用户批评 Claude Projects 权限设计
一名 Reddit 用户认为,Claude Projects 可用一个协调者统筹 iOS、Android 和 Web 版应用,但强制自动模式、频繁权限提示及缺少远程模型控制影响了使用体验。该用户还称项目线程可能更慢,有时甚至不会回应,不过这些说法尚未得到独立验证。
这则反馈呈现了 Claude Projects 在远程协调与权限、性能及模型控制之间的用户体验取舍。
In my opinion, Projects is Anthropic's best idea and also a big miss.
The idea is great. You get one "god thread" for each project (Alex Flinn's term). A project can be your app across all its versions (iOS, Android, web) or something specialized like marketing. You talk to a single coordinator, and it gets the work done, updates you, and can even set up cron jobs. A perfect productivity booster, right?
Here's how they're ruining it:
1. You're forced into auto mode. Auto mode is safer, I get that. But power users keep running into permission prompts or getting blocked by the agent, and that's counterproductive. Bypass permissions should at least be an option.
2. Every new session needs you to tap "Allow." Why? The real issue is that Anthropic doesn't trust remote work. Every time you try to get something done remotely, permission prompts flood your work.
3. It's slow. You can feel how much slower a Projects thread is than a normal coding thread. Sometimes the thread goes to sleep, and when you send a prompt it doesn't answer at all.
4. You can't pick the model or settings remotely, even for local models. Threads feel like mini sessions instead of full-power ones. There's no model or effort picker, no context window control, no modes.
The concept is exactly what we need. The execution makes it feel like Anthropic built it and then didn't trust anyone to use it. Give us the full-power sessions and the option to opt out of the guardrails, and this becomes a killer feature.
Anyone else feeling this? Would love to hear if you've found workarounds.