Theo makes his coding agent win a debate before it can merge
His "council" skill forces the agent to argue its changes with Opus 5.5, Muse and GPT-6.1 first.
Developer Theo (@theo) shared a "council" skill for his coding agent. According to his post, the agent must argue its changes with Opus 5.5, Muse Wonder Full and GPT-6.1 Sol before it is allowed to merge code.
The idea is a kind of automated peer review: instead of trusting one model's output, several models from different labs challenge the work, and a change only goes through once it survives the debate.
Why it mattersAI agents are now writing and merging real code, and developers are experimenting with ways to catch mistakes before they ship. Having models check each other is a cheap way to add a second opinion, though it adds cost and time to every change. The post picked up roughly 1,500 likes.
