Skip to main content

Security review scope

Status: no independent security review has been done. This page is the scope the sponsor asks reviewers to use. An independent review is a gate before any real-client write traffic; it is not needed for the static site or the fictional browser simulation.

In scope​

The specification (Part 1: Core, Part 2: Additions, threat model), the reference endpoints (reference/, alt-python/), the MCP adapter, the conformance suite, and this static website.

Priority areas​

AreaQuestion
Authorization boundaryCan an agent or host approve on the person's behalf, or reuse, extend or widen an approval?
Content-digest bindingCan anything be sent that differs from what the person approved?
Replay and idempotencyCan a handoff be delivered twice, or an old webhook or receipt be replayed?
Signed receipts and callbacksCan a receipt, recipient event or 0.2 webhook be forged, re-attributed or accepted unsigned?
Key rotation and binding (0.2)Does a revoked or rotated key stop being trusted at the right time?
SSRFCan callback or endpoint URLs make a host fetch internal addresses?
Cross-tenant leakageCan one client, firm or person read another's CaseBriefs, previews or handoffs?
Recipient identityCan a request reach someone other than the one named recipient?
Retention and deletionDoes deletion do what it says, and are the limits stated truthfully?
Secrets and dependency provenanceAre test credentials clearly fake and isolated? Are dependencies pinned and minimal?
Prompt injectionIs CaseBrief and firm-supplied text always treated as data, never as instructions?

Out of scope​

Denial-of-service against olip.ai or its hosting, social engineering, physical attacks, and any test that touches real people or real firms.

Reporting​

Report privately to the security contact. Do not open public issues or post findings in public. Unresolved findings are tracked privately; the sponsor publishes only a summary that the reviewer has agreed to, after a fix or a documented decision.