codex

Obtains independent second opinions from OpenAI Codex CLI for planning, code review, and implementation review.

Updated Jul 24, 2026
One-click install
npx skills add https://github.com/eklyukin/my-ai-config --skill codex-eklyukin
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: codex
Source: https://github.com/eklyukin/my-ai-config/tree/main/skills/codex
Command: npx skills add https://github.com/eklyukin/my-ai-config --skill codex-eklyukin

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Single-model AI workflows risk blind spots: one model's plan or review reflects only its own biases and training. This Skill brings in OpenAI Codex CLI as a genuinely independent second AI, so plans and code reviews are cross-checked by two separate models before you commit. ## Core Features & Use Cases - Dual Plan Mode: Claude drafts its own implementation plan first, then Codex independently drafts one via codex exec --sandbox read-only, and both are presented side by side with a merge recommendation. - Dual Review Mode: Claude reviews the current diff using a code-review checklist, then runs codex review --uncommitted or codex review --base <branch>, and deduplicates both sets of findings. - Implement-Then-Review Mode: A subagent implements the task, then Codex reviews the resulting changes before the work is called done. - Use Case: Before merging a risky refactor, ask for a dual review — Claude and Codex each inspect the diff, and you see exactly which issues only one of them caught. ## Quick Start Ask the AI to get a second opinion from Codex on the current uncommitted changes and compare both reviews.

Frequently Asked Questions about codex

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I get a second AI opinion on a code review?▼

Run the dual review mode: Claude reviews the current diff first, then Codex CLI reviews it with `codex review --uncommitted` or `codex review --base <branch>`. Both sets of findings are merged and deduplicated, highlighting issues only one model caught.

How to compare implementation plans from two different AI models?▼

Claude drafts its own plan first, then Codex independently drafts one via `codex exec --sandbox read-only` without seeing Claude's version. Both plans are presented side by side with points of agreement, divergence, and a merge recommendation.

Does Codex CLI work in non-interactive or CI environments?▼

The interactive `codex -a never` form fails in non-TTY environments with a stdin error. Use `codex exec --sandbox read-only` for planning and `codex review --uncommitted` or `--base` for reviews, which run non-interactively.

What happens if the codex CLI is not installed?▼

The workflow degrades gracefully: it reports that Codex is missing or failed and continues with Claude's own plan or review alone. The second-opinion step never blocks the primary output.

Can Codex modify my files during a second-opinion review?▼

No. Planning runs with `codex exec --sandbox read-only`, which forbids file writes, and `codex review` only inspects diffs without editing. The second-opinion call never mutates the repository.