CONTINUITYRAmEx Labs
CORE 0.2 Evaluation

A security review, interrupted.

The agent changes.
The responsibility remains.

Replace a reviewer. Withdraw a permission. See what happens to its powers—and to the work it leaves unfinished.

it is 2 a.m. do you know what your agent is doing?

Security. Control. Inheritance.
Durable survival of powers and responsibilities.
Record durably, locally or on chain.

This is your agent’s brain on

CONTINUITY

THE USEFUL DISTINCTION

Starting a replacement is easy.
Knowing what it inherits is harder.

A process supervisor can start Agent B. Continuity records which role B occupies, which powers it has, what happened before, and which unfinished duties survive the change.

That matters when a security review outlives its original investigator. The replacement should receive enough authority to do its assigned work. It should not silently inherit every permission the previous agent had.

Identity

Who is acting? Agent A and Agent B remain distinct.

Authority

What may they do now, under the selected rules and history?

Responsibility

What unfinished duty remains, and who is assigned to perform it?

Choose your holder of record.

Keep records locally, or choose a chain-backed deployment to make selected history publicly verifiable. Cryptographic commitments can let authorized reviewers check private records without publishing their contents. This tutorial runs locally; each deployment has its own privacy, availability and finality requirements.

What the demonstration establishes

The same deterministic Core engine used by the reference application denies the obsolete signed request, preserves the OPEN duty, and checks the successor’s permission independently. A fresh local process can replay the retained history.

What it does not establish

This local tutorial uses public test keys and a trusted host. It does not isolate hostile programs, identify an attacker, prove that an incident is true, or guarantee that an investigation is finished. Protocol termination does not kill an operating-system process.

In a real integration, every consequential operation must go through a protected executor. An old ALLOW result is not permission to act later. An AI may propose work; it does not decide cryptographic or policy validity.