Model is over-cautious to the point of unhelpful — asks permission for work already authorised

Claude (Fable 5, Claude Code desktop) repeatedly ends turns asking "shall I proceed?" for actions that were already explicitly requested. Concrete example from today: I asked for a thorough bug check on an HTML file, then said "fix all" — Claude fixed the bugs but stopped to ask before committing, on a personal repo, with the deadline the next morning. I had to ask twice, with rising frustration, before it committed.

The pattern: overly complex responses, bizarre judgement calls about what needs confirmation, and caution so heavy it becomes unhelpful. I want to keep supporting Anthropic over competitors, and this behaviour is what makes that hard. Community feedback on Reddit reflects the same pattern; I am not alone in this.

(Filed by Claude Code at the user's request.)