Workflow for project development

[VictorKdy]

Hello,
I recently viewed your instructional videos on AI-Assisted Model-Driven App Development and found them really impressive. My current skill level is not enough to review so much AI-generated output, but I could see how it would be a great help in the near futur.

As a intermediate solution architect I'm starting to use CLI commands and visual studio code. Watching these videos led to several questions. I think having answers would be super useful for myself and others.

Visual Studio Code Project Setup: How do I/Why create a VS Code project structure the same as the one demonstrated in your video, including files like requirements.md, the specified folder hierarchy and datamodel ? Do you create one for each project you are working on ?

pacx IntelliSense/Metadata Support: Is it possible to enable IntelliSense for pacx commands within VS Code, and can this include metadata (e.g., table names, column names) from a connected environment?

Data Validation: In the second video, you implemented data validation using JavaScript, would it be more advisable to perform this validation server-side using a C# plugin?

Thank you

[neronotte]

Hi @VictorKdy,
thanks for taking a look at my videos 😊 Let me answer your questions point by point.

VSCode Project Setup

I work in a consulting firm and I have defined a standardized approach to local project folders. This allows my team to work consistently across different projects for different clients.

Basically the structure of the folder looks like this:

• 📂 .github # folder containing the standard instructions to be provided to copilot
  • 📄 copilot-instructions.md
  • 📄 powerplatform-javascript.instructions.md
  • 📄 ...
• 📂 .vscode # this is where I define the MCP Servers used locally for this project
  • 📄 mcp.json
• 📂 datamodel # this is where I document the structure of the datamodel of my app
• 📂 docs # this is the folder containing the project
  • 📂 pacx
    • 📄 ...
  • 📄 requirements.md
• 📂 solutions # this is where I archive my solutions, exported via pac solution clone or pac solution sync
  • 📂 ...
• 📂 src # this folder contains the source code that can be intended as a "deliverable" for my project
  • 📂 Plugins
  • 📂 WebResources
    • 📂 greg_
      • 📂 images
      • 📂 pages
      • 📂 scripts
    • 📄 .pacxproj
    • 📄 .wr.pacx
• 📂 test # this folder contains the source code that is not a "deliverable" for my project (unit tests, ui tests, custom console apps, ...whatever...)

I also have setup a powershell script to initialize everything, but I cannot share it... it is pretty simple to create though.

pacx IntelliSense/Metadata Support

PACX is a command line tool, thus it is accessed via terminal. Actually I don't know whether is possible to add intellisense to powershell terminals... any hint on this will be helpful!

Data Validation

This is a long discussed topic between Power Platform / Dynamics architects.
Answer is: it depends.

• Client side validation provides immediate feedback to the user. It acts in the exact moment you update a field, and the feedback can be provided directly near the field that caused the error.
• Server side validation is a must if you need to grant consistency regardless of how and where the record has been created. But it lacks this immediate feedback on the UI side, because it is triggered only when the record is saved.

So, my suggestion with the hat of an architect is: you have to do both 😊 if you want your solution to be robust.

My engineering background, however, always remembers me that doing both has side effects to be aware of:

• you have twice the logic to implement and maintain
• if something changes in the specs, you must remember to change it in two places
• if you have too many plugins (or too much logic in a given plugin), it slows down the creation of records, and this may lead to performance issues on the long run.

...long story short: it's a design decision each Power Platform architect must take basing on the actual needs of the project, and his own sensibility.

[VictorKdy]

Thank you for the detailed response. I am realizing now it is a much cleaner way to work : It allows to have everything in the same place : the data model, Js, plugins, specs, etc.

I do have followup questions based on your response.

VSCode Project Setup

Data Model Sync: If the data model is created via mcp and then updated on the platform, is there a way to sync those changes back to the local project files, and to update the Mermaid diagram?

requirements.md : Do you use a specific template or a prompt for the formatting of the requirements.md file?

mcp.json: Based on your videos, is the mcp.json file referencing https://github.com/neronotte/Greg.Xrm.Mcp?

Resources in src: Regarding the src folder, while you rely on mcp or the plugin registration tool to imported Js or plugins, how are resources like table icons are imported? Do you still import these directly from the platform UI? I read your article [here](https://dev.to/_neronotte/how-to-quickly-set-dataverse-table-icons-2lbb) but I guess you would still have to import the files in the solution.

PACX IntelliSense/Metadata Support

I have no idea if it doable directly from the github project but I'm sure it can be installed as addon for vsc.

I was asking the question because currently, I rely on a browser for the PACX wiki and use FetchXML from the XrmToolBox to fetch logical names, which is inefficient. IntelliSense would be ideal, but incorporating the PACX wiki or a way to access commands and -options within the project folder and leveraging Dataverse DevTools can be an improvement. Do you have any recommendations to improve efficency using CLI instead of the platform UI ?

Data Validation / Plugin Development

I often use Visual Studio for plugin development instead of VSC, but your video makes the latter look much faster. It looks like from VSC you avoids manually adding NuGet packages for the XRM SDK, setting up a key, and initializing tracing because it leverages a PluginBase interface instead of the standard IPlugin. Is this correct?

[neronotte]

Hi @VictorKdy ,

VSCode Project Setup

if you want a clean representation of your datamodel that can be understood by GitHub Copilot you can use pacx script commands to generate a local .ps1 file with all your datamodel definition.

For requirement.md, I don't have a specific template: every project is different.

mcp.json yes, is referencing my AppMaker MCP server. In the upcoming videos I'll show also how to manage Azure DevOps MCP server and Playwright MCP Server to work all togheter.

Icons are webresources. You can upload them via pacx webresources push (or pacx wr push, my favorite shortcut) and then apply them automatically via name-matching using pacx webresources applyIcons (or pacx wr iconize, my favorite shortcut).
If you take a look at the wiki you'll see a lot of useful commands.

PACX Intellisense/Metadata Support

You can incorporate the pacx wiki in your project folder. Simply type pacx --export. Personally i find the CLI already waaay to efficient than the UI, if you know what to do. Moreover, CLI commands can be easily generated by AI (Copilot).
Just think about the creation of the datamodel of an app. Before pacx, if you had to create a datamodel with 20 tables, each one with 10-20 colums, it could have taken a whole day. Now, it's a matter of minutes.

Data Validation / Plugin Development

Me too. I find Visual Studio more suitable for .net development, while VSCode works better on javascript/typescript/react stuff.

It looks like from VSC you avoids manually adding NuGet packages for the XRM SDK, setting up a key, and initializing tracing because it leverages a PluginBase interface instead of the standard IPlugin. Is this correct?

Actually no, the key difference is that I create the plugin project via Power Platform CLI (pac plugin init command), instead of starting from a plain .net 4.6.2 DLL project. pac plugin init scaffolds all the necessary dependencies. PluginBase then... I left it for the sake of the video. In real life scenarios I have my own version, way more complete than the standard one.

[VictorKdy]

I did my homework: I wrote a PowerShell script to create the folder structure, and did the following :

pacx auth create to create the authentication for the project
pacx solution create to create a solution in the environment
pacx project init to create the file into the project folder
pacx wr init in /src
pacx --export in /docs/pacx

Is this the correct workflow ?

Then I followed along your video: AI-Assisted Model Driven App Development (part 1).
I then asked to create a minimal datamodel.md which was great. And to generate a .ps1 script based on pacx commands. I don’t understand where the agent is supposed to source those commands. I ran pacx --export in docs/pacx, but I’m not sure the agent read the files because the commands were invalid.
-Also, please share the contents of your copilot-instructions.md because my agent didn’t generate content the same way as yours.
-How did you get the agent to use the pacx form clean in the PowerShell script without explicitly writing it? Was it pulled from copilot-instructions.md?

Using the CLI is new to me, therefore I apologize if some of these questions are basics.

Also, concerning your response above :
To implement icons, your suggested process is to first download the icon files, place them into the webresources project folder under the corresponding table name, and then execute the pacx webresources push and pacx webresources applyIcons commands?

Could you expand on why you find Visual Studio Code more suitable for .NET development?
And, you mentioned creating your own public class Plugin to replace PluginBase? Could you share some of the advantages of doing something like that?

Thank you.