Multi-Agent Debate
A moderated discussion between persona agents where each argues from their buyer perspective, producing a consensus publish/revise/reject decision.
Multi-agent debate is the highest-fidelity quality gate in Knowledge OS. After individual persona agents score content independently, a moderator agent facilitates structured debate rounds. Agents challenge each other's assessments, surface blind spots, and converge on a recommendation. The debate format catches issues that single-reviewer skills miss: the technical evaluator might approve copy that the skeptical buyer rejects as too polished. Debate rounds run via SendMessage-aggregate pattern for 4+ participants to avoid the silent-stall problem where parallel file appends drop writes. Output is a consensus matrix with per-agent scores and a final recommendation.
Where it shows up:
Related Terms
Agent Layer
Persona Agent
An AI agent assigned a specific buyer or expert persona that evaluates content from that perspective...
Methodology
Quality Gate
An automated checkpoint in a skill chain that evaluates output against defined criteria before allow...
Methodology
Anti-Slop
A 4-category quality framework that detects and eliminates AI-generated writing patterns: banned phr...
Methodology
Skill Chain
A sequence of skills where each step builds on the previous output, creating a pipeline greater than...