External Review

Overview

A same-family reviewer shares the author's blind spots. This skill sends the work to a critic from a different model family so the gaps one model misses, the other can name. The critic reads the primary artifacts itself and returns a brutally honest, venue-calibrated critique plus concrete deliverables the author can act on.

This skill does not reinvent a reviewer backend. It routes the review brief through the Codex bridge documented in superpower-writing:collaborating-with-codex. A different-family critic is the point: Claude drafts the paper, Codex reviews it.

Core principle: The critic forms its own assessment from primary files. Never pre-digest, summarize, or pre-judge the work for it. A filtered brief re-introduces the very blind spots cross-model review exists to remove.

Advisory, never autonomous. Every verdict, score, and matrix this skill returns is advisory. Surface it to the user and let the user decide. Never auto-reject a paper, never flip a claim STATUS, never mutate .writing/ state on a review verdict. The review is input to a human decision, not a gate that fires by itself.

Never fabricate. Do not invent the critic's verdict, scores, or quotes when a backend is unavailable. If no reviewer backend can be reached, say so and stop. A made-up review is worse than no review.

When to Use

Trigger this skill when: