@JaMarr John Johnson what are your thoughts on what he does in his example, where the only thing in the CLAUDE.md it telling it to go to the AGENTS.md and follow the instructions. If I'm understanding things correctly other models will natively crack open the AGENTS.md like Claude does with its MD file? So wouldn't't doing that make this more adaptable to other model so you can with much cleaner? Or am I missing something?

I just noticed something interesting. I wanted to start a new workspace so I can reorg size for the 10000000000th time. I thought, let me start simple and put this example in thee before anything else. Just for a very simple start. I don't know if it was an update to Claude or VSC but it used to put all the words on one line. Now each word is a different line. I did try it with Codex and that did it all on one line. So I would think it was a change to Claude? Just an investing observation. But concerning. Slight changes like this can cause a huge mess if not careful. I can easily correct it and add some instructions but you don't know what you don't know until you know it LOL So getting kicked in the butt an having to correct is always a possibility. One o the reasons I always like to remain in th loop for everything. To catch little hiccup like this.

OMG not long ago I had a really bad one. I have a workspace that I thought was pretty cool. The product I support at work in a large enterprise has a API with a bunch of endpoints Every version there are updates, additions, deprecations. I have it go to the vendor doc's, create a MD for dry API endpoint with instructions, etc. Basically the webpage. Then I have it compare the different versions and give me reports on what changed. SO when a new version. is pushed I an se the change. I noticed the other day several version had the same build number. When I asked laude what that was it told me that it couldn't access the actual documents so it just made them up. WTF! So several expletives and coming later, like it cares. It was able to acces the links.

Not yet but when I get a good foundation and am working successfully on some projects that it adds value I will. I just haven’t had a use case that it’s going to add value. I was considering setting up a workspace to do research on things I do at work and help me get supporting data, validate, etc. if I understand loops correctly, or even loop of loops. Something Nate Jones spoke about recently on his podcast.

@Sam Yarbrough I have been trying to setup process to use a script. For instance, the product I support at work I started a workspace to do tasks. There are three environments, Development (dev), User Acceptance Testing (UAT) and Production (Prod). With this product there are several different components that connect to it to offer different functionality. We upgraded Dev 1st then UAT then push to Prod. So different versions at times. My workspaces sometimes have version specific tasks. Like API calls. so I have a full listing with tables, components, versions, etc. So I can read that (Claude rather) to know what version is installed. I wanted to dynamically update those tables. Claude would be able to log in via the API, pull the list of installed and connected components and get the versions and update the table. BUT that's. tokens and could make a mess of things if there is a hiccup. So rather that]n that I'm working with Claude to buildout a better config file and a script I can run when I want or instruct Claude to run it and that logs in, gets the data and updates the MD file. That way it's. not automatic and n to wasting tokens. But this will give me freedom of running the script outside of Claude or inside or even a scheduled task on a server to update on a regular schedule. That to me is how I understand the process, don't involve AI if you can do it without. If you can and want to edit the MD or create from scratch great. Always good to know how to do that and I have to work every day to learn better practices of doing that. But I will usually have Claude create the basic setup then I can edit as needed. Starting from a blank slate I get writers block. But I do often tell a model to scan and check if there are gaps in my setup and teak me where is diviates from the ICM process. Sorry did I hit what you were asking or am I babbling again? :)

@Sam Yarbrough Yup, and I am have another one that pulls down all the vendor docs for the APIs, creates a MD for each under a version folder. So each version has its one folerd, 12.2, 13, 14.4, etc. So every release I have it update with what's new. I can read those files offline or on a server where Internet is blocked. Then I have some routines to create different PDF's for Managers, Developers, etc. with update info. For developers I ask it to created a PDF that has all different es, deprecated calls, updated ones, etc. them. I also created a corporate brand workspace so It also creates the PDF per the required corporate colors fonts, wording, etc. It's kinda cool. Nobody has really received it and use it but me. But that's OK. I'm going to ry to redo it on my home system in a generic. setup so anybody using this product can download from GitHub and use. Maybe the vendor will pay me a million dollars for it. hahahahahaha