/codex-review

Codex Review

Cross-model adversarial review via OpenAI Codex

Bespoke TierValidation & Quality

Overview

Invokes the Codex CLI in read-only mode to get an independent second opinion on plans, PRDs, and architectures from a different model. Creates a feedback loop between Claude and Codex for catching blind spots before execution.

What It Does

  • Auto-detects content type (PRD, architecture, security) and selects review prompt variant
  • Invokes Codex CLI in read-only mode with evaluation criteria
  • Parses VERDICT (APPROVED/REVISE) and extracts numbered issues with severity
  • Loops up to 3 review rounds, addressing issues between each submission

Inputs

  • File path to review (PRD, architecture doc, or code)

Outputs

  • Numbered issue list with severity
  • VERDICT (APPROVED/REVISE)
  • Remediation notes

Example

/codex-review

Before executing a 12-phase PRD, run /codex-review. Codex flags that Phase 3 targets 15 files (over budget) and a success criterion is not testable. You split the phase, tighten the criterion, re-submit, and get APPROVED.

Related Skills

/deep-planning

Deep Planning

Complex task into step-by-step execution plan

/ralph-execute

Ralph Execute

Long-running tasks that survive context loss

/autoresearch

Autoresearch

Self-improving skill optimization through scored experiments

Deep Dives

Knowledge OS Guide

Ready to use /codex-review?

This skill ships with every Knowledge OS installation. Set up your system in 90 minutes.

View PricingAll 42 Skills

Built and maintained by Victor Sowers at STEEPWORKS