Repository navigation
v2.24.1
Follow-up answers work again, and failed updates name their culprit
Small but important one. Follow-up messages in a conversation were all dying with a ChatGPT 400 error — first message fine, everything after rejected. That's fixed. And if an update ever fails while backing up your install, the updater now logs exactly which process is holding the files, instead of a bare "in use" error.
GitHub release title: exactly v2.24.1 (version only, no descriptive suffix — the descriptive title lives in this body heading).
What changed
- Second message works again: history turns were all sent as
input_text, but the API only accepts that for your messages — the assistant's own prior replies must go asoutput_text. Every follow-up since the regression was rejected on arrival; now each turn carries the part type its role requires. - Updates diagnose themselves: when the pre-install backup move is blocked, the updater asks Windows Restart Manager who holds the install folder and writes names and process IDs into
updater.log— on every retry and in the final failure result. Plus a pre-move snapshot: leftover Nex/WebView2 processes and registered antivirus. Next time an update stalls, the log tells us what to fix instead of guessing.
Bug Fixes
- Chat follow-ups rejected (HTTP 400
input_text): fixed as above; covered by a new unit test posting a user → assistant → follow-up sequence and asserting part types.
Commit Log
7e04057fix(chat): use output_text parts for assistant history turns3c11720feat(updater): log locking processes when backup move is blocked59a78dechore(release): bump to v2.24.1