Microsoft Fabric Skills

Many AI coding assistants understand generic code patterns but lack workload-specific knowledge for cloud services. This repository packages that missing knowledge as reusable "skills": compact instruction sets that tell an AI tool which Fabric APIs, query languages, and authentication flows to use so the tool can perform real Fabric tasks reliably.

What Sets It Apart

• Skill-driven separation of concerns — skills encode the protocol (endpoints, query syntax, auth scopes) while MCP servers provide live access. So what: teams can audit and version the agent's decision logic independently of runtime connections to production data.
• Workload-focused bundles — authoring, consumption, operations, and Power BI authoring bundles group only the guidance a workflow needs. So what: you install a smaller surface for a given role (e.g., data engineering vs. dashboarding) and reduce accidental actions.
• Multi-tool support and auto-discovery — includes configs and rules for GitHub Copilot CLI, Claude Code, Cursor, Windsurf and others. So what: the same skill content can be reused across different AI coding tools without rewriting prompts.
• Designed as guidance, not an execution engine — skills are Markdown instructions; MCP servers do the live execution. So what: this keeps the repo reviewable, auditable, and safe to include in projects while enabling optional runtime connectors.

Who It's For and Trade-offs

Great fit if you manage Microsoft Fabric environments and want AI assistants to help with authoring, diagnostics, or migration planning — especially teams that use Copilot CLI or other agent-enabled IDE workflows. Look elsewhere if you need a turnkey, hosted AI agent that executes on your behalf out of the box: skills supply the knowledge model and examples but do not themselves run queries or hold credentials (MCP servers or your tooling must provide execution and tokens).

Where It Fits

Use this repo when you want reproducible, reviewable agent behavior for Fabric tasks: embed skills in developer environments, pair them with MCP servers for safe live access, and pick focused bundles to limit surface area for specific roles (data engineers, analysts, operators).