OASF — OPEN AGENT SECURITY FRAMEWORK

An agent must authenticate with a verifiable identity before executing a governed operation.

Credentials are stored in encrypted vaults and never exposed to agent runtime memory.

Sessions are cryptographically bound to the originating device using ECDSA P-256 keypairs.

Agent permissions are scoped to the minimum required resources with explicit allow-lists.

Agent containers run with read-only filesystems, no-new-privileges, and seccomp profiles.

Outbound network access is restricted to explicitly allowed domains and ports.

Sensitive host paths are blocked and filesystem writes are confined to designated volumes.

Each agent runs in an isolated process namespace with resource limits enforced.

All secrets are encrypted at rest and in transit using AES-256-GCM with per-tenant keys.

Supported governed agent actions, API calls, and data-access events are logged with tamper-evident integrity.

Agents only receive the data they need; PII is tokenized before reaching agent context.

Agent outputs are scanned for secrets, PII, and prompt injection before delivery.

Skills submitted through the supported marketplace workflow can undergo code review, dependency audit, and sandbox testing.

Organization-wide policies are enforced at the gateway layer before agent execution.

Scoped evidence reports map selected controls to SOC 2, ISO 27001, and GDPR review workflows; they do not certify OpenSyber or make an environment automatically compliant.