This is a pre-PR code review orchestrator that spawns parallel reviewer agents, each with a specialized persona, then merges and deduplicates their findings into a single structured report. By default it runs interactively and applies safe fixes to a clean working tree without pushing. Pass mode:agent to get JSON output for automation instead of markdown tables. It parses arguments for scope (PR links, branch names, base: refs, plan: paths) and uses a P0-P3 severity scale with action routing metadata so downstream tooling knows whether a finding needs manual design work or can be applied programmatically. Use it before opening a PR or inside larger workflows that need deterministic, machine-readable review output.
npx skills add https://github.com/everyinc/compound-engineering-plugin --skill ce-code-review