adversarial-debate

Run the FREE-MAD debate protocol, which assigns Bull, Bear, and Market agents to argue the thesis across five rounds. The Bear agent operates in anti-conformity mode, challenging every Bull argument with counter-evidence and quantified downside scenarios.

FREE-MAD is a free-form multi-agent debate method where heterogeneous models argue opposing positions. Research cited in the skill shows 91% accuracy with heterogeneous models versus 82% for single-model analysis, using anti-conformity prompting to prevent premature agreement.

Yes, the debate works with or without MCP connections. You can upload documents or provide data manually, and all three agents share that fact base, disagreeing on interpretation rather than data.

Yes, a single model can run all three agents, though accuracy drops from 91% to 82%. The anti-conformity prompting still forces the Bear agent to challenge every Bull argument rather than converge.

Debate quality depends on the underlying fact base, so data gaps are explicitly flagged as unresolved disagreements. The output identifies killer risks that could not be rebutted, but final investment decisions still require human judgment and additional due diligence.