data-goblin/power-bi-agentic-development

power‑bi‑agentic‑development

What it is – A marketplace of Claude Code / GitHub Copilot plugins that teach AI agents (Claude, Copilot, etc.) how to work with Microsoft Power BI, Fabric and related tooling. The repo ships a collection of skills, agents, hooks and optional UI mods that can be installed individually or as a whole marketplace.

Key concepts

• Agentic development – using LLM‑driven agents to build, manage and optimise artefacts (semantic models, reports, pipelines, etc.).
• Skills – declarative instruction files that give an agent domain knowledge and workflow steps (e.g., create-pbi-report, tmdl).
• Agents – autonomous subprocesses that can run long‑running or multi‑step tasks (e.g., pbip-validator, deneb‑reviewer).
• Hooks – deterministic scripts that run after a tool action to validate files (DAX reference checks, PBIR schema validation, RDL structure checks, etc.).
• Mods – optional UI panes for Claude Code that surface report trees, model metadata, etc.

How to install

1. Register the marketplace in Claude Code or Copilot CLI:

   claude plugin marketplace add data-goblin/power-bi-agentic-development
   # or
   copilot plugin marketplace add data-goblin/power-bi-agentic-development
   

2. Install the specific plugin you need, e.g.:

   claude plugin install tabular-editor@power-bi-agentic-development
   # or
   copilot plugin install data-goblin/power-bi-agentic-development:plugins/pbip
   

   (Only the marketplace root is a catalog; you must pick a child plugin.)
3. Verify with /plugin list and /skills list inside the respective CLI.

Main plugin groups

Plugin	Type	Notable items
goblin‑mode	Skills	help-me-get-started, improve-my-agent-setup – onboarding and health‑check for your agentic setup
tabular‑editor	Skills, Commands, Agent	BPA rule creation, C# scripting, TE CLI automation, /suggest-rule command
pbi‑desktop	Skill, Agent, Hooks	Connect to Power BI Desktop, capture live DAX queries, automatic DAX reference and metadata validation
pbip	Skills, Agent, Hooks	Author/validate PBIP project files, TMDL, PBIR, structural validation
reports	Skills, Agents, Mod	End‑to‑end report authoring (create-pbi-report), design canon, theme handling, reviewers for Deneb, SVG, Python, R
paginated‑reports	Skill, Hook	Author/validate RDL paginated reports, PDF/Excel rendering
custom‑visuals	Skills	Build Deneb, Python, R, SVG, and .pbiviz custom visuals

Typical workflow

1. Install the marketplace and the plugins relevant to your project (e.g., pbip for PBIP files, reports for report authoring).
2. Invoke a skill via the CLI or a Claude Code command, e.g. /skill create-pbi-report.
3. The skill guides the agent through the steps, calling underlying tools (Tabular Editor CLI, pbir CLI, Fabric commands).
4. Hooks automatically validate artefacts after each write, surfacing errors before they reach production.
5. Optional agents run in the background (e.g., query‑listener) to capture live DAX queries or continuously validate a PBIP project.

Why it matters

• Turns repetitive Power BI/Fabric tasks (model authoring, report styling, validation) into LLM‑driven assistants that can be invoked with natural‑language commands.
• Provides a single source of truth for best‑practice rules, validation schemas and automation scripts, reducing manual error.
• Works with both Claude Code and GitHub Copilot CLI, letting teams choose their preferred LLM platform.

Maturity & maintenance

• Weekly releases; version numbers like 26.40.7.
• A breaking transition occurred between versions 26.26‑26.38; pin to 26.25 or earlier for stable behaviour.
• GPL‑3.0 licensed, community‑maintained with detailed install/uninstall notes for Windows path‑length issues and known Copilot CLI bugs.

Getting started

• Watch the short YouTube walkthrough linked in the README.
• Begin with goblin-mode’s help-me-get-started skill to bootstrap your environment.
• Add project‑specific plugins only when needed to keep the LLM context small.

---

All information above is taken directly from the repository’s README; no external assumptions have been added.