What problem does it solve? Authors cannot objectively grade their own writing because session context contaminates judgment. This Skill dispatches zero-context cold judges that grade a post's reading experience with quoted evidence, so publish decisions rest on how a stranger actually reads the text. ## Core Features & Use Cases - Zero-context judging: Builds a clean payload (title, summary, body) with no author identity, edit history, or repo context, then dispatches one cold subagent per vendor that scores seven dimensions (hook, one idea, pull, voice, fluency, honesty, landing) on a 0-3 anchored rubric with mandatory quoted evidence. - Cross-vendor second opinion: Runs a second judge from a different vendor (Codex host uses Claude, Claude host uses Codex) with a byte-identical prompt, then relays both scorecards verbatim with convergence analysis and written dispositions. - Audience persona lenses: Optionally simulates a specific reader type (e.g., an Israeli LinkedIn engineering lead) returning read/share verdicts and funnel-stage reactions without scores, kept strictly separate from the craft grade. - Use Case: Before publishing a blog post, run the judge to get a letter grade (A-F out of 21 points), a PUBLISH/REWRITE verdict, the top three fixes ranked by expected lift, and a second-vendor scorecard to confirm the call. ## Quick Start Ask the assistant to run stark-story-judge on your draft post file to get a cold-reader grade and verdict before publishing.
