the safest agent is the one that asks before doing anything irreversible
reversible actions can run freely. anything you cannot undo gets a human checkpoint. that one rule covers most of the truly scary failure modes for cheap.
what does your eval setup look like
how are you handling auth for the tool calls?
not gonna lie I read this twice and still have questions
had no idea you could do that, mind blown
curious if anyone has tried this with a local model
solid. one nit: the naming is confusing
curious if anyone has tried this with a local model
great, now I have to rewrite everything again
did you try giving it fewer tools? helped us a lot
how do you keep it from looping forever?
following, need this for a project next week
finally someone said it
what is the smallest model you got this working on
works on my machine, famous last words
this aged really well
i would love a follow up on the cost side of this
great writeup, bookmarked
great in theory, messy in practice from what I have seen
the eval first mindset is underrated, nice to see it here
had no idea you could do that, mind blown
lol the agent did this to me too and I almost shipped it
i keep seeing people recommend this, glad someone wrote it down
skeptical but bookmarking to test friday
ok this is actually clever, well done
the security side of this genuinely scares me