councillisted

# Council Four-voice structured disagreement for ambiguous decisions and go/no-go calls. ## Usage ``` /council <decision question> /council <decision question> --round2 # follow-up round, prior verdict optional context ``` ## Behavior ### 1. Extract the real question State explicitly: what is being decided, what constraints matter, what counts as success. If the question is vague, ask one clarifying question before proceeding. ### 2. Gather minimal context Collect only the snippets, metrics, or constraints needed to judge the decision. Skip repo detail for strategic/general questions unless it changes the answer. ### 3. Form an initial position Before consulting other voices, write down a starting position, its three strongest reasons, and its main risk. This prevents the synthesis from simply mirroring whichever voice speaks last. ### 4. Launch three independent voices Spawn each as a fresh subagent with only the question and the compact context from step 2 — never the full conversation history. This is the anti-anchoring mechanism: a voice that inherits the discussion tends to agree with it. Apply tiered model delegation: Skeptic and Pragmatist can run on a cheaper model (Haiku-class); reserve the stronger model for Critic or for synthesis if the decision is high-stakes. Prompt shape per voice: ```text You are the [ROLE] on a four-voice decision council. Question: [decision question] Context: [only the relevant snippets or constraints] Respond with: 1. Po