A security triage agent that learns from its own mistakes. When it gets a severity call wrong, Phoenix MCP turns the miss into a regression test and a better prompt — automatically.
5 real CVE alerts · One self-improve loop · Zero manual rules
One button. The agent runs 5 CVE alerts, discovers its mistakes, improves its own prompt through Phoenix MCP, then reruns. No human rules needed.
Note: The full loop runs 3 subprocess rounds on Cloud Run with Phoenix MCP sidecars. Expect 5–10 minutes total.
Send any security alert to the improved agent. It consults the project module graph, checks exposure paths, and returns a triage decision with evidence.