disler/fusion-harness ? reverse-engineered prompt

Build me a Claude Code style extension harness that lets me use two AI models together instead of choosing one.

I want three slash commands. One that asks both models for their opinion side by side, one that lets both models work in parallel and then merges their answers into one final response with clear attribution, and one that makes a validator define the acceptance criteria first, then has the builder work until the result passes.

Make it feel like a real workflow for coding tasks, with clean separation between the roles, readable output, and a clear summary of what each model said or changed. It should support setup from the command line, run the agents in isolated sessions, and keep any temporary files out of the repo. If you need to check current docs for the agent tools or command syntax, look them up online.

I care most about the experience being smooth, easy to install, and useful for actually building and reviewing code with two models working together.