Copy critique

Copy Critique

Standalone evaluation skill. No generation. Receives copy, returns scores and actionable feedback.

Overview

This skill evaluates copy that already exists. It never generates copy from scratch. Input is existing copy text. Output is a structured critique: 7-dimension scores, line-level issues with concrete fixes, and a summary assessment. Optionally rewrites flagged sections when requested or when AI-tell score is high.

Input

Required:

• Existing copy text to evaluate

Optional context (include when available for a more accurate voice match score):

• audience: who the copy is for
• goal: what the copy should achieve
• brand_voice: tone, vocabulary, personality guidelines

5-Step Evaluation Process

Step 1: Receive copy and context

Take the copy as submitted. Note any optional context provided. Do not modify the copy before evaluation begins.

Step 2: Anti-slop scan

Load quality-frameworks/references/anti-slop.md.

Scan the copy for:

• Every banned word and phrase listed in the Banned Words and Phrases section
• Em dash usage
• Tricolon patterns that read like jingles
• AI tell-tale phrasing (hedging language, hollow transitions, vague superlatives)
• Paragraph openings with repetitive grammatical patterns

Flag each hit with the specific text and its location in the copy.

Step 3: Score against 7-dimension rubric

Load quality-frameworks/SKILL.md. Score the copy against all 7 dimensions.

Dimension	Scale	What it measures
Clarity	1-10	Reader understands on first read, no re-reading required
Specificity	1-10	Concrete details vs. vague generalities; numbers, names, proof
Voice Match	1-10	Matches the brand's tone, vocabulary, personality (score 5 if no brand_voice context provided)
AI-Tell Score	0-10	Detectable AI patterns. Lower is better.
Persuasion	1-10	Applies persuasion principles; moves reader toward action
Action	1-10	CTA is clear, specific, friction-free, appropriately urgent
Overall	1-10	Holistic quality judgment across all dimensions

Score each dimension before moving to Step 4.

Step 4: Build line-level issue list

For every issue found in Steps 2 and 3, produce an issue entry:

• Reference the specific line or phrase (quote it exactly)
• Explain why it is a problem (banned pattern, vague claim, broken rhythm, etc.)
• Suggest a concrete fix: "Replace X with Y" not "Consider improving X"
• Assign severity: high, medium, or low

Severity classification:

Severity	Criteria
high	Banned pattern present, major AI tell, factually wrong statement
medium	Vague claim with no specifics, weak or missing CTA, inconsistent tone
low	Minor style preference, optional improvement, rhythm issue

Do not write a vague suggestion. Every fix must name the specific replacement text or rewrite.

Step 5: Rewrite flagged sections (conditional)

Rewrite flagged sections only if:

• The user explicitly requests a rewrite, OR
• AI-Tell Score is 5 or higher

If neither condition is met, return the issue list without rewriting. Do not rewrite copy the user did not ask to have rewritten.

Output Format

Return evaluation results in this YAML structure:

scores:
  clarity: X
  specificity: X
  voice_match: X
  ai_tell_score: X
  persuasion: X
  action: X
  overall: X
issues:
  - line: "the specific text quoted from the copy"
    problem: "why this is flagged"
    fix: "concrete replacement or rewrite"
    severity: high|medium|low
summary: "1-2 sentence overall assessment"

If a rewrite was requested or triggered (AI-Tell Score >= 5), append a rewritten_sections block after summary:

rewritten_sections:
  - original: "the original flagged text"
    rewrite: "the replacement"

Hard Rules

Fixes must be concrete. "Replace X with Y" is acceptable. "Consider improving X" is not. Every issue entry must contain a specific proposed replacement, not general advice.

No generation without evaluation. If no existing copy is provided, report that this skill requires existing copy and stop.

Scoring without context. If brand_voice is not provided, score Voice Match at 5 and note the missing context in the summary.