What does CueCrux give security teams?

From a security team's perspective, an agent fleet is a set of fast-moving credential holders that act at machine speed and forget to mention it. The controls that work for humans, standing roles, quarterly access reviews, best-effort logs, do not fit. CueCrux supplies controls designed for the actual threat model.

Authorisation is per request and fails closed. RCX issues short-lived capability tokens scoped to specific tools, a tenant and a trust tier, self-issued locally and verifiable offline, so there are no long-lived grants quietly accumulating. When policy says no, the denial is not a swallowed exception: it is a signed RefusalReceipt with a reason code, which turns every blocked attempt into forensic material. The off, observe and enforce postures let you baseline real traffic before turning enforcement on.

Identity is unavoidable. Every connection binds to a passport and every tool call carries it, so there are no anonymous writes to investigate around. Any action in the store resolves to who acted, at what tier, under which grants.

Incident response gets a replayable trail. Receipts are Ed25519-signed and hash-chained on a BLAKE3 append-only spine, so an intruder or a malfunctioning agent cannot silently edit history; altering, inserting or removing records breaks the chain detectably. Verification runs offline with corecruxctl verify-store, which means the evidence can be validated on a clean machine during an incident, when trusting the possibly-compromised environment is exactly what you must not do.

Containment ends with proof. Revoking a compromised agent's access is a signed pull, acknowledged per agent, with an attestation issued when zero holders remain. The difference between believing access was removed and holding an attestation that it was is the difference security reviews are paid to notice.