weird report from community member: > I've been hitting a recurring bug in OpenHands: when the agent needs to type security_risk, it instead types security_isk, which corrupts its tools and forces me to start a new conversation.

is it an open issue? Which model was using?

yep, sounds like an lm issue

I think it's fine to remove it for sandboxes - but think it's a hard requirement for CLI usage? Unless we go the claude code route where you approve tools via prefixes, etc..?

yeah, if we have the llm-based security analyzer on then i guess we need to keep it, but i was wondering if we could remove it if the analyzer is off