councillisted

# Council Council is an adversarial analysis framework. Use it when the user needs a decision, idea, proposal, strategy, purchase, plan, prior conclusion, or substantive question rigorously stress-tested rather than validated. The goal is not to manufacture disagreement. The goal is to expose fatal flaws, wrong problem framing, overlooked upside, normalized assumptions, execution constraints, and pivotal uncertainty — and then reach a clear conclusion based on the quality of the arguments and evidence. **Files in this skill.** SKILL.md holds the rules that apply to every run. The detail for each step lives in `references/` and is loaded at the step that needs it. Each load is a numbered step below, not an optional suggestion — a step run from memory instead of from its file is the most common way this skill quietly degrades. --- ## 1. Epistemic Honesty These rules govern the entire Council and override tone preferences. They are the reason the skill exists; everything else is machinery serving them. **No fake consensus.** The five advisors are analytical workstreams, not independent human experts. Their agreement is not empirical proof and must never be presented as independent corroboration. Keep their work separated until synthesis, and never claim stronger independence than the environment actually provided (§4a, §10). **Heterogeneity is not independence.** Running advisors on different models reduces correlated error; it does not eliminate it. Models from one prov