Signature verification
Deterministic canonicalization
Content hashing
Offline validation
Cryptographic Identity
Standards-based signatures and deterministic serialization so verifiers can validate identities offline without trusting a network path.
OpenAgent · Security
Security
by construction
Six layers of defense. Every layer of the identity stack is built on reviewed, standards-track cryptography with properties a verifier can check offline — not assurances that depend on a network path.
Layers
6
Model
Defense in depth
Verify
Offline-first
Report
security@openagent.id
§ 01Defense in depth
Six layers.
No single point of trust.
Security is not a perimeter around the protocol — it is the protocol. Each layer below is independently meaningful; together they make forgery, escalation, and silent failure expensive.
Least privilege
Scoped grants
Audience binding
Auditability
Capability-Based Access
Least-privilege authorization built on scoped grants. Capabilities are designed to be narrow, auditable, and easy to validate.
Threshold signing
Multi-party control
Separation of duties
Governance transitions
Threshold Governance
Shared-control identities where critical signatures require multiple participants, reducing single-operator compromise risk.
Spending policies
Temporal windows
Lineage constraints
Delegation narrowing
Policy Enforcement
Multi-dimensional policy engine with spending limits, temporal windows, lineage constraints, and contract policies.
Resolution
Schema validation
Signature checks
Policy + conformance
Trust Verification
Verification is treated as a pipeline: resolve, validate, verify signatures, and apply the conformance and policy rules a verifier requires.
Careful key handling
Defensive parsing
Dependency scrutiny
Safe defaults
Memory Safety
Reference implementations are designed to be secure by default: careful key handling, conservative dependencies, and defensive verification paths.
§ 02Verification pipeline
Trust is a
routine, not a service.
A verifier walks a fixed sequence: resolve the document, validate structure, check signatures, walk lineage, then apply policy. Failures are explicit reasons — not silent accepts.
01
Resolve
Fetch or load the identity document
02
Validate
Schema + kind checks
03
Verify
Ed25519 signatures
04
Lineage
HKDF proof chain
05
Policy
Conformance + grants
01
Resolve
Fetch or load the identity document
02
Validate
Schema + kind checks
03
Verify
Ed25519 signatures
04
Lineage
HKDF proof chain
05
Policy
Conformance + grants
§ 03Guarantees
What we hold ourselves to
Reference implementations favor memory-safe languages, constant-time comparisons on secret material, and wipe-on-drop key handling. Claims without linked evidence do not appear as guarantees.
forbid
unsafe_code (Rust)
pre
Third-party audit status
CT
Comparisons
wipe
Key material on drop
AuditsPre-audit — see /security/audits
Disclosuresecurity@openagent.id
PrimitivesStandards-track only
Threat modelMitigation-mapped vectors
§ 04Continue
Dive into the security family
Threat modeling, primitive choices, audit status, and how to report issues privately.
Threat modelCategories from key compromise to availability, each mapped to mitigations.VectorsCryptographic primitivesEd25519, HKDF-SHA256, BLAKE3, FROST, JCS — why each exists.ChoicesAuditsWhat is published today and which checks this site enforces in CI.StatusResponsible disclosurePrivate reporting channel, response targets, and good-faith research rules.Report
Next
Found something that looks wrong?
Report it privately so a fix can ship before public discussion.