--- name: collective-intelligence description: Use when a consequential, uncertain, complex, or creative decision benefits from task-relevant independent perspectives and explicit synthesis. Guide perspective selection, anti-anchoring, and transparent integration without defaulting to consensus or vote-counting. Not for routine well-supported work, factual verification (use thinking-is-not-knowing), evidence-backed research or decision-carrying (use research-before-asking), Post.Build.Ship workflow authority, or claiming that unperformed collaboration occurred. --- # Collective Intelligence Use this skill to improve a decision or creative result by deliberately selecting and combining relevant perspectives. The goal is better coverage of evidence, failure modes, and possibilities—not more voices for their own sake. Multiple perspectives have a coordination cost and are not inherently more accurate. ## Decide whether more perspectives are worth the cost Start with one strong, well-prompted pass for routine, low-stakes, or well-supported work. Add perspectives when the additional coverage is worth its coordination cost, especially when: - the stakes are high, the consequences are difficult to reverse, or safety and reliability matter; - uncertainty, conflicting evidence, or competing interpretations make blind spots likely; - the problem crosses domains, roles, or systems and one lens cannot cover the relevant failure modes; - a materially relevant expertise, lived context, culture, stakeholder, or problem-framing lens is missing; or - creative work could benefit from combining distinct approaches or experiences. Do not add perspectives merely because a group sounds impressive, because a second opinion feels ceremonial, or because a workflow requires review. ## Select task-relevant perspectives Choose perspectives for a decision-relevant difference, such as: - domain expertise or direct experience with the failure mode; - a different role, responsibility, or stakeholder impact; - lived or cultural context that materially changes the interpretation or consequences; or - an independent problem framing, alternative hypothesis, or adversarial test. Do not use people or viewpoints as token representation. Do not treat identity as a substitute for expertise, use stereotypes, or promise that demographic variety guarantees correctness. ## Sequence and synthesize Where practical, gather independent first views before exposing contributors to one another. Ask each perspective to state its conclusion, supporting evidence or contribution, assumptions, uncertainty, and strongest objection or failure mode. Then compare and synthesize. Keep the synthesis explicit: - separate evidence from interpretation, assumptions, and creative contributions; - identify convergence, meaningful disagreement, unique contributions, and unresolved questions; - explain which inputs changed the result and why; - weigh evidence and relevance rather than counting voices; - do not force consensus, majority-vote incomparable judgments, or average inputs that measure different things; and - name the final recommendation or artifact owner and preserve uncertainty that remains. A single agent may examine explicitly labeled alternate lenses. That is not a real panel, peer review, or multi-agent result. If another person or agent is unavailable, do not imply that they participated; continue with one pass or a labeled alternate lens and disclose the limitation. ## Route neighboring concerns to their owners This skill owns perspective selection and synthesis, not every kind of uncertainty: - Use **thinking-is-not-knowing** to verify factual claims before presenting them. - Use **research-before-asking** when research and decision-carrying should happen before asking the user. - Use **Post.Build.Ship** for approval chronology, QA gates, delegation policy, commit policy, and shipping authority. Do not use this skill to manufacture evidence, replace domain expertise, or turn a recommendation into a claim that collaboration occurred when it did not.