An XDA author used a Codex plugin to review, without modifying, a DVD-management app that Claude Code had declared complete. Codex found crashes with malformed stored data, corrupted characters, duplicate screen-reader output, keyboard-focus loss after changing filters and missing search-result announcements. The Codex CLI is invoked from the Claude window, but usage is billed to the Codex account. This is one small case, not proof that either model is always better. It illustrates separating successful execution from actual quality verification.