Fan out 4–5 role-specialised audit agents (regression / security / performance / coverage / style) over a code change, then merge findings with a noise-filtered severity model. Use as a wider-net alternative to a single review pass, or as a complement to prism-review when a change needs breadth.
