Key Management & Trusted Agents

Guidance for operating trusted agents: signing descriptors, authenticating deployments, rotating credentials, and responding to compromise.

See also: Trust & Validation · Cloud Deployment

Trusted Agent Overview

Trusted = registered operators with authenticated agents
Benefits: reduced validation overhead, priority scheduling (policy)
Trust is revocable; untrusted mode remains always available

Agent Descriptor Signing

Obtain operator credentials from the project’s registration process
Sign canonical agent descriptor (capabilities + build provenance)
Advertise over an authenticated channel

Rotation:

Short-lived credentials; rotate regularly (e.g., 24–72h)
Automate rotation via CI/agents; audit key usage

Secure Storage & Access

Store keys in cloud KMS or local secure enclave
Limit blast radius with scoped IAM
Use separate credentials per deployment/environment

Compromise Runbook

Revoke credentials immediately
Quarantine associated agents in orchestrator
Rotate all dependent secrets
Redeploy from known-good binaries with fresh credentials

See also: Threats 5.6/5.7 in Threat Model

Attestation (Optional)

Where supported, emit attestation references alongside signed descriptors
Use for additional confidence; treat as advisory, not absolute

Checklist

Descriptor signing in CI
Automated rotation
Audit logs enabled
Quarantine playbook tested