@shivam_kushwaha16 Pretty granular. Harden evaluates actions at runtime instead of giving the agent blanket permission. So things like reading files, editing code, shell commands etc. can be treated differently, and higher-risk actions can still require approval while routine ones run automatically. We make the approval requirement selective enough so that you don’t have to click “allow” on every step.
Replies
The approval step is a nice safety layer for coding agents How granular are the permissions when the agent runs different actions?
Harden
@shivam_kushwaha16 Pretty granular. Harden evaluates actions at runtime instead of giving the agent blanket permission. So things like reading files, editing code, shell commands etc. can be treated differently, and higher-risk actions can still require approval while routine ones run automatically. We make the approval requirement selective enough so that you don’t have to click “allow” on every step.