Skills Layer

Codex Review

Cross-model adversarial review that invokes Codex CLI to get a second opinion from a different model family before committing to a plan.

Codex Review sends your plan or PRD to OpenAI's Codex CLI for an independent evaluation by a different model family. The adversarial perspective catches blind spots that same-model self-review misses: architectural assumptions Claude takes for granted, edge cases a different reasoning style surfaces, and scope creep that looks reasonable from inside the planning context. The critical timing rule: Codex Review runs before Ralph execution starts, not during it. At plan-authoring time the cost of changing direction is near zero. Mid-build, conflicting cross-model feedback compounds into context thrash and contradictory instructions. In-build quality review stays targeted to /edit-content, /skeptical-buyer, and /review-content instead.

Where it shows up:

PRD reviewArchitecture decisionsHigh-stakes plan validation

Build your Knowledge OS

90 minutes from zero to your first skill chain.