Agent kill chain

Borrowed from intrusion analysis, a kill chain describes an attack as a series of steps. For agents, no single tool call may look alarming on its own: reading a config file is normal, and making an HTTP request is normal. The attack lives in the sequence, read the secret, then exfiltrate it, which only reveals itself when the calls are correlated.

Catching kill chains means looking beyond one call at a time. Correlating an agent's actions across a run surfaces the dangerous pattern, lets a control plane block or hold a later step based on an earlier one, and turns a scatter of low-signal events into a single, explainable incident.