Pitfall verification

Use this skill after finishing any PRD, spec, plan, or code artifact — before declaring the work done. It is NOT a generic review. It is a targeted check that typical pitfalls for this type of artifact, in this domain, do not apply here.

Invoke with: /superpowers-gstack:pitfall-verification

This skill is a multi-model orchestrator

For anything beyond a trivial change, one model is not enough — there is always more to find, and a single training distribution has fixed blind spots. So this skill does not stop at Claude's own pitfall pass. For substantive / ship-worthy changes it runs, as ONE automatic flow:

1. Self-pitfall (Claude, lens 1) — the rounds below, max 2.
2. Codex (lens 2) — /codex review on the patched artifact.
3. Third model house (lens 3) — /superpowers-gstack:third-lens-review on the patched artifact.
4. Adversarial synthesis — combine all findings, finding-is-real-until-refuted.

Stages 2–4 run automatically. Do NOT stop after the self-pitfall rounds to ask the user whether to run Codex or the third lens — run them. The whole point is that there is nothing extra to remember: invoking pitfall-verification on substantive work is the multi-model review. The only thing that skips stages 2–4 is the trivial tier (see "Tier gate" below) — a typo or doc fix gets the free self-pitfall pass and nothing else.

Stage 0 — name the target, compute the tier floor