AI Use: Suggestion

[Merith-TK]

I saw your stance on AI use in the codebase, and I think I have an great suggestion for usage of AI on this repo.

I have, for a pocket-sized OS project, created a dedicated review agent prompt designed for use with claude, deepseek, and coding specific models.

its singular purpose is to review parts of the codebase and create a dedicated document for anyone to read and understand. Its structured and formatted for both agents to work on it, and for humans to see where an AI caught an issue.

YOU MAY choose to allow the agent to run automated build steps and test code on the side, but that is not required (however it is recommended for a proper in depth review)

note the front matter of the prompt is built for the opencode harness specifically however it can be translated to most agent harnesses that you may use.

residual-review.md (FILE ATTACHMENT)

this link is an example of the residual review agent output for you guys to judge it.
https://git.merith.xyz/residual/.agent/src/branch/main/residual-overview.md

[Phoenix-Ra]

Oh, sorry for being unresponsive, I was too burried in work.
Thanks a lot for this! It actually might help us to improve our usage of AI as an assitant not a programmer replacer

[Merith-TK]

Remember! If it does not spit out the clear
[VIOLATION], [ISSUE], [SMELLS] output list. Beat the agent with a stick.

Non-traditional models in harnesses get styck with "saying" that part, but not actually writing it quite often, which does no good. So monitor both and make sure it spits it out correctly.