Both review your branch.
Separately and at the same time. Security holes and bugs by default, or whatever you point them at, like business logic.
Two reviews of your code.
One consensus.
Claude Code and Codex each review your branch for security holes and bugs, then work together to fix what they find.
/dual-review in Claude Code$dual-review in Codex
dual-review · 9 files, +871/−0 vs origin/main
Claude Code Opus 5 host · reviews, refutes, fixes
Codex gpt-5.6-luna counterpart · reviews, refutes
rounds ≤3 · fixes land one commit each on your branch
Round 0 · two blind reviews running
Round 0 · Claude 12 · Codex 10 · each checking the other
Round 0 · 11 agreed · 6 disputed · 3 dropped
How it works
Each checks the other's findings.
A finding stays only if the other model can't show it's wrong.
You get one list.
The findings they agree on. Anything they disagree on is marked for you to decide.
They fix what you approve.
The other model reviews each change until nothing new turns up.
Install
Clone it into your skills folder.
Needs Claude Code and Codex, signed in. Install the skill in one or both.
Claude Code
/dual-review
git clone https://github.com/drydockai/dual-review ~/.claude/skills/dual-reviewCodex
$dual-review
git clone https://github.com/drydockai/dual-review ~/.agents/skills/dual-reviewAlso try /dual-review quick · scan only · whole repo · 5 rounds · resume · use <model>