What problem does it solve? When a design decision or policy question has no obvious answer, a single model's opinion is unreliable. This Skill runs a structured multi-LLM discussion where independent models state positions, respond to disagreements, and attack a draft conclusion, while the current session's LLM moderates and adjudicates a final answer. ## Core Features & Use Cases - Three-round structured debate: Round 1 collects independent opinions, Round 2 confronts each participant with conflicting anonymous positions and moderator questions, Round 3 stress-tests the aggregated draft for logical holes. - Moderator adjudication: The session LLM builds a topic map, drafts an aggregated proposal, and rules on each criticism by evidence strength rather than majority vote, with one extra round allowed for fatal objections. - Configurable participants via llms.json: Enable or disable discussants by editing entries with name, command, and timeout; all participants run through OpenCode in read-only plan mode. - Use Case: Ask whether to adopt a new architecture for a repository. The Skill checks out a worktree, gathers independent opinions from three models, maps their disagreements, forces responses to conflicts, and delivers a single adjudicated conclusion with remaining uncertainties. ## Quick Start Ask the assistant to run a discussion on whether the team should adopt approach A or B for the current design problem.
