I’m curious how everyone is documenting their features, if at all.
Do you do it before implementing or after?
I suspect folks aren’t so disciplined on this but curious what your methods are. Wonder: Am I under or over documenting.
1 Like
I use the replit agent to keep current a User Guide, a project narrative, a database schema and other descriptive documents, like an Admin Guide. They are stored in the form of a markdown file that the agent creates
4 Likes
Same here. Usually after a feature is working 95% as I want, I’ll have it document in a separate MD, and usually will take it into Notion for safekeeping.
2 Likes
What does the process typically look like for this in like a mid-market org?
I’m on the not so disciplined side but will start building out support docs in like a month or two after I’ve implemented a few more features.
Coderabbit.ai has been summarizing during pull requests with architecture diagrams and descriptions, but I haven’t really stored this anywhere (and just started using it, so only 3 pull requests with this info).
You really don’t want to know what medium and large orgs do! I’ve worked for big IT consultancies and major banks. The amounts of paper produced to document everything will make your eyes water.
I hope we are passed the mountains of paper, because quite simply, you can just document as you need it. i.e., have an agent you can talk to about exactly what you need to know - at any time.
Coderabbit sounds very interesting if it documents the system as it goes. But otherwise, the answer is the same as it is for everything - just ask the agent. And you don’t even really need to tell it the details of what is required, as the LLMs have so much knowledge these days of what constitutes good documentation.
So just say who it is for, the style, their technical levels, etc. And let it generate something specifically for them, on the fly. Or, as I suggest above, let the users ask for themselves.
Yeah, it’s pretty cool. Here’s the diagram from the same PR that I posted the poem from:
This was just automatically generated at the PR.
2 Likes
This is impressive. And to Steve’s point, it’s interesting to see what larger organizations do. I’m in the big bank industry, and the results do vary. In any case, your workflow diagram is impressive and very much how we do in the banking tech world, depending on the team.
So your diagram shows the data/layer flow for when a user does a particular action. Having the Replit Agent do the work is something I often do but I’ve been wondering of how I can automate this per new features or changes.
Yeah, Coderabbit.ai you can use though GitHub (I think it can be installed in a few other like GitLab, AzureDevOps, BitBucket). It has integrations so I’m guessing you could integrate with whatever you use for documentation and have it automatically push when you merge, PR or something.
2 Likes
I wouldn’t be surprised to find a prompt could do the charts too:
Hey replit agent, analyse the user flow on the xyz pages and then: [2 options]
1. Create an image showing this flow diagramatically
2. Write it out as a detailed prompt for me to pass to chatgpt so it can produce an image showing this flow diagramatically
1 Like
I really appreciate your responses and sharing. Thank you!
Document with help by Replit Agent. Read and verify as best as possible. Update as needed. Store in special folder in files structure. Move to other sources. Create different types of documentation: admin, user, marketing, sales. You can repurpose the documentation and use in other places: chatbot for sales/marketing, user training, marketing material, marketing website, etc.
1 Like
I cover some of this in my video, i pasted a link to the time i cover it and show the structure. TLDW is after i build a successful feature or update a feature succesffully i document everything with the agent, i archive old documentation if i am doing major updates or refactors in case there is an issue we have references to how it was working vs what changes were made and the agent can refer to this depreciated code to avoid issues we used to have, i include some of this in the replit.md (or did) now it seems to update it every time so i took it upon me to keep my documentation current.
1 Like
Let me preface this by saying
I run my code locally on my machine and use augment and claude code for complex work. I use replit agent for devops work related to patches, security audit…etc.
For simple changes that only touch one page /api/db table I use OpenSpec to create the proposal for the change request. when feature is built / tested and I’m happy with the work I run archive to move the specs to archive.
For complex changes that touch code that is re-used in multiple places (try to minimize this) I use specKit. It helps with creating and reviewing the documentation, dividing up the work in phases…etc.
I’m not updating docs I treat the documents as proposals but not as how the system works. I didn’t find need to update all docs.
You can google openspec and speckit and see how you can run it locally with claude/augment or other agents.