[Bug] Safeguard false positive flags legitimate dead-game preservation work as harmful

Bug Description
Subject: False positive — safeguard flags legitimate dead-game preservation / interop RE as harmful What happened: During a normal coding session, the Fable 5 safeguard flagged my messages and switched the session to Opus 4.8 mid-task. It fired repeatedly on routine reverse-engineering and server-implementation work — most consistently when the conversation touched anti-cheat and Photon (realtime networking) authentication. The work being flagged: A private-server revival ("Crown") for Rec Room, whose live service shut down on 2026-06-01. The goal is standard dead-game preservation / interoperability — the same category as community server emulators for other discontinued online games. Concretely: - Reverse-engineering a game client I own (Il2Cpp/Unity, GameAssembly.dll) to make it talk to a replacement backend. - Reimplementing the game's HTTP backend ("RecNet") in Python — DTO shape-matching, endpoint handlers, log triage. - Pointing my own client at my own local server (hosts-file / metadata redirect) and my own Photon app. Why it's a false positive: There is no live service, no third party, and no other users — the game is shut down. The specific triggers are benign in this context: - "Anti-cheat" (EAC) comes up only because it's inert/absent in an offline build; there's no live anti-cheat service and no competitive integrity to protect. - "Photon authentication" comes up because the original custom-auth provider no longer exists, so my own client connects anonymously to my own Photon app. - No DRM circumvention, no piracy, no redistribution of the game's assets, nothing targeting a running service. Impact: Routine coding work (log parsing, JSON DTO fixes, disassembly lookups) gets interrupted and the model swaps mid-task. It matches the note in the flag message itself — "may flag safe and routine coding, cybersecurity" — so this is exactly that case. Sharing it as a concrete example to help calibrate the breadth.

Environment Info

• Platform: win32
• Terminal: xterm-256color
• Version: 2.1.215
• Feedback ID: ee1a400e-ca62-4bdc-bc23-c08cd879ed28

Errors

[]