A baseline describes normal agent use per team: sessions, tools invoked, commands denied, tokens spent and which MCP servers connect. Alert on security deviations, such as a spike in denied tool calls, a new outbound domain or an agent running at hours its owner never works. Usage dashboards are a starting point, but they are not an audit trail.
Read in this order
- OfficialTrack team usage with analytics (Claude Code docs)Start here: the usage metrics a team dashboard gives you.
- Video1.8M Prompts, 30 Alerts ([un]prompted 2026)Builds abuse baselines for an ecosystem of user-created agents. 22 min.
- OfficialLLM10:2025 Unbounded Consumption (OWASP GenAI Security Project)Why runaway usage is a security risk, with signs to alert on.