companion-context-protocol

Threat Model

Status: Draft, pre-1.0

This document describes the security and privacy assumptions behind the current CCP draft. It is not a certification claim, complete risk assessment, or substitute for an implementer’s own threat model.

Assets

CCP is designed to protect:

Trust Boundaries

The current draft assumes these boundaries:

In Scope

The draft is intended to reduce these risks:

Out Of Scope For This Draft

The current draft does not yet define:

Implementers must address these items in their own deployment architecture until the draft defines them or explicitly adopts an existing primitive.

Actor Risks

The term Actor currently covers people, organizations, systems, agents, and services. These have different trust postures:

Future drafts should add more explicit actor typing or delegation semantics if design partners need interoperable policy decisions across these roles.

Grant Risks

PermissionGrant is a schema-backed authorization record, but grant transport and proof are intentionally not standardized yet. This creates interoperability risk:

Until this is specified, compatible servers should treat grant identifiers as lookup keys into trusted authorization state, not bearer secrets, and should fail closed when grant status, issuer authority, subject pet, requester, purpose, scope, facility boundary, or service window cannot be verified.

Cross-Profile Inference

Visibility precedence rules prevent many single-response leaks, but they do not fully prevent inference across multiple authorized calls. For example, a requester with separate commerce and care-facility grants could compare omissions, partial responses, timestamps, or summaries to infer restricted context.

Current mitigations are:

Future drafts should define stronger guidance for rate limits, query correlation, per-request minimization, and profile-combination policy.

Facility Truth Risks

Facility Truth is expected to cover mostly public operational facts, but public-by-nature does not mean risk-free. Incorrect hours, stale service eligibility, invented certifications, or outdated booking links can still harm facilities and clients.

Facility Truth should therefore require provenance, freshness, source identity, omission reasons, and clear distinctions between public facts, partner-only facts, and restricted internal facility data.

Security Review Questions

Design partners should review: