dispatching-code-review-subagentslisted

# Dispatching Code Review Subagents Core principle: split review only when independent reviewers can produce better signal than one reviewer. The main agent routes work and synthesizes results; it does not duplicate the reviewers' full review. ## Workflow Execute these steps in order. Do not skip steps. Do not stop after synthesis. ``` Step 1: Scope — Inspect diff, decide reviewer count and review lenses. Step 2: Review — Dispatch reviewer subagents in parallel. Step 3: Synthesize — Deduplicate findings, merge, spot gaps. Step 4: Verify — 4a: classify & plan; 4b: verify per 4a plan, produce report-ready fields; 4c: for subagent output only, confirm exists & sanity-check, filter rejected. Step 5: Report — Assemble verifier output into report table, no re-judgment. ``` ## When NOT to Use This Skill - The diff is trivial (typo, missing import, single-line config) — review inline directly. - No subagent dispatch capability is available. - A single localized change with no cross-domain risk — one quick self-review suffices. ## Decide Reviewer Count **One reviewer** when the diff is small, localized, or owned by one domain. **Multiple in parallel** when the change spans independent domains, distinct lenses would produce different findings, or the diff is large enough that one reviewer would sample instead of inspect. Use layered review only when one pass changes the next (e.g., correctness findings reshape how a second reviewer examines tests). If pas