Evaluates agent actions for security risks, enforcing least-privilege policies with allow, deny, or confirmation decisions and secret redaction.
## 1.0.0 - Initial release. - Policy-driven ALLOW / DENY / NEED_CONFIRMATION decisions. - Default strict mode, TLS required for HTTP. - Workspace-root file jail + denylist for sensitive locations. - Header + regex secret redaction. - Outbound secret detection and blocking. - Structured audit record per decision (traceId, policyVersion, fingerprint).