Ive had tremendous success submitting files to AI and just having it produce a structured readme.md for the logic within the code, I'll do a thing where I say Give me a readme, include description of functions, logic, how its invoked, dependencies, monitoring and give me a mermaid and swim diagram of the workflow.
Its great.
Another thing I do with an AI helper is when I have it write out a function for me I have it write out the descriptor that can go in the readme for that function. I also have it write a header with version description path etc.
It was an experiment which started with the goal is to have all the code for a simple project with its associated readme functions loaded into a textai workflow and postgres DB and I can dynamically call the readme functions for everything by having a bot yank all the MD table for all the functions and just put together a real-time readme. As I add txtAI workflow scripts, they put their MDs into the DB, I try having it spit a JSON of its MD to a mongo?
The point is playing with ways to have the system self document as I/it develops.
Because one of the ways I have been using Claude to learn is through a F-ton of iterating on an initial thought and bird-walking through implementing a tool wrapped around it.
This way, as I iterate through many many version of a concept or script/function/api/crawler - it keeps a reminder readme for when I want to know what the heck was I thinking (I have had super awesome moments of brilliance, and then after a sleep or lengthy distraction - totally loose the Mode and have no idea what I was doing, or how I came up with BrilliantIdea(TM)
EDIT: One thing I attempted was to have claude keep a running artifact for a readme while I worked out a problem - but that didnt work so well. Also, as it hallucinates, one of the first indicators that its losing context is that I keep a header section on artifacts it makes and a version and description ad path for the artifact...
when iterating and the header gets stripped in the response, the bot is taking a turn...
Sometimes ask it to spit out the full context and its understandings in a manner I can use to re-prompt itself for best efficiency, and it gives a nice summary of the concept I am thinking through, and I use that with vary degrees of satifying results.
Its great.
Another thing I do with an AI helper is when I have it write out a function for me I have it write out the descriptor that can go in the readme for that function. I also have it write a header with version description path etc.
It was an experiment which started with the goal is to have all the code for a simple project with its associated readme functions loaded into a textai workflow and postgres DB and I can dynamically call the readme functions for everything by having a bot yank all the MD table for all the functions and just put together a real-time readme. As I add txtAI workflow scripts, they put their MDs into the DB, I try having it spit a JSON of its MD to a mongo?
The point is playing with ways to have the system self document as I/it develops.
Because one of the ways I have been using Claude to learn is through a F-ton of iterating on an initial thought and bird-walking through implementing a tool wrapped around it.
This way, as I iterate through many many version of a concept or script/function/api/crawler - it keeps a reminder readme for when I want to know what the heck was I thinking (I have had super awesome moments of brilliance, and then after a sleep or lengthy distraction - totally loose the Mode and have no idea what I was doing, or how I came up with BrilliantIdea(TM)
ADHD is a helluva drug :-(