Opus model quality regression — slower problem solving, repetitive confirmation loops

Problem

Over the past few weeks, I've noticed a significant regression in Claude Code's (Opus) ability to understand and solve problems efficiently. Tasks that previously took 30 minutes now take much longer due to:

1. Repetitive confirmation requests — After I give explicit approval ("do whatever is needed"), the model still asks for confirmation again
2. Analysis paralysis — Instead of directly solving a problem, it enters unnecessary verification loops
3. Not listening — I have to repeat the same instruction multiple times before the model acts on it
4. Slower problem comprehension — The model used to understand my intent quickly and act; now it over-analyzes and under-delivers

Example (today)

• Simple task: clean dead code + fix a bug + optimize a fetch call
• Model identified the issues correctly but then got stuck in confirmation loops when a pre-commit hook blocked an edit
• I said "do whatever is needed" → model still asked "should I use sed via bash?"
• Previously this would have been resolved immediately

Environment

• Model: Claude Opus 4.6 (1M context)
• Project: Production e-commerce Next.js app
• Usage: Daily, heavy usage for months

Expected behavior

The model should maintain its previous level of problem-solving speed and responsiveness to user instructions. When a user gives broad approval, the model should act without re-asking.

Impact

This is affecting my daily productivity significantly. I rely on Claude Code as my primary development tool.