What problem does it solve? A single model's edit or judgment of a draft post is one opinion with no way to measure disagreement. This Skill dispatches the same editing or judging skill to claude, codex, and gemini in parallel with byte-identical payloads, then verifies each candidate against the skill's mechanical rules so you can reconcile real model differences instead of trusting one output. ## Core Features & Use Cases - Parallel three-model panel: Dispatches one of four post skills (voice, story-edit, blog-sharpen, story-judge) plus the input document to claude, codex, and gemini simultaneously, capturing stdout, exit codes, latency, tokens, and cost per seat. - Mechanical verification: Marks each candidate CLEAN or DISQUALIFIED against the skill's rule table (em-dashes, frozen numbers, cut-only, scorecard shape, quote anchoring), with disqualification not overridable by judgement. - Anchored merge and calibration report: For rewrite skills, merges CLEAN candidates against the source document as ground truth; for story-judge, writes a calibration report with score matrix, spread, verdicts, convergence, and open disagreements. - Use Case: You have a 2,000-word draft post and want to know which model edits it best. Run the jury with blog-sharpen, get three verified candidates, and produce an anchored merge.md with a full audit trail. ## Quick Start Run the jury on my draft post at drafts/launch-post.md using the blog-sharpen skill and merge the clean candidates.
