WIO

got it, keeping IAM for auth and pushing authorization into the DB is a cleaner split than i was picturing. thanks for walking through it.

the "knowing what good looks like" bit is what I keep wondering about. when the agent is running OCI CLI commands for you, how do you actually review that it didn't leave something too permissive, like those 2 IAM policies? do you read the diff or just test the outcome?

the just-in-time depth part is what sticks for me. but the sonarqube story kinda shows the catch, you only dug in because CI flagged it, so what's the trigger when nothing flags it and the agent's code just quietly rots?

Interesting take on putting a plain function API under WebMCP. How does glottis handle a shop that changes its exports between versions, does the caller just break? Also curious if the bound tools get messy once an agent is holding five or six cart objects at once.