Agent-Set Expectations and Automatic Retry

Agents can now set their own expectations on procedure stages instead of requiring them to be authored by hand ahead of time. When an expectation fails, the agent is automatically re-invoked to retry — up to two attempts — with a summary of what went wrong passed back into its context.

This only changes what happens after a failure; agent behavior on the first attempt is unchanged. Teams rolling this out should start with non-gating expectations, which surface issues without blocking the workflow, before turning on ones that hold up a case.