wtf is dual-ai-skills?

This project provides two Claude Code skills that put Claude and OpenAI's Codex CLI to work checking each other instead of relying on a single AI model's judgment. The idea is that any AI coding assistant tends to review its own work with the same blind spots it had while writing it, so having a second, differently trained model act as an independent critic catches more real problems. The first skill, dual plan, has Claude draft an implementation plan while Codex reviews it against the actual codebase in read only mode. Claude then responds to each of Codex's points, and the two go back and forth for up to three rounds. If they still disagree after that, the unresolved points are shown to you directly rather than silently dropped, and you approve the final plan before any code gets written. The second skill, dual review, runs Claude's own code review skill and a Codex review in parallel on the same set of changes before a merge. Findings both models agree on are treated as strong signals and shown first. Findings only Codex raises get checked against the actual code before being reported, with a chance for Codex to defend a rejected finding once. Every result is labeled as agreed by both, Claude only, Codex only, or disputed. To use this you need Claude Code and the Codex CLI both installed and logged in, since the whole approach depends on using two separate AI vendors. It can be installed as a Claude Code plugin through the plugin marketplace, or copied by hand into your skills folder, and either way you can trigger it just by asking to dual plan or dual review your work instead of typing a slash command. The project is released under the MIT license.