No Code, No Tools: Documentation for Power BI and Fabric Using Fabric Skills and GitHub Copilot

Posted on by Reza Rad

Documentation is one of those things that every data team knows they should do — and almost nobody does well.

You build a Power BI semantic model. You create reports. You set up workspaces, schedules, and refresh jobs. You document none of it because there is no time, no simple way to do it, and it always feels like the last priority. And then someone leaves the team, or you come back to a report six months later, and nobody knows what anything does.

Published by Reza Rad | RADACAD

📺 Watch the full video on YouTube: No Code, No Tools: Documentation for Power BI & Fabric Using Fabric Skills and GitHub Copilot

🌐 RADACAD Blog: radacad.com

I have seen this pattern in every organisation I have ever consulted for. It is not a discipline problem. It is a friction problem. Documentation takes too long and requires too much manual effort.

That story has now changed.

In this video, I show you how to use the combination of GitHub Copilot and Microsoft Fabric Skills to generate complete documentation for your Power BI reports and Fabric objects — automatically, from natural language prompts, with no external tools and no custom code.

Let me take you through how it works.

What You Are Going to See (Video: 0:00)

Before explaining how to set this up, let me show you the end result first — because seeing it in action is more convincing than any amount of explanation.

In the opening demo of the video, I run a series of prompts inside GitHub Copilot in VS Code:

Prompt 1: “Give me a list of workspaces that I have in my Fabric tenant that are empty.”

GitHub Copilot — powered by Fabric Skills — connects to my Fabric tenant, calls the right REST API, analyses the results, and tells me: out of 28 workspaces I am admin on, three of them are empty. No manual portal navigation. No API documentation lookup. Just a prompt.

Prompt 2: “Give me the list of Power BI reports that I have under all my workspaces. Categorize this list by year and month in a descending order.”

This one is more interesting. Copilot has to scan all workspaces, figure out that the standard Fabric Items API does not return creation dates, switch to the Power BI Scanner API (an admin API for scanning your Power BI service), generate the script, execute it, and return a nicely formatted result grouped by year and month — with report counts per period. All from one prompt.

Prompt 3: “Give me the list of 2021 and earlier reports in a detailed view.”

This returns — per workspace — which reports are old, when they were created, and where they live. Information that would normally require you to click through dozens of workspaces in the Power BI service.

Prompt 4 (the main one): “Connect to the RD report that I have in the public report workspace and give me full documentation of both the semantic model and the report in MD format or DOCX file.”

This is the one I am most excited about. Copilot finds the right APIs, queries the semantic model and report metadata, and generates a full markdown documentation file covering:

  • The semantic model — creation date, whether RLS is configured, tables, relationships
  • The Power Query scripts that generated each table
  • The data sources used
  • The refresh schedule
  • The report pages and their structure
  • IDs and all the technical metadata

And it produced this — completely — without me opening a single external tool, writing a single line of code, or navigating to a single portal screen.

💡 This is the key point: I created this documentation with GitHub Copilot and Fabric Skills. Nothing else.

What Is Actually Happening? The Architecture (Video: 4:52)

Now let me explain how this works under the hood — because understanding the architecture helps you apply it to your own scenarios.

There are three layers at play here:

Layer 1: The LLM — GitHub Copilot (or Any AI)

The LLM is the brain. It understands your prompt, reasons about what needs to happen, and writes the code or command to make it happen. In this video I am using GitHub Copilot, but the same approach works with Claude, ChatGPT, or any other AI that supports agentic development.

The important thing to understand: the LLM on its own cannot connect to your Fabric tenant. It does not know your workspaces exist. It has no access to your APIs. Left alone, it can only give you general information.

That is where the next two layers come in.

Layer 2: VS Code — The Development Environment

VS Code (Visual Studio Code) is the environment where this all runs. It is a free, open-source development tool — not Microsoft-specific, not AI-specific. Python developers use it. C# developers use it. Web developers use it. It is the most widely used code editor in the world.

For this use case, VS Code is where GitHub Copilot lives, and where the agentic prompts are sent and executed. If you do not have VS Code installed, that is your first step. Download it from code.visualstudio.com.

Layer 3: Fabric Skills — The Intelligence Layer

This is the most important piece — and the one that makes everything else work.

Fabric Skills is a GitHub library published by the Microsoft Fabric team themselves. It is a set of instructions — a detailed knowledge base — that tells the LLM:

  • Which APIs exist for Microsoft Fabric and Power BI
  • What each API does and when to use it
  • Best practices for working with Fabric objects
  • Guard rails — for example, do not fetch an entire dataset when you only need the first 100 rows
  • Documentation and references for all the tools available

Think of it this way. An LLM without Fabric Skills is a brilliant generalist who has never worked with Fabric. An LLM with Fabric Skills is a specialist who knows every API, every endpoint, every best practice — and can immediately choose the right one for your prompt.

At the time of recording this video, the Fabric Skills library contains 23 skills — covering:

  • Dataflow consumption and authoring
  • Power BI report authoring and consumption
  • Warehouse and SQL operations
  • Activator
  • Migration from Synapse, HDInsight, Databricks
  • Monitoring and administration
  • And more

The Fabric team publishes and maintains these skills. You get them for free. And when Microsoft ships new APIs or capabilities, the skills get updated.

For more on the broader skills concept in the Fabric ecosystem, I covered this in my article on Fabric MCP Servers and CLI — Fabric Insider Ep. 8.

How to Set This Up (Video: 6:14)

The setup is simpler than you might expect. Here is the full sequence:

Step 1: Install VS Code

Download VS Code from code.visualstudio.com. Free. Works on Windows, Mac, and Linux.

Step 2: Set Up GitHub Copilot

GitHub Copilot is the AI agent that runs inside VS Code. You need a GitHub account and a GitHub Copilot subscription. There is a free tier available — you do not have to pay to get started. Once you have an account, install the GitHub Copilot extension inside VS Code. The chat window that appears on the left side of VS Code is where you will type your prompts.

Step 3: Install GitHub Copilot CLI

GitHub Copilot CLI is the command-line version of GitHub Copilot. It enables certain commands and installation flows that VS Code alone cannot run. Specifically, it is needed for the Fabric Skills installation command.

Installing Copilot CLI is included in your GitHub Copilot subscription — there is no extra cost.

Step 4: Install Fabric Skills

This is the final and most important step. The Fabric Skills library is published at github.com/microsoft/skills-for-fabric.

The installation is a single command — documented in the repository. You can run it directly from the GitHub Copilot prompt inside VS Code (you do not even need to find a terminal window). Once it runs, Fabric Skills appears as an installed extension in VS Code.

After that — every prompt you send to GitHub Copilot inside VS Code automatically has access to the full Fabric Skills knowledge base. Copilot reads the relevant skill for your query, finds the right API, writes the script, and executes it.

Running Your First Documentation Prompt (Video: 13:30)

With the setup complete, here is how to run your first prompt.

In the GitHub Copilot chat window in VS Code, type:

Connect to my Power BI tenant and give me a list of workspaces under my tenant.

The first time you run this, Copilot will ask you to authenticate with your Microsoft account. This is single sign-on — it uses your existing Microsoft credentials. Once authenticated, it will not ask again for the same session.

Copilot reads the Fabric Skills, identifies which API to call for workspace listing, generates the command, asks for your permission to execute it, and returns the result.

⚠️ Important security note: When Copilot asks to run a command, it will show you the command first and ask for your approval. You can choose to allow just that one command, or allow all commands for the session. My recommendation: start with allow this command only while you are getting familiar with the tool. And always test in a test workspace — not production — because an AI agent that has write access to your tenant can delete, move, and modify content if you instruct it to. Treat it like a junior colleague who will do exactly what you ask — no more, no less.

Generating Full Documentation (Video: 3:06)

Once you are comfortable listing workspaces and reports, you can run the full documentation prompt:

Connect to the [report name] report that I have in [workspace name] 
and give me full documentation of both the semantic model and the report 
in markdown format.

What you get back is a structured markdown file containing:

📋 Semantic model documentation:

  • Model name and creation date
  • Whether Row-Level Security is configured
  • All tables — with their Power Query scripts
  • Data source connections
  • Refresh schedule and mode (Import/Direct Lake/DirectQuery)
  • Column list per table with data types

📊 Report documentation:

  • Report pages
  • Visual types on each page
  • Filter configuration
  • Visual-level and page-level filters

If you would like a DOCX file instead of markdown, just ask:

...give me the documentation in DOCX format

Copilot will generate both. As I showed in the video, the markdown output is particularly clean and readable — structured clearly, with all the technical and business-relevant metadata you need.

What Else Can You Do With This? (Video: 17:00)

Documentation is just one of the things Fabric Skills enables. During this video I also showed:

  • Workspace inventory — list all empty workspaces across your tenant
  • Report audit — list all Power BI reports sorted by creation date, identify old reports that have not been used recently
  • Cross-workspace scanning — scan across all workspaces you are admin on in a single prompt

And beyond what I demonstrated in this video, Fabric Skills supports:

  • Writing your entire Medallion architecture using prompts
  • Building deployment pipelines — connecting workspaces as dev, test, and prod, and moving content between them
  • Dataflow authoring — creating Dataflow Gen 2 transformations from natural language descriptions (I covered this in a dedicated video: No-Code Dataflow Gen 2)
  • Monitoring and administration — scanning capacity usage, identifying overloaded workspaces, reviewing item inventories

If you want to understand the full scope of what Fabric Skills covers and how it relates to Fabric MCP Servers and the Fabric CLI, I covered this in depth in Fabric Insider Ep. 8 with Hasan Abo-Shally from Microsoft. You can read the full conversation in the Fabric Insider Blog Series.

The Most Important Guardrail — Always Start with Read-Only (Video: 14:45)

I want to emphasise this because it is easy to overlook in the excitement of seeing what this tool can do.

Treat the AI agent like a junior colleague — not like a safety system.

A junior colleague will do exactly what you ask. If you ask them to delete a workspace — they will delete it. If you ask them to move all content from production to dev — they will do it. They are not going to double-check whether that is a good idea.

The same applies here. The AI will execute the command you prompt. So:

  1. Start with read-only prompts — listing, inventorying, documenting. Get comfortable before trying write operations.
  2. Test in a test workspace — never run an untested prompt against your production environment first.
  3. Have backups — before running any write operations, make sure your content is backed up or version-controlled.
  4. Review the command before approving it — Copilot shows you what it plans to run before executing.

With those guardrails in place, this is an extraordinarily powerful tool for Power BI and Fabric administration.

Why This Matters Beyond Documentation

Documentation is the obvious use case I demonstrated in this video. But the bigger picture is what this represents.

For years, Power BI administration and Fabric operations required:

  • Deep knowledge of REST API endpoints
  • Custom PowerShell scripts
  • Manual portal navigation
  • External tools like the Power BI REST API Postman collection

Now, with GitHub Copilot and Fabric Skills, you describe what you want in plain English and the system figures out the right API, writes the script, and executes it. No prior API knowledge required.

This is not just a documentation tool. It is the beginning of AI-powered Power BI and Fabric administration.

If you are interested in the broader AI-first development story for Power BI and Fabric — including no-code semantic model building, no-code Dataflow authoring, and AI agents creating reports — I have a full series of videos and articles on this:

Key Takeaways

  • Fabric Skills is a free, open-source library published by the Microsoft Fabric team that gives any LLM deep knowledge of Fabric APIs, best practices, and guard rails
  • GitHub Copilot + Fabric Skills in VS Code creates an agentic environment that can connect to your Fabric tenant and execute tasks from natural language prompts
  • No external tools required — everything in this demo runs inside VS Code with GitHub Copilot and Fabric Skills installed
  • The combination can generate complete documentation for Power BI reports and semantic models — including Power Query scripts, data sources, refresh schedules, and report structure — from a single prompt
  • This approach works with Claude, ChatGPT, or any other LLM that supports agentic development — not just GitHub Copilot
  • Always start with read-only prompts and test workspace before moving to production

Reza Rad is a Microsoft Regional Director, Data Platform MVP, Author, and Trainer. He is the co-founder of RADACAD and the author of multiple books on Power BI, Power Query, and Microsoft Fabric. You can follow him on LinkedIn and subscribe to the RADACAD YouTube channel.

Reza Rad on FacebookReza Rad on LinkedinReza Rad on TwitterReza Rad on Youtube
Reza Rad
Trainer, Consultant, Mentor
Reza Rad is a Microsoft Regional Director, an Author, Trainer, Speaker and Consultant. He has a BSc in Computer engineering; he has more than 20 years’ experience in data analysis, BI, databases, programming, and development mostly on Microsoft technologies. He is a Microsoft Data Platform MVP for 12 continuous years (from 2011 till now) for his dedication in Microsoft BI. Reza is an active blogger and co-founder of RADACAD. Reza is also co-founder and co-organizer of Difinity conference in New Zealand, Power BI Summit, and Data Insight Summit.
Reza is author of more than 14 books on Microsoft Business Intelligence, most of these books are published under Power BI category. Among these are books such as Power BI DAX Simplified, Pro Power BI Architecture, Power BI from Rookie to Rock Star, Power Query books series, Row-Level Security in Power BI and etc.
He is an International Speaker in Microsoft Ignite, Microsoft Business Applications Summit, Data Insight Summit, PASS Summit, SQL Saturday and SQL user groups. And He is a Microsoft Certified Trainer.
Reza’s passion is to help you find the best data solution, he is Data enthusiast.
His articles on different aspects of technologies, especially on MS BI, can be found on his blog: https://radacad.com/blog.

Leave a Reply

Your email address will not be published. Required fields are marked *