investigate

An adversarial AI debate surfaces design flaws and tradeoffs by forcing two different models to critique each other's reasoning. This structured risk-analysis process highlights hidden assumptions before you commit to an architecture decision.

You launch the debate by using the investigate keyword with optional flags for rounds, codebase context, and focus areas. The backend then starts an investigator agent to produce a convergence report.

Yes, the investigation runs asynchronously in the background. You can continue your development work while the investigator agent processes the debate and generates the final convergence report.

Running a structured adversarial debate is a highly effective way to stress-test design tradeoffs. Configuring multiple rounds with specific codebase context forces a thorough evaluation of risks prior to implementation.

No external dependencies are required. The skill operates independently to facilitate the multi-provider debate and generate a convergence report containing the identified flaws and tradeoffs.

You should avoid this adversarial debate approach for trivial or single-solution tasks where no meaningful design tradeoffs or architecture risks exist to evaluate, as it adds unnecessary processing overhead.