hypothesis-teams
A Claude Code plugin that tests difficult questions with subagent teams arguing for and against hypotheses.
What does it do?
Enter a question in Claude Code to establish the verification criteria and shared facts before forming supporting and challenging teams for each hypothesis. Team members research from different perspectives and debate within their teams, passing key findings between rounds through consolidated notes. After shifting the budget toward promising hypotheses, a separate agent checks the claims again. A final verdict table and a solo research baseline for the same question make it possible to compare the team’s results with its token cost. Progress records and results are saved as files in the current project’s research folder.
Key features
- Set verification criteria and shared facts before research
- Form subagent teams to support and challenge each hypothesis
- Debate within teams and share findings between rounds
- Have an independent agent recheck claims and produce a verdict table
- Compare team costs and results with a solo research baseline
- Save records for each round and final results in the project’s research folder