[1123581321]

A classifier is probably nice for those who currently allow --dangerously-skip-permissions, but it's not for those who have been trying to only allow the right commands to always run. It only lowers the odds of something bad happening. Maintaining a massive allowlist that parses nested bash commands is safer. (I do this. It fits in a 2MB binary that runs on a hook, and it includes what I've put in Claude's allowlist after parsing and tokenizing nested bash.)

[splitbrain]

Is that available somewhere? Maybe as a blog post on how you set this up?

[1123581321]

I have it on Github. https://github.com/michaeldhopkins/safe-chains

All of the commands and flags are listed here. https://github.com/michaeldhopkins/safe-chains/blob/main/COM...