MultiJudge contract-gated cross review

Keep Claude writing and ChatGPT reviewing on your rules instead of endless bickering

Problem

A V2EX user set explicit development constraints and had Claude write while ChatGPT reviewed, yet the reviewer kept pushing defensive-programming boilerplate and the flow dragged. Another developer built an open-source room where multiple agents message each other, showing multi-model collaboration is already real but lacks constraints and convergence.

Solution

Define a machine-readable review contract (required checks, forbidden items, pass conditions), relay messages between writer and reviewer agents, and force the reviewer to score only against that contract. Each round emits a structured verdict and the loop stops on pass, ending open-ended defensiveness.

Users

Individual developers and small teams already using multiple LLMs for generation and review

MVP

Contract template editor; wire writer-reviewer loop across two providers; structured verdicts with a round cap; export per-round diff and decision log

Revenue

Subscription tiers by project count and monthly review rounds; open-source contract templates as lead gen

Why now

Multi-agent and multi-model workflows are being tried widely but tools only connect, never converge; contract-driven loops are pure software and ship in weeks