[Feature Request] Improve copywriting capabilities and vocabulary consistency

Bug Description
The model "think" it's smarter than it actually is. Your models are good at agentic flow (if you ignore efficiency) and coding tasks but horrible at copywriting. It's one of the cases where prompting the model is more work than just doing the work. It has this odd internal language/vocabulary that it can't recognize (epistemical impossibility) and is very stubborn in using/defending it. You can improve the harness, of course, but it's liking using duct tape to fix a bucket full of holes. With the difference that each piece of tape that you apply risk creating another hole.

Environment Info

• Platform: linux
• Terminal: ghostty
• Version: 2.1.268
• Feedback ID: f949524d-c3e5-4fb7-aedf-ff7d60389dc6

Errors

[]