LLM Council

Run a structured multi-agent council for decisions where being wrong is expensive. Use five distinct thinking styles, force anonymous peer review, then synthesize a clear verdict and save two artifacts:

• council-report-[timestamp].html
• council-transcript-[timestamp].md

Gate the Request

Run the council only when the user presents genuine uncertainty, multiple options, meaningful tradeoffs, or real downside risk.

Use ordinary reasoning instead of the council for:

• factual questions with one correct answer
• low-stakes preference calls
• simple creation tasks like "write me a tweet"
• pure summarization or extraction tasks

If the request is vague enough that the council would debate the wrong thing, ask exactly one clarifying question, then proceed.