I quite like this idea, but am still unhappy about how I don't have control over the interactive permission prompt.
I've been thinking about this a lot, and what I'd love to have is an ability to get these interactive prompts with a timeout (e.g. 5mins), and if I don't respond within the time, have it be treated as a reject, so it'll have to either come up with a workaround, or halt with an explanation of why the rejected approach is required for success.
I haven't looked at the code yet, but am wondering if there's way for me to implement it myself, but from an earlier exploration, I couldn't find a way to regain control after giving the user a prompt within the same terminal session.
Moreover, given how Claude Code often gives me silly interactive prompts for compound commands, when each individual part would be approved, I think that Claude Code should be given a way to know what is auto-approved and what isn't, so it'll always strive to use auto-approved commands when possible. For example, I just got a silly request to approve this, blocking it from proceeding (where each is obviously safe):
That compound command issue is infuriating, and really what drove me to make this.
FWIW, the prompt is easily adjustable in a `.claude/PERMISSION_POLICY.md` file in the project.
It's also quite easy to remix the script with claude to meet your needs. Right now it prompts the user and runs the script, so it's a race. If you added a delay in the script and increased the hook timeout in `.claude/settings.json` you should be able to accomplish what you're looking for pretty easily.
falcor84|5 hours ago
I've been thinking about this a lot, and what I'd love to have is an ability to get these interactive prompts with a timeout (e.g. 5mins), and if I don't respond within the time, have it be treated as a reject, so it'll have to either come up with a workaround, or halt with an explanation of why the rejected approach is required for success.
I haven't looked at the code yet, but am wondering if there's way for me to implement it myself, but from an earlier exploration, I couldn't find a way to regain control after giving the user a prompt within the same terminal session.
Moreover, given how Claude Code often gives me silly interactive prompts for compound commands, when each individual part would be approved, I think that Claude Code should be given a way to know what is auto-approved and what isn't, so it'll always strive to use auto-approved commands when possible. For example, I just got a silly request to approve this, blocking it from proceeding (where each is obviously safe):
> git log --oneline -1 && echo "---" && git status
> Command contains quoted characters in flag names ... Do you want to proceed?
defrex|2 hours ago
FWIW, the prompt is easily adjustable in a `.claude/PERMISSION_POLICY.md` file in the project.
It's also quite easy to remix the script with claude to meet your needs. Right now it prompts the user and runs the script, so it's a race. If you added a delay in the script and increased the hook timeout in `.claude/settings.json` you should be able to accomplish what you're looking for pretty easily.