by-hypothesis-debate

Multi-agent adversarial debate selects a winning protein design strategy by spawning conservative, aggressive, and diverse hypothesis agents, then ranking their proposals with a reflection agent against a fixed scoring rubric.

Adversarial hypothesis debate for antibody design coordinates competing agents exploring different design spaces, then validates outputs against a strict schema to choose a campaign plan before committing GPU compute.

Invoke the orchestrator after your research bundle is ready to spawn three hypothesis agents, run the debate, and write the winning campaign plan directly to campaign_config.yaml.

You need a research bundle containing research.md, recommended_hotspots.json, design_recommendation.json, validated_findings.json, and a scoring rubric to run the debate and produce ranking artifacts.

Yes, multi-agent ranking applies to novel-target campaigns and scenarios with multiple viable design modalities, using a reflection agent to resolve ties and output a decision summary.

The skill requires pyyaml as its dependency and produces artifacts including ranking.json, decision_summary.md, and an updated campaign_config.yaml with logged audit trails for compliance.