[Feature Request] Reduce false positive safety flags for basic, non-harmful code

Bug Description
i was coding a harmless basic tour operation for cruises extremly basic and harmless its anoying to be flagged when im working on my own operations and a simple and basic code. Here's what happened in plain terms: the system runs a broad classifier over the conversation, and it pattern-matched on the kind of language this session is full of — auth tokens, session credentials, service-role keys, rotating login users, "provision," "revoke." All of that is completely legitimate (it's just how the demo-sandbox login machinery works), but out of context those words look like cybersecurity/hacking territory to a dumb filter. So it threw a [cyber] flag and, as a precaution, handed the conversation from Fable 5 to Opus 4.8.

Environment Info

• Platform: darwin
• Terminal: WarpTerminal
• Version: 2.1.252
• Feedback ID: 596b90de-1963-41f9-9040-0156648ef93e

Errors

[]