What Your Aperture Agent Logs Actually Reveal

What Your Aperture Agent Logs Actually Reveal

What Your Aperture Agent Logs Actually Reveal

This talk shows how to turn Aperture by Tailscale agent logs into practical insights. Coding agents generate large volumes of logs, but most teams don’t know how to extract useful signals from them.

We’ll walk through real failure modes observed across coding-agent sessions — from risky tool usage and leaked secrets to agents drifting into suboptimal trajectories — and show how analyzing these patterns can improve both developer productivity and security posture.

We’ll cover approaches such as clustering common trajectory drift patterns, identifying high-risk actions, and using the insights to improve agent configuration (agents.md / claude.md) so agents make better decisions over time.