content-scorer

name	content-scorer
description	Expert-panel scoring discipline for marketing content: assembles a panel of 7-10 domain experts, scores each dimension, and iterates toward a 90/100 aggregate in up to 3 rounds. Use this skill when the user says "expert panel this", "score my content", "rate these variants", "run this through the panel", "quality check my copy", or wants rigorous multi-angle criticism before publishing. Not for grammar or spelling checks, SEO keyword audits, or brand voice enforcement.
license	MIT
metadata	{"skillsmith-see-also":"content-angles","skillsmith-invocation":"user","skillsmith-maturity":"experimental"}
user-invocable	true

content-scorer

Publish-ready marketing content has passed a skeptical audience — not just the author's judgment. This skill simulates that audience as a panel of domain experts who score independently and force revision until the aggregate reaches 90/100 or 3 rounds are exhausted.

Assemble the panel

Build 7–10 experts tailored to the content type and audience. A blog post for a technical SaaS audience needs a skeptical practitioner, a conversion copywriter, and a first-time reader who has never heard of the product. A cold email needs a busy VP who deletes 90% of their inbox before the second sentence. Always include:

• AI Humanizer — scores whether the prose reads as human or generated. Weight: 1.5×. Non-negotiable on every content type.
• Target Reader — scores whether someone who matches the intended audience would keep reading and take the desired action.

Name each panelist, state their role and the one axis they score hardest on. Scores are 1–10 per panelist; the aggregate is the weighted mean rounded to /100.

First-pass scoring

Run every panelist independently. Report each score, the sharpest criticism per panelist in one sentence, and the top 3 weaknesses across the panel. Do not average the criticisms — contradictions between panelists are signal, not noise; surface them.

Iterate toward 90

If the aggregate is below 90, address the top 3 weaknesses, revise, and re-run the panel. Limit: 3 rounds total.

Rules for revision:

• Fix what the panel agreed on; flag what they disagreed on to the user rather than picking a side.
• Do not add length to satisfy a score. If a panelist wanted more depth, that depth should replace something weaker, not pad the end.
• The AI Humanizer's 1.5× weight means a humanizer score below 7 blocks a passing aggregate. Fix prose before other dimensions.

Report honestly

After the final round, return:

• The revised content
• The round-by-round score summary (aggregate + top critiques per round)
• A honest statement if the aggregate is still below 90: what is holding it back and what a future revision should address

Do not inflate scores to end early and do not hide a below-90 result in qualifications.