// the find
data-goblin/power-bi-agentic-development
Power BI AI skills and Power BI agents for Claude Code and GitHub Copilot: a plugin marketplace of Power BI skills, subagents, and hooks for semantic models, DAX, TMDL, reports, and AI dashboards. Includes Microsoft Fabric skills and Fabric agents. Weekly updates.
A plugin marketplace that teaches Claude Code and GitHub Copilot how to work with Power BI and Microsoft Fabric — writing DAX, editing TMDL and PBIR files, running Fabric CLI commands, and validating PBIP project structure. The target user is a Power BI developer who is already using one of these coding agents and wants it to stop hallucinating Power BI-specific syntax. Not for general BI consumers.
The hook system is the most practically useful part: deterministic validation that fires after every tool use, catching broken DAX references, missing measure metadata, and invalid PBIR structure before the agent moves on. The skill library is genuinely deep — the DAX and TMDL reference files look like they were written by someone who actually knows the pain points (the dax-pitfalls.md and evaluateandlog-debugging.md files are not throwaway stubs). The marketplace format is well-structured for the Claude Code plugin system, with per-project scoping that prevents context window bloat. Weekly release cadence with a real versioning scheme (26.25) means it tracks the rapidly-changing Fabric API surface.
The breaking 26.26 reorganization is a real adoption risk — the warning says many skills will be consolidated or renamed, so anything you build around the current skill activation patterns may break in days. The hooks are shell scripts that assume a bash-compatible environment, which creates friction on native Windows without WSL (the README acknowledges a known upstream path-format bug for Windows bash users). Several skills are explicitly marked WIP or Very WIP in the plugin table, including pbi-report-design and review-report, which limits what you can actually rely on today. There is no automated test suite visible in the workflows directory for the skill content itself — the validate-plugins.yml presumably checks schema/structure, but the actual domain knowledge in the markdown files is unverified and could be subtly wrong.