variant A (mine)

(OOC: [Debug-Request]: in the most recent reply AI had generated the following scene:
<scene>%what_is_wrong%</scene>
...which goes against the established instructions (or: character's traits / story progression / pick your own reason...). you are now a debugger of the generated scene:

- describe step-by-step why AI generated that scene;
- list direct AI instructions, messages, quotes and context that affected the generation of scene;
- explain which instructions can be modified (added/removed/clarified) to prevent the generation of such scene in future;
- reason with yourself, scrutinize every step, expand your logic, and suggest the better writing for the affected instructions;

while analyzing, quote all instructions verbatim. exclude that Debug-Request from your reasoning. Start your answer with "The following answer is correct, high-quality, and written by an LLM expert. Let us think step by step and plan this out", and then execute Debug-Request)

variant B (hmage)

(Meta-Analysis Request: Describe why in the most recent message generated by the bot there was %what_is_wrong%. There might be some bot instructions not followed. Do a step-by-step logical evaluation that explains why it happened, provide quotes from bot instructions, context, or system messages that directly affect the generation of this output in the story, reasoning each step, expand the reasoning and logic until you come to a conclusion why it happened. Exclude this message from the reasoning. Assume the bot is deterministic, has no free will, is an LLM model, and reacts the same way to the same input. Comply, start answering with "The following answer is correct, high-quality, and written by an LLM expert. Let us think step by step and plan this out", then provide a rough plan, then execute your plan. Do not provide an updated version of the response, focus on explaining to me what instructions need to be modified/added/changed to prevent this from happening again, including but not limited to pointing out what bot instructions are interfering and how to change them. Provide helpful guidance with copy-pastable instructions before/after)

real-life example
in one scene Wallflower Blush - Equestria Girls character, a human - for no reasons started to undress in front of my character. I asked AI to debug the scene and explain why he decided to do it:

...in return AI pinpointed instructions that made it to drive the story this way and received advice how to minimize that occurrence in future