cc-pair pairs Claude Code and Codex as coder and reviewer

On 2026-10-30 23:16, a V2EX user shared a self-built tool called cc-pair. According to the post, the author found that when the same agent reviews code it just wrote, it almost always lets it pass and finds little to flag, so the tool pairs two agents the author already uses, Claude Code and Codex, with one acting as author and the other as reviewer. In each round, the author finishes a change set and hands it to the other agent to review. The reviewer returns a verdict, pass or needs changes; the author then checks each point, fixing what should be fixed or explaining why it disagrees, and sends the work back for another round until it passes. If the round limit, 3 by default, is exceeded, the process stops and a human steps in to decide. The roles are symmetric, so Claude Code and Codex can take turns writing or reviewing. Two modes are available. In the automatic mode, run pair start --as claude or --as codex in one terminal; it launches the author and automatically pulls in the other agent as reviewer each round without a second terminal, on macOS / Linux. In the manual mode, two terminals each run one agent and the user asks them to start the review, which is the mode used on Windows. No subsequent developments or official statements about the tool were found, so its current status cannot be confirmed.