance, V-Order, bin-compaction
fabric-lakehouse-views-perf-remediate	MLV refresh optimization, incremental refresh, CDF, lineage

Spark and Compute

Apache Spark workloads, compute configuration, PySpark optimization, and UDF tuning.

Skill	Purpose
fabric-delta-spark-perf	Delta Lake on Spark: file sizing, Z-order, partition pruning, caching
fabric-spark-compute-perf-remediate	Pool sizing, autoscale configuration, resource allocation
fabric-spark-compute-remediate	Compute error resolution, pool failures, session management
fabric-spark-perf-remediate	Shuffle tuning, join strategies, broadcast thresholds
fabric-pyspark-perf-remediate	Pandas UDFs, Arrow integration, Python worker tuning
fabric-udf-perf-remediate	Serialization overhead, vectorized UDFs, SQL alternatives

Notebook

Fabric notebook execution and performance.

Skill	Purpose
fabric-notebook-perf-remediate	Session startup, cell latency, environment config, libraries

OneLake

OneLake storage-layer performance.

Skill	Purpose
fabric-onelake-perf-remediate	I/O throughput, shortcuts, cross-region latency, tiering

Power BI

Semantic model performance and security.

Skill	Purpose
fabric-pbi-perf-remediate	DAX optimization, model refresh, Direct Lake, rendering
fabric-pbi-security-remediate	RLS, OLS, workspace access, sensitivity labels

REST API

Fabric REST API integration and performance.

Skill	Purpose
fabric-rest-api-perf-remediate	Latency, throttling, batch ops, long-running op polling
fabric-rest-api-remediate	Auth failures, payload issues, error resolution

Real-Time Intelligence

Fabric Real-Time Intelligence workloads.

Skill	Purpose
fabric-rti-perf-remediate	Eventhouse ingestion, KQL performance, RT dashboards

Network and Security

Network connectivity and security configuration.

Skill	Purpose
fabric-network-remediate	Private endpoints, managed VNet, firewall, outbound access

Platform Monitoring

Cross-cutting performance monitoring and intelligent diagnostics.

Skill	Purpose
fabric-performance-monitoring	Capacity metrics, throttling, CU consumption, workload mgmt
fabric-iq	Fabric Copilot/IQ, natural-language diagnostics, insights

Repository Structure

copilot-claude-skills-fabric/
├── README.md
├── LICENSE
├── CONTRIBUTING.md
├── CHANGELOG.md
└── skills/
    ├── fabric-data-agent/
    │   ├── SKILL.md
    │   ├── LICENSE.txt
    │   ├── references/
    │   ├── scripts/
    │   └── templates/
    ├── fabric-data-agent-perf-remediate/
    │   └── ...
    └── ... (23 skill directories)

Each skill is self-contained. Install individual skills or the entire collection.

Skill Anatomy

Every skill follows the same internal structure:

fabric-<workload>-<category>/
├── SKILL.md           <- Entry point (< 500 lines)
├── LICENSE.txt        <- Apache 2.0
├── references/        <- Deep-dive docs (on demand)
│   ├── common-failures.md
│   └── architecture.md
├── scripts/           <- Diagnostic scripts
│   ├── health-check.sql
│   └── audit.ps1
└── templates/         <- Notebook/runbook starters
    └── diagnostic.sql

Progressive loading means Claude reads only the name and description from YAML frontmatter during discovery. The full SKILL.md body loads when your prompt matches. Scripts, references, and templates load only when Claude needs them during a workflow.

Naming Conventions

Skills follow a consistent naming pattern:

fabric-{workload}-{category}

Segment	Example Values	Meaning
fabric	—	All skills target Fabric
workload	lakehouse, spark, pbi	Fabric service area
category	perf-remediate, remediate	Type of guidance

Category definitions:

• perf-remediate — Performance diagnosis and optimization
• remediate — Error resolution and functional fixes
• access-control — Security, RBAC, and permissions
• perf — Performance best practices and patterns (no active remediation)

Requirements

• Claude Desktop, Claude Code, VS Code with Copilot, or any tool supporting the Agent Skills spec
• Microsoft Fabric workspace (any SKU) for running diagnostic scripts
• Fabric Notebook or SQL analytics endpoint for executing bundled SQL scripts

No API keys or additional dependencies are required. The bundled scripts are standard Spark SQL, T-SQL, and PowerShell.

Contributing

Contributions are welcome. See CONTRIBUTING.md for guidelines on adding new skills, updating existing content, reporting inaccuracies, and submitting diagnostic scripts.

All skills must pass the validation checklist before merge:

• Valid YAML frontmatter (name and description)
• name is lowercase with hyphens (max 64 chars)
• description states WHAT, WHEN, and keywords (max 1024 chars)
• SKILL.md body under 500 lines
• Large workflows split into references/
• Scripts include usage comments and error handling
• All resource paths are relative
• No hardcoded credentials or secrets
• Apache 2.0 LICENSE.txt included

License

This project is licensed under the Apache License 2.0. See LICENSE for the full text.

Microsoft Fabric, Power BI, OneLake, and related product names are trademarks of Microsoft Corporation. This project is not affiliated with or endorsed by Microsoft.

Acknowledgments

Built with information from Microsoft Learn and real-world Fabric operational experience. Skills follow the Agent Skills specification.