[Bug] Safety classifier false positives with bug-bounty skills loaded
Status Open
Reported on v2.1.288
Maintainer reply None cached
Activity 0 comments · opened Oct 3, 2026
Bug Description
Benign messages (e.g. just saying "Hi") are being stopped by the safety classifier / falling back to Opus 4.8 with a [cyber] tag. These are false positives — no cybersecurity request was made. Session has bug-bounty skills loaded, which seems to bias the classifier.
Environment Info
- Platform: linux
- Terminal: gnome-terminal
- Version: 2.1.288
- Feedback ID: ad06b53d-5eba-4603-8223-b52ea0365f03