Proposed security boundaries

Consent should have visible edges.

The concept separates user intent, policy decisions, secrets, execution, and reconciliation. The public prototype does not handle real financial data.

Architecture proposal only · Controls are not presented as implemented, audited, or ready for production.

Separation by purpose

Intent

Keep the user’s requested scope and constraints explicit.

Policy

Evaluate the proposal against defined rules and return allow, deny, or needs-review.

Secrets

Keep credentials, tokens, card data, and unrestricted account access outside model context.

Execution

Allow only an approved, bounded proposal to reach a future adapter.

Reconciliation

Record the strongest state a future source can actually confirm.

Evidence

Keep proposal versions, decisions, approvals, and state events inspectable.

Conceptual trust boundary

  1. MODELbounded intent fields and explanations
  2. POLICY LAYERallow / deny / needs review
  3. SECRET BOUNDARYcredentials excluded from model
  4. FUTURE ADAPTERapproved proposal only
  5. RECONCILIATIONaccepted / settled / credited

Authentication, authorization, storage, providers, observability, incident response, and independent review remain TBD.