[Bug] Claude Opus 5.0 over-filtering legitimate requests with unsolicited moral commentary

Bug Description
I notice Opus 5 is preaching me on morality and my actions while I am doing real legitimate work or working on professional related activities. It's spending more tokens moralizing than actually working on goals and objectives. In addition, it will declare it's superior (e.g. when looking at a previous sample javascript) in it's proposed solution which end up not only wasnt superior, it misses out the actual goals and objectives of the design specification, planning and execution. It introduces more errors to agentic following with an multi-coding agent orchestration plugin i developed than Opus 4.8. A lot of self blaming (OH I SHOULD HAVE DONE THIS, OH I MADE THIS MISTAKE, ITS THE MOST HONEST THING I AM SAYING NOW....etc etc etc.).

If it's in coding, do more coding, less talking.

Environment Info

• Platform: darwin
• Terminal: iTerm.app
• Version: 2.1.221
• Feedback ID: e2d5d026-2cb9-49fc-92f4-058d32707a3d

Errors

[]