research-graph: a CLI that checks whether a multi-agent run actually held together (MIT, built with Claude Code)

Most of the controls in my own multi-agent runs were just instructions sitting in a prompt. I would tell one model to produce an analysis and another to review it. Nothing in that arrangement told me whether the artifact changed after the review, whether the