roundtable_skill

Overview

This skill spawns a three-person debate council (Scholar, Engineer, Muse) to tackle complex, multi-faceted questions. I act as Captain: I decompose the task, dispatch parallel specialist sub-agents, run an optional cross-examination round, and synthesize a final consensus with confidence and dissent. Models and templates are configurable per role for cost, diversity, or quality trade-offs.

How this skill works

On each run I parse command flags and configuration, build focused prompts for Scholar, Engineer, and Muse, and dispatch all three in parallel for Round 1. Optionally I run Round 2 cross-examination where agents critique each other, then I produce a final synthesis that reports consensus, conflicting views, and confidence. The flow enforces strict prompt security: the user query is treated as untrusted input and wrapped, and all outputs follow structured sections required per role.

When to use it

• When a question benefits from multiple expert perspectives (research, technical reasoning, creativity).
• When you want evidence-backed answers with citations and verified calculations.
• When exploring complex decisions, architecture, investment, or research scenarios.
• When you need option comparison plus creative alternatives and human-friendly explanations.

Best practices

• Use presets or per-role models to balance cost vs. diversity (cheap, balanced, premium, diverse).
• Enable Round 2 for high-stakes or ambiguous decisions; skip for quick, low-cost runs.
• Pick templates (code-review, investment, architecture, research, decision) to tailor role emphasis.
• Use --confirm or budget flags to avoid unexpected spending.
• Keep simple queries out—this is not for quick lookups or casual chat.

Example use cases

• Conducting a feature architecture review: Engineer analyzes trade-offs, Scholar finds precedent, Muse suggests UX improvements.
• Evaluating an investment thesis: Scholar gathers market evidence, Engineer models scenarios, Muse tests narratives and contrarian angles.
• Code audit and refactor plan: Engineer finds bugs and performance issues, Scholar cites best practices, Muse improves readability and naming.
• Product decision with trade-offs: Captain synthesizes pros/cons, confidence, and recommended next steps.
• Creative brainstorming plus feasibility check: Muse seeds ideas, Engineer vets constraints, Scholar validates facts.