Security model
Security decisions come from executable policies, never from a model's reading of a warning. Anything ambiguous, failed or unverifiable results in no signature.
Schema 1.0 · pre-release · not deployed
Threats addressed01
TEE guarantees02
Signing keys03
Production release gates04
No production release proceeds until every gate is met. None are met today: Ravelin Wall is pre-release.
Incident response05
01Contain
Engage the global stop or pause the affected agent through the independent control path. New authorizations are denied immediately; nothing already denied can be signed.
02Revoke
Revoke compromised API credentials and, where needed, the signing identity registered with the smart account. Revocation of a signer is an on-chain action.
03Investigate
Reconstruct the timeline from the append-only audit trail by request ID, policy version and reason code. Reservations show what was authorized but not finalized.
04Recover
Release the stop only after health checks pass and an authorized operator confirms. Enclave or measurement changes are re-attested before signing resumes.
05Disclose
Notify affected organizations with the scope, the window and the audit references. Publish a post-incident note in the changelog.
06Learn
Add an adversarial test that reproduces the incident and a release gate that blocks its regression.
The repository is private during pre-release, so report suspected vulnerabilities directly to the maintainers, following SECURITY.md. A public private-disclosure channel is a release gate and will be published before the first public release. Do not test against systems you do not own.