Published on [Permalink]
Reading time: 2 minutes

Claude Code learns to ask less

This one is for developers and vibe coders. Everyone else, feel free to skip.

The short version: Anthropic are listening, refining their tools based on feedback, and shipping at a pace that is genuinely hard to keep up with. The tools keep getting better, which just encourages everyone to hit their servers harder.

Every time Claude Code wants to run a shell command, edit a file, or call an MCP tool, it stops and asks. The first time, this feels right. By the thirtieth time in a session, it feels like a co-worker who needs written authorisation to use the stapler.

The blunt solution has always been --dangerously-skip-permissions. The name does the work. It is the nuclear option: no prompts, no checks, full autonomy. Developers reach for it constantly because the alternative (approve every mkdir) is genuinely unusable for longer sessions.

Anthropic has now shipped something better. /fewer-permission-prompts analyses your recent Bash and MCP tool calls and generates a prioritised allowlist, written directly into .claude/settings.json. It learns from what you actually do and codifies that into a structured permission layer rather than blowing the layer away entirely.

This is the right architecture. The bypass flag exists for containers and VMs where Claude genuinely cannot cause lasting damage. The allowlist approach is for the rest of us who work on real machines with real consequences. You get the flow state without handing over the keys to the whole house.

What makes it worth noting is not just the feature itself. It is the reasoning behind it. Constant permission prompts were a documented friction point in the community for months, and the response is not “here is a bigger override switch.” It is “here is a smarter default that shrinks the problem by understanding your actual workflow.”

That is how tooling should evolve. Calibrated, not capitulated.


Sources:

✍️ Reply by email