What problem does it solve? Drafting release notes that match a specific product voice is subjective and hard to verify. This Skill replaces intuition with measurement: it writes changelogs in the Vercel/Cursor house style and grades each draft against a baseline style profile computed from 45 real published changelog entries, iterating until the numbers match. ## Core Features & Use Cases - Style-Grounded Drafting: Encodes the qualitative rules of Vercel and Cursor changelogs — short declarative sentences, second person, present tense, minimal em-dashes, no exclamation marks, bullet-heavy detail. - Quantitative Grading Loop: The stdlib-only Python script scripts/style_profile.py measures sentence length (median/p95), punctuation density per 100 words, fragment and long-sentence percentages, pronoun rates, and bullet share, then reports PASS/FAIL per metric against references/baseline_profile.json. - Revision Guidance: Each failing metric maps to a concrete editing instruction (split compound sentences, reframe around "you", move detail into bullets), with a loop capped at ~4 rounds. - Use Case: A developer tools team shipping instant rollbacks asks for a changelog entry; the Skill drafts it, grades it against the baseline, revises failing metrics, and delivers prose statistically indistinguishable from a real Vercel announcement. ## Quick Start Ask the AI to write a changelog entry for your feature in the style of Vercel or Cursor and iterate until the style grader passes.
