autoconference-skill

Multi-agent research orchestration improves metric optimization by spawning N parallel autoresearchers that explore a shared goal, followed by adversarial peer review to catch overfitting and invalid comparisons before synthesizing validated findings into a fused output.

Adversarial peer review in multi-agent research workflows is a structured phase where an Opus reviewer challenges autoresearcher claims to identify measurement noise, overfitting, and invalid comparisons before allowing findings to propagate to the final knowledge synthesis.

To run parallel research with round-based peer review, provide a conference.md configuration file specifying your research goal, mode (metric or qualitative), and allowed or forbidden changes, then execute iterative conference rounds to generate persistence artifacts and a final synthesis report.

Yes, you can use qualitative research modes for rubric-based reviewer judgments. The multi-agent conference workflow supports qualitative evaluation through reviewer rubrics, providing stronger validation than self-evaluation alone without relying on numeric metrics.

No, multi-agent knowledge synthesis does not select a single best run. The final synthesis output combines complementary findings from all parallel autoresearchers into a unified report, controlling convergence and budget limits rather than using a winner-takes-all approach.

You should avoid using parallel autoresearchers for knowledge synthesis when your research requires a single self-optimizing loop without cross-researcher knowledge transfer, or when you cannot allocate the budget for iterative round execution and adversarial peer review artifacts.