councillisted

# council Five independent critics, one synthesizer, no majority voting. Single round. Adapted from Chris Blattman's [claudeblattman](https://github.com/chrisblattman/claudeblattman), including the rule that the synthesizer ranks by how much of the argument rests on a finding and never by how many critics raised it. ## When to use - Stuck between two framings of a paper, R&R, or grant, and you want adversarial pressure on both. - A plan, design, or skill draft is almost right and needs failure-mode pressure before it ships. - The target is heterogeneous (idea plus design plus venue choice, or a strategy memo) so a manuscript-shaped review does not fit. Not for a complete manuscript you want line-edited and refereed against a journal bar: that is `review-paper`. Not for numeric or reproducibility audits. This skill produces a checklist and never edits the target. ## Inputs | Argument | Default | Meaning | |---|---|---| | target | required | file path, or an inline paragraph describing the idea, plan, or decision | | `--chef-skill` | off | swap in the skill-design roster | | `--critics=a,b,c` | default roster | explicit roster; unknown names become free-form role prompts | | `--n=K` | 5 | number of critics, hard cap 5 | A file target may be `.tex`, `.md`, `.qmd`, `.txt`, `.pdf`, `.R`, `.py`, or a `SKILL.md`. Read it once in the main thread before dispatching, and pass the contents inline to the critics so five subagents do not each re-read it. This setup assumes the Re