/gsd:debug doesn't seem to perform well with GPT-5.3 Codex model

[Ehtz]

I haven't had a lot of success with /gsd:debug with GPT-5.3 Codex xhigh model. Maybe it's a bit too rigid for codex. It has been performing better at debugging without the gsd command.

Is this the case for other people?

[trek-e]

GSD workflows are primarily tested and optimized for Claude models. Non-Claude models may interpret the structured prompts differently.

v1.31.0 adds a --diagnose flag for debug (PR #1409, pending merge) which separates diagnosis from fix — letting the model analyze the problem without immediately attempting a repair. This may work better with GPT models that struggle with the combined diagnose+fix flow.

In the meantime, /gsd:set-profile quality ensures the strongest available model is used for debugging.

---

Generated by Claude Code