Agent Security Protocols

A set of security protocols for running AI agents in production: credential management, output handling, access control, and audit logging. Each protocol is operational, a specific practice with an implementation pattern.

Protocols covered: storing API keys via wrapper scripts rather than environment variables, running secret scanning on a weekly cron, output sanitisation patterns for agents that write to external systems, and the principle-of-least-authority approach to AUTHORITY.md.

Also covers: logging patterns that make agent actions auditable, Docker network isolation, and compliance considerations for agents that handle personal data.

What is included: the full protocols document, a secret scanning cron script, an AUTHORITY.md template with security annotations, and a secrets management setup guide for Ubuntu.

Best for operators running agents in production where security and auditability matter.