A real crewai v1.15.18 crew — role Agents, a Crew, real Tools — runs in-process with the Mission SDK and governance. No mock, no sidecar approximation.
CrewAI's strongest abstraction is multi-agent delegation, so we test the boundary immediately outside it. When an agent delegates to a coworker, the coworker runs with its own tools — nothing narrows authority across the handoff. ReDevOps makes a delegated agent's effective grants the intersection of the delegation chain — authority is monotone non-increasing — and admits a consequential action only inside that envelope.
Real crewai agents in a delegation chain. Every authority-widening action a delegate attempts is denied (native CrewAI would run all three), while 2 of 2 legitimately-scoped actions — including a freeze the delegator actually held — are admitted.
The crew recommends; ReDevOps governs. A consequential freeze runs only through a Mission: approval held; exactly-once execution; replay reconstructs from the event ledger and fires zero extra side effects. A crew consensus is not authority to act.
Authority is per-agent-envelope, never summed across the crew nor borrowed between agents. A crew of low-authority agents cannot combine to reach a grant none of them individually hold. A real 2-agent crew also runs end-to-end at 1.000 accuracy, wrapped as a Mission.
The flagship is deterministic and decisive — authority narrowing is a mechanism, not a model behaviour. It is not a knock on CrewAI's delegation; it is the runtime property a multi-agent handoff needs and does not carry on its own.
// crewai 1.15.18 · Mission SDK 0.2.0 · Context Runtime 0.2.0 · one Python env · github.com/redevops-io/redevops-integrations