[Bug] Fable 5 cyber safeguards falsely flag legacy game engine crypto patterns as offensive content

Bug Description
False positive report: Fable 5 cyber safeguards repeatedly flag benign game development work. Context: I'm porting a legacy Windows MMORPG client (custom C++/DirectX game engine from the early 2000s) to the web using Claude Code. The work is character rendering, skeletal animation, equipment/item attachment to joints, and proprietary 3D asset format parsing. Nothing offensive, no exploits, no malware, no security tooling. What happens: The trigger is not my prompts. My messages are things like "now let's render the character's weapons and equipped items". The fallback fires the moment Claude reads a source file from the codebase (e.g. a character rendering class). The legacy engine code naturally contains patterns like Windows CryptoAPI calls (RC4/SHA1 used for asset obfuscation by the original developers), binary format parsing, and memory/buffer manipulation — standard for a 2000s game engine. The classifier appears to flag these file contents as offensive cyber content. Impact: The session falls back to Opus 4.8 on nearly every file read, making Fable 5 effectively unusable for this entire codebase. There is no way to phrase prompts around it since the trigger is the repository content itself. Request: Please tune the classifier to distinguish legacy game engine code (asset decryption, binary parsers, renderers) from actual offensive security tooling. Reading crypto calls in a 20-year-old game client is not exploit development. Environment: Claude Code, model claude-fable-5, macOS. Happy to provide session IDs or code samples if useful.

Environment Info

• Platform: darwin
• Terminal: Apple_Terminal
• Version: 2.1.198
• Feedback ID: 718eba99-85d6-4537-ad60-8e5e42ada710

Errors

[]