adversarial-squad

# Adversarial Squad Deployment (harness) The operational **HOW**. The doctrinal **WHEN/WHY** is the procedure `adversarial-squad-deployment` (`packs/built-in/procedures/adversarial-squad-deployment.procedure.yaml`), which sits under the `brownfield-onboarding` paradigm. This skill changes **no** mission type or guard; it is a technique the orchestrator opts into. ## When to use A squad is worth its tokens at a high-leverage point-cut where one reviewer's blind spot is expensive: - **after `/spec-kitty.specify`** → pre-spec investigation (scope, prior art, live repros) - **after `/spec-kitty.plan`** → post-planning brownfield check (foldable issues, split-brain, deprecations) - **after `/spec-kitty.tasks`** → post-tasks anti-laziness pass (fakeable DoDs, decomposition realism) - **before merge** → architectural-gate / cross-base sweep - **ad-hoc decision** → proponent + adversaries + synthesizer (e.g. delete-vs-migrate) Do NOT use it as a rubber stamp, and do NOT wire it as a mandatory gate. ## The recipe 1. **Frame one sharp question** and pick the point-cut. A squad answers a question; it is not a vibe check. 2. **Select 3–4 distinct profiles by lens** (bounded). Complementary, not redundant: - `architect-alphonso` — structure / seams / topology - `debugger-debbie` — live-evidence, coverage, "would this catch the regression?" - `reviewer-renata` — anti-laziness, contract-vs-implementation, fakeable assertions - `randy-reducer` — duplication / dead code ...