how should I fix a wrongly implemented change?

[sanzoghenzo]

Hi there, first of all sorry if I got this totally wrong, I'm starting to use agentic coding and this tool and I still have to figure out how to not made them fail epically.

My setup is OpenCode + OpenSpec + GPT 4.1 (the only one provided by and allowed at my company).

I'm using it on a brownfield project, but I didn't go through the chore of building the specs from the existing codebase. (should I? maybe that's the root of all evil?)

I just created a change to add a new function that relies on concepts (read: an input object) already in the codebase. The proposal/specs/design/tasks.md files seemed ok, so I went ahead and used the apply command, but then I realized that there was an important omission (something I deemed implicit giving the function context) in the design document that made the implementation totally wrong.

What is the best way to deal with cases like this?

• should I start a new change with the corrections to be made? should I first archive the current change? how to stop the llm think that it has to document the changes or ensure backward compatibility for a change that isn't even committed?
• is there a command to "rollback" and make the llm change the design docs of the current open change? should I just ask it nicely without using commands/skills?
• Should I just git rm the code files, edit the design documents and retry with the apply?

Meanwhile, somewhere a vibe-coder with a 50k$/month revenues project is laughing at me, stuck at the very basics of this AI coding stuff... 😅

[sanzoghenzo]

Never mind, I saw multiple issues with this same question, and the answer seems to be "edit the files and re-apply", (even though those issue are still open).

It would be nice if this information clearly states that as a real world scenario....