Simple "autocorrect" llm response feature

[Alphaquest2005]

as you know the llm will return a variety of messages for the same purpose for example to run a mulitple operations linux systems use "&&" my windows system use ";" ... i will like a simple feature to just "auto correct" the llm so replace "&&" with ";"... also i would like the llm to choose from existing approved commands rather than just generating another slightly different command to do the same thing. a new command could get approval same way and added to approved list its used to determing the action rather than letting llm just generate slight differences and i have to add each one to approve list

[porocode11]

A good idea would also be to have an way to edit the commands, since sometime even telling the llm the right command, it still refuses to use it.

An option to edit the commands on the fly would be nice (or autocorrect like you said)

[Alphaquest2005]

Yeah its easy nothing fancy... But i was thinking about it it might sound crazy but i think it might work A big problem with vibe coding is the context. Thats what makes a experienced programmer more efficient than expert programmer on code base the experienced programmer knows... In fact the reason why we refactor is to understand the code better... The llm has the same problem... Imargine if roo code had its own ollama llm that is untrained but the llm has access to all that the programmer dose and interacts the advanced llm... This roo llm has the ability to analyse and learn the code base vectorizing it, making relationships looking for improvements. So eventually this llm can add context to user prompt to advanced llm... And learn from the user is like training in the live environment. Because as a programmer i try to teach the llm how to solve problems, or what techniques to use. But all that is lost because the advanced llm is stateless... Cuz honestly roo code is losing alot of knowledge and training by not capturing the use interaction...it can be private or shared where other personalized llms can exchange tips and tricks on problem solving. And it will make roo code much more productive... I am still thinking about it. regards, Joseph

…

________________________________ From: porocode11 ***@***.***> Sent: Wednesday, April 30, 2025 7:02 PM To: RooVetGit/Roo-Code ***@***.***> Cc: Alphaquest2005 ***@***.***>; Author ***@***.***> Subject: Re: [RooVetGit/Roo-Code] Simple "autocorrect" llm response feature (Discussion #3015) A good idea would also be to have an way to edit the commands, since sometime even telling the llm the right command, it still refuses to use it. An option to edit the commands on the fly would be nice (or autocorrect like you said) — Reply to this email directly, view it on GitHub<#3015 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AFJQ6ZPYETT5W3BZLBPLDGT24FI6VAVCNFSM6AAAAAB4BIZKMSVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTEOJZHE2DCNI>. You are receiving this because you authored the thread.Message ID: ***@***.***>