Repo Prompt Blog

To answer your question directly "it depends". There are a lot of different subtasks to coding. Defining requirements, sketching/discussing architecture, coming up with a technical plan, executing it, creating tests, etc. Different interfaces are better at different things. Cursor/Windsurf agentic flows can be great at well defined tasks that are directly actionable within the abstractions of the existing codebase. If you want to build new things, you need to be a little more methodical, and this can mean using chat interfaces to have a dialogue with AI. I use RepoPrompt [0] + Cursor Tab Complete (since it's easily the best on the market), but I don't use the in-IDE cursor/windsurf agentic flows unless I have a small task I need doing where I can describe it perfectly. RepoPrompt is an underhyped context selection tool that includes - A prompt formatter so you can easily dump it + prompts into chat windows - A built-in diff format prompt template that gets auto-inserted so you can auto-apply the resulting changes. - A unique "Chat" mode that's an API frontend for multi-turn conversations with dynamic code context, templated prompting, etc. - An incredible "delegate edit" feature where the results of your conversation get sent to cheaper/smaller LLMs to actually enact. The cost savings and increase in coherency are unreal. It's not unusual for a single turn to generate something like 50 specific edits across a dozen files that are completely coherent. Right now my main…

I have some question that is relevant that the author or someone knowledgeable and have tried those might answer. Does XML work better when feeding context to LLM or plain text or something like json? Did anyone try or have a theory about difference (as an input to the LLM not about processing the files/folders)? Also wouldn't this be a good WASM usage or even just plain webpage? I know that processing large codebases like chrome/Linux would be better served via something else like Go program but would this even be useful? The context window of our best models is the real limit here and not the serialization speed of the context. using RAG is different league and out of consideration here of course. I liked repoprompt [1] features during beta test but they then moved to $200 price tag which is hard for me to justify for my use case. Also it is Mac only and I mainly develop on Linux which is why I think something like webapp or WASM will be the more suitable option. I maybe wrong on several points here but no problem, I welcome any correction or explanation [1] https://repoprompt.com

If you want to find an advocate for Codex that can give a pretty good answer as to why they think it's better, go ask Eric Provencher. He develops https://repoprompt.com/. He spends a lot of time thinking in this space and prefers Codex over Claude, though I haven't checked recently to see if he still has that opinion. He's pretty reachable on Discord if you poke around a bit.

One approach to improving CC's search is to use Repo Prompt https://repoprompt.com

Not a startup, and doesnt help you still have to choose, but I paid 200usd for RepoPrompt (macOs app) its a very niche app, and havent used it as much since buying it, but there's that https://repoprompt.com/

https://www.repoprompt.com is better. You need more granular control if you're planning to use this in real large codebases.

How does this compare to a tool like RepoPrompt? https://repoprompt.com

Related: https://repoprompt.com

Have you seen Repo prompt yet? https://repoprompt.com/