dangerous commands blocked in about 40ms. everything else reviewed in 2 to 4 seconds.
permission request
model
approve
block
what it does
a model reads every permission request in context and approves or blocks it, so the agent never waits on me.
the code on githubthe code on github