Third-lens review

The third lens in superpowers-gstack's multi-lens review. Lenses 1–2 are Claude self-pitfall (pitfall-verification) and Codex (/codex review). This skill adds lens 3 — a different model house (different training distribution → different blind spots) reading the already-patched artifact (via OpenRouter for distant houses, or the codex CLI for the countersynthesis role) — followed by a mandatory adversarial synthesis.

Invoke with: /superpowers-gstack:third-lens-review

The governing principle (field-proven): cross-model agreement = high confidence; cross-model disagreement = where the value is. A third house finds architecture-level mistakes ("you never wired it together"), degraded-state bugs, and challenged core assumptions that two Western houses both took for granted — even after they already fixed 14 issues.

When to invoke (tiering)

Normally you do not invoke this skill by hand. pitfall-verification is a multi-model orchestrator and calls this skill automatically as Stage 3 of its chain for high-stakes changes — so the third house fires as part of the standard verification flow, with nothing extra to remember. Invoke it directly only for an ad-hoc third-house read outside that flow.

This is not for every change. The lens count scales with stakes, and each added lens must add a house or a role, never another generalist: