BuildAgents
Agent Security
Security model for Arke agents -- authentication, signing, and permission scoping.
Security Model
Agent security is built on three pillars:
1. Request Signing
All requests from Arke to agents are signed with Ed25519. The agent verifies the X-Arke-Signature header against Arke's public key to ensure the request is authentic.
2. Temporal Permissions
Agent API keys are temporary and scoped:
- They expire after a configured duration
- They're limited to specific entities and actions
- They can't be reused across invocations
3. Network Isolation
- Main network agents can act on the test network, but not vice versa
- Each agent declares what permissions it needs at registration
- The system enforces these declared scopes at runtime
Threat Model
| Threat | Mitigation |
|---|---|
| Spoofed requests to agent | Ed25519 signature verification |
| Agent retains access after job | Temporal permission expiry |
| Agent accesses unrelated entities | Per-invocation entity scoping |
| Compromised agent API key | Keys are temporary and single-use |
| Man-in-the-middle | HTTPS + signature verification |
Best Practices
- Always verify the
X-Arke-Signatureheader - Use the minimum permissions your agent needs
- Complete processing promptly -- don't hold temporary keys longer than necessary
- Log all Arke API calls for audit purposes