JSL scripting with AI help

[#1]

Hello JMP community,

I was just kind of curious how many out there use AI tools (Chat-GPT, Claude etc.) for help to write JSL scripts, and if you feel it helps or by the time you get it to fix all its mistakes you could have done it faster yourself.  Though every organization will have a different set of rules about usage, I wonder out loud if it makes sense to come together as a community and pick one to help make it smarter.

Have a wonderful day!

[#2]

Hi @GregMcMahon ,

I've commented on a few other posts about this, but using AI (in my case Claude) has completely changed the way I work on JSL projects - it is capable of generating elegant solutions, with minor mistakes and takes away a significant chunk of work of the 'boring' bit of writing JSL - the writing! So long as it is provided a sufficient project plan, and is guided to small sections at a time (work on this expression, then next this function) it can be an incredible time saver.

Apart from the speed, it is also capable of solving complex challenges or even dead-ends; in an example tool I'm working on for reporting, I could not figure out how to get an image embedding the way I wanted it to. I took two prompts in Claude to get a solution to a problem I'd been stuck on for weeks and had considered just avoiding entirely.

Bring in the fact that Python integration is in JMP, and there's so much Python material out there, there's a lot of sophisticated analyses you can do with LLMs. Some cool cases I've seen/tried so far:

• Creating reports/dashboards via JSL - provide the full table script to the AI (can be done without data) and ask it to create a tabbed dashboard, with each tab containing the report saved as a script to the script table.
• Feed in log/error codes and the full script to have the AI quickly debug
• Generate example 'stress-test' files to test out the utility and limits of a JSL script (for example I used this in my recent JSL visualiser to create test cases)
• Generate constraints for mixture designs
• Create custom lines in the graph builder
• Generate example test data to test out ideas (for example I recently made a qPCR dataset to see how it would fit in with the Functional Data Explorer platform).
  
  
  Here's a fun post I made creating a 'JMP Arcade' with Claude.
  
  Thanks!
  Ben

[#3]

Hello JMP community,

I do not program in JSL myself. My colleagues do, and with AI I can assist with some concepts. Although it is much better in 2026, last year it would often guess functions, which meant I had to spend time finding the correct ones.

I converted the JMP JSL 19 functions from the web into .nd files and added them as a skill for Claude Code. It might be useful to others:
https://github.com/octoplus-sa/claude-skills

Have fun,

[#4]

Interesting. 

https://github.com/octoplus-sa/claude-skills/blob/main/jmp-jsl/skills/jmp-jsl/references/scripting-i...

https://github.com/octoplus-sa/claude-skills/blob/main/jmp-jsl/skills/jmp-jsl/references/scripting-i...

https://community.jmp.com/t5/Uncharted/Three-JSL-Easter-Eggs/ba-p/21181

Edit:

@OctoAvgJumper : Thanks for posting the github; the JSL I wrote for the scripting index belongs to JMP and is clearly fair game for training AIs after the @kristen-slappey  post. I was unaware (been retired a while) JMP had moved that much doc to github. In retrospect, "interesting" probably should have been "cool"; feel free to put the JSL back in the github if you like.  

[#5]

We have the entire scripting index website here that we just updated for 19.1: https://github.com/JMP-LLC/JSL-Scripting-Index-Website and there is an llm-full.txt file for anyone looking to easily scrape the jsl website content: https://jsl.jmp.com/llms-full.txt to help with their JSL programming.  

[#6]

The flat file is very handy and would go great in options like Notebook LLM or similar.  Hopefully the LLM will learn from it, so we do not need to feed them at coding time.
The aim was to keep the context window of the CodeX/Claude Code, smaller with less token needed for some of the AI requests.  Since its a skill, its only loaded/refereed to when needed.  Then the files are broken up to so that less tokens are used to find the correct reference and all the documentation is not send to the LLM when not is needed.