docs: add provider security model for sdk integrators

Introduces security-model guidance covering provider invariants, expected auth failure modes, and integration-safe handling.
This commit is contained in:
Andy Dixon
2026-02-26 13:38:45 +00:00
parent 1d62f0affd
commit fc80ee1659
2 changed files with 31 additions and 3 deletions
+27
View File
@@ -0,0 +1,27 @@
# BlackWall Provider Security Model (SDK View)
Last reviewed: 2026-02-26
This document summarises provider-side security guarantees that SDK integrators should design around.
## Threat Assumptions
- Provider is internet-exposed.
- Attackers can register and attempt tenant-boundary abuse.
- OAuth/WebAuthn endpoints are abuse targets and may rate limit.
## Invariants Integrators Should Assume
1. Tenant/project/user/client scope is enforced server-side and can change between request steps.
2. OAuth authorize can fail with `invalid_scope` if requested scopes are not currently allowed for the client.
3. Token/refresh exchange can fail with `invalid_grant` when subject/client state or scope policy changes.
4. UserInfo can fail for revoked/invalid/inactive subjects even before nominal token expiry.
5. Admin and portal workflows are distinct scope domains; do not mix them in client UX flows.
## Integration Guidance
- Keep requested scopes minimal and aligned to the provider client configuration.
- Treat `access_denied`, `invalid_scope`, and `invalid_grant` as expected security outcomes, not transport errors.
- Build retries only for transient failures; do not blindly retry authz/authn denials.
- Keep enrollment URLs/tokens and OAuth tokens out of URLs, logs, and analytics tags.
- Use strict HTTPS and secure session-cookie handling in the relying party app.