CLI Companion Council

`ed companion council`

Asks several lenses the same question in sequence, then runs a synthesis pass whose only job is to locate the crux: the one fact none of them has, which would settle the disagreement if it were known. The disagreement is the output. A single confident answer would tell you less.

Usage:

ed companion council "<question>" [--personas analyst,coach,skeptic] [--json]
                                  [--endpoint <url>]

Name	Type / values	Default	What it does
`--personas`	comma separated lens ids	`analyst,coach,skeptic`	Which lenses sit on the council. Two is the minimum.

Persona ids must be exact; unknown ids and lists shorter than two are refused. The command requires a configured reasoning provider.

Each lens answers through its own retrieval policy and evidence weighting, so they can genuinely disagree rather than paraphrasing one another. The plain output prints each answer, then where they agree, where they diverge, the crux, and the question worth answering before you decide.

--json shape: {question, agreement, divergence, crux, cruxQuestion, model, answers: [{persona, label, answer, abstained, grounding, citations}]}.

This is materially more expensive than one ask. Each selected lens runs its own multi-stage pipeline, one after another, and a final model call synthesizes their answers. Each lens response also writes its normal turn and retrieval telemetry. Default to ed companion ask and escalate here.