[FEATURE] Please include "Yes to all" option

[lukehutch]

Right now when Claude Code stops, it gives you 2-3 options:

1. Yes
2. [not always present] Yes, and don't ask again for similar commands in [path]
3. No, and tell Claude what to do differently (esc)

Please provide another option between 2 and 3:

2.5. Yes, and auto-respond "Yes" to all other requests for confirmation until all outstanding tasks are finished

[lukehutch]

This is particularly bad because Claude Code gets slower and slower the longer you use it, so sometimes you have to kill it and restart the session, which resets all the "Yes, and don't ask again" responses you gave in the previous session.

[dgtise25]

100% agree, when using Claude code shell for an extended period, it slows down and eventually starts to show 'offline' a lot which is a sign to me to kill the session and start again. Having a 'Yes to all' would be super helpful, especially in a development area where things are less critical in terms of breaking things . But I think a simple question at the start of every Claude code session would easily sort this even easier, i.e. 'Are we working in a development or production environment?' --> If Dev, relax all rules on running basic commands and asking for permission to do so, thus making Claude more autonomous. However, if noted as Production, apply the opposite stance and ask questions for any potential destructive commands

[lukehutch]

I would also like to have an option to encourage Claude to never stop until it has finished all the objectives I have given it. It will get 80-90% of the way there and then stop with "I did all these great things! Here are some other things that still need work" -- so do that work, for crying out loud...

[dgtise25]

+1 for the above too, seems we all have the same struggles. This is especially true for projects, for example I am doing my bug fixing run with Claude on a codebase of circa 300k lines of code, and after each section it stops to give me a nice summary of what it has done, despite the fact I have asked explicitly for a summary only at the end in order to avoid these delays! How nice it would be to give it this task, go to bed and find it all completed by the time you wake up...Utopia! :)

[ldenblyd]

i agree it's not perfect but i also think the user should have the choice to allow all, with warnings.

edit i just found the option --dangerously-skip-permissions

so to allow claude fully automated just run claude --dangerously-skip-permissions

[lukehutch]

There should also be no need to ask a user whenever Claude wants to do something that is a read-only operation, such as recursively grepping things, or even piping via xargs through another program that is also read-only.

Writing to the temp dir should also be allowed without confirmation.

[github-actions[bot]]

This issue has been inactive for 30 days. If the issue is still occurring, please comment to let us know. Otherwise, this issue will be automatically closed in 30 days for housekeeping purposes.