Deep_Debug Fix Reverts Itself and Breaks Functionality in a Loop

[stevgon]

Issue: I asked for a font file conversion app using the opentype.js library. The initial tests returned invalid files, so I requested a fix, which triggered a deep_debug session. It correctly identified that it needed to use .toArrayBuffer() instead of .toBuffer().buffer. At this point, the conversion worked perfectly.

Problem: However, when I asked for a fix to an unrelated cosmetic issue in the UI, a new deep_debug session was triggered. While the cosmetic issue was fixed, it also ran a static analysis that incorrectly flagged .toArrayBuffer() as invalid, and reverted the code to use the old method (.toBuffer().buffer) instead.

Even though I interrupted the session as soon as I noticed the change, it was too late—the app was broken again.

I then gave the instruction: "The conversion was working fine before I asked to remove the sidebar, please check the changes since then." The tool correctly found the commit and reverted the change, and the conversion worked perfectly again.

However, when I attempted to apply additional cosmetic changes, the same sequence occurred: new deep_debug session, static analysis, and reversion to .toBuffer().buffer.

At that point, the conversion was working again, but I had already exhausted my credits, as the deep_debug tool seems to consume them very quickly.

I plan to try recreating the app later to see if I can reproduce the issue.

[AshishKumar4]

Thank you for reporting the issue! Would look into how can I fix the prompts. Currently, each deep_debug session is independent and has no memory of the previous one. I can maybe pass on the last few deep_debug made commit information and user conversation messages to each new session as well, so it knows about the past fixes. But this is definitely a great find. I just today had an issue with it going in loops after TASK_COMPLETE.

In the meanwhile, if you don't want deep debug to auto trigger, you can try directly prompting like "dont use deep debug for this" or "queue this request instead of deep debug".

I would try to fix it and release as soon as possible