Council

Develop useful alternatives against the user's actual objective. A draft is optional. If the request is specifically to critique existing work, invoke the neighboring adversarial-review workflow once; do not run both workflows automatically. Discussing councils does not invoke one.

Read references/host.md and references/panel-execution.md. Follow that complete execution contract and the shared Quality and revision contract in council-synthesis.md. The shared scripts/council.py defaults to this development workflow. With no saved preference or explicit override, use three fresh local workers: skeptic, pre-mortem, completeness checker, then a fresh synthesis worker. Chief of Staff is an optional fourth development member only on request. All four persona definitions are bundled.

Use the persona's development guidance: propose an answer or approach, alternatives, tradeoffs, evidence, and a next step. Do not turn development into four copies of a critique. Members work independently; the host is not a panel member or the synthesizer. Preserve the user's voice and constraints in any resulting draft.

For open creative, strategic or design work, ask for meaningfully different approaches and the reasoning and tradeoffs behind them. Give concrete analytical questions in the shared brief instead of relying on persona labels as expertise. Present distinct directions, their strongest reasons and what can be combined; recommend without manufacturing consensus or an averaged compromise. Candidate prose is welcome when requested; otherwise use the discussion object that helps the user choose. Preserve settled decisions, uncertainty and rejected suggestions. Maintain the short working summary of accepted repairs, rejected suggestions and held questions, and verify selected revisions in the actual final artifact when revision is authorized.

Saved provider preferences apply only when the user has asked to change their default. Exact provider/count requests override saved defaults. Do not infer a request to use paid services from the subject's importance. A later review of the revised result uses adversarial-review with a new archive and fresh reviewers.