What does it mean that the agent audit?

Inventory your existing agents both designed and default to understand what is running.

Example: You decide to track every repeated behavior for 48 hours — what you do when you wake up, how you respond to Slack messages, what you eat when stressed, how you start meetings. You discover 30+ behavioral agents running on autopilot. Some you designed (the morning run, the weekly review). Most you didn't (checking your phone within 90 seconds of waking, defaulting to agreement in group settings, reaching for sugar at 3pm). You now have a map of what's actually running — not what you think is running.

Try this: For the next 48 hours, set a recurring hourly timer. Each time it fires, write down exactly what you were doing and whether that action was deliberate (you consciously chose it) or automatic (it happened without a decision). After 48 hours, sort your entries into two columns: Designed Agents (behaviors you intentionally installed) and Default Agents (behaviors running without your explicit authorization). Count each column. The ratio tells you how much of your operating system you actually wrote.