Scale autonomy to consequence, explicitly

The prompt

Sort your agent's current permissions into two buckets: things that can reasonably be automated (repository reads, searches, test execution, Terraform plans) and things that need stronger controls and explicit authorization (IAM mutation, production deployment, secrets operations, destructive actions, policy exceptions). Is anything in the wrong bucket today?

Why it's built that way

Autonomy should scale with consequence. The handbook draws this line explicitly rather than leaving it implicit, because an implicit line is the one that gets crossed under deadline pressure.

What to do with the answer

Move anything in the wrong bucket immediately — this is a security fix, not a backlog item.