[Skill] Build Project

xulihang

Well-Known Member
Licensed User
Longtime User
This skill finds a suitable B4JBuilder.exe to build for the current project.

Put it in the skills folder to use (e.g. ~\.claude\skills).
 

Attachments

It would be very useful, for everyone, I believe, if those with more expertise could create a mini-guide on how to harness the power of the latest LLMs within programming environments like B4J, B4A, and B4i
After all, using ChatGPT or Claude via prompts is something well within everyone's reach these days

It is a different story, however, when it comes to leveraging these systems via agents capable of generating code, attempting to compile it, and if compilation fails correcting it and retry the compilation phase
That second scenario, which requires configuring various systems to work together, is not always something everyone can easily handle
 
I don't write skill for the model, the model write it himself. I just answer to the questions he gives me. Sometimes I just give it a one sentence tips so he can remember and follow next time.
 
So, are you talking about using an LLM via the command prompt?
No, what I meant was figuring out how to use it via an agent

I describe the system, and the agent should be able:
1- to generate the code (like detailed description)
2- test it (with test described on a document)
3- and correct the code if it doesn't produce the expected result, then goto point 2
4- say 'finish'
 
So, are you talking about using an LLM via the command prompt?
No, what I meant was figuring out how to use it via an agent

I describe the system, and the agent should be able:
1- to generate the code (like detailed description)
2- test it (with test described on a document)
3- and correct the code if it doesn't produce the expected result, then goto point 2
4- say 'finish'
yes, I am talking about using coding agent but I don't use Claude.

I am boring today and vibe coding a new library for B4X.

Steps:
  1. I wrote my message on B4X forum.
  2. I copied the exact message of post #1 and paste to https://chatgpt.com/
  3. When I reach a point (around 3-4 conversations) as you see in that Chit Chat, I asked it to generate a prompt.
    1792081610248.png

  4. Continue in the AI coding agent.
    1792081762828.png


    Because he told me he can't compile the project, I told him he actually can.
    1792081886181.png


    The model I am using is smart enough to do "everything" himself, including to create and do the test.
    He wrotes README.md, TODO.md, Milestone.md and sometimes Backlog.md and Naming.md
    1792081968867.png


    The rest is history... (in other words, continue to vibe code)
Bonus: I even ask him to make git commit.
 
Last edited:
I describe the system, and the agent should be able:
1- to generate the code (like detailed description)
2- test it (with test described on a document)
3- and correct the code if it doesn't produce the expected result, then goto point 2
4- say 'finish'
Without dilluting the OPs post, this is one of the things I have been trying to achieve with B4XDaisyUIKit-Skills.

The objective is that you give it a prompt and it generates an app for you that you are able to build and run in your B4A IDE. You can then extend the app with more functionality.

As a result, there are a couple of demo apps that have been featured here that are basically done with the skills with minimal user intervention. In all these only the prompt was given to an AI model after installing the skills and it produced these app from its own "thinking". I hope this helps.






 
Back
Top