voice-editor

Voice doesn't live in first drafts—it lives in editing choices. This skill guides the editing process that turns AI's generic output into something distinctively yours.

<quick_start> Provide:

1. Content to edit (paste or file path)
2. VOICE.md location (default: ~/.claude/voice/VOICE.md; create one with /voice-analyzer)

The skill will:

1. Run slop detection on the content
2. Identify voice mismatches against your profile
3. Perform multi-pass editing (Structure → Voice → Polish)
4. Output edited content with change annotations
5. Suggest VOICE.md updates based on patterns found </quick_start>

<the_editing_reality> The 30-40% Rule:

Even with a perfect style guide, expect to edit 30-40% of AI output to make it distinctively yours. This isn't a problem—it's the feature.

30-40% editing effort is dramatically less than 100% writing-from-scratch effort, while producing work that sounds more like you than unguided AI ever could.

If you're editing more than 50%: Your style guide needs refinement, or your prompts need work.

If you're editing less than 20%: You might be accepting too much generic content. </the_editing_reality>

<editing_workflow> Focus: Does the organization serve the content?

Opening Mechanics (from writing-quality):

• Does opening create a SPECIFIC question? Not vague mystery - readers need enough context to form a particular question they want answered
• Does first paragraph contain change or threat of change? Brains notice disruption of status quo
• Is first paragraph under 50 words / 3 sentences?
• First line hits 3+ of: question, character, setting, tone, movement?

Questions to ask:

• Does the opening hook or just clear throat?
• Is there a clear progression of ideas?
• Does each section earn its place?
• Is anything redundant or missing?
• Does the conclusion land or just stop?

Scene/Section Structure:

• Does each section follow Goal → Conflict → Outcome?
• Are transitions connected by "because" not "and then"?
• Are characters/subjects in motion during slow passages?

Actions:

• Cut throat-clearing openings ("In this article, we'll explore...")
• Reorder sections for better flow
• Merge redundant sections
• Add missing transitions
• Strengthen weak conclusions
• Add change to opening if missing - start with what changed, not background
• Make question specific - "Why is the other side of the bed cold?" not "Something is happening"

Check against VOICE.md:

• Sentence length patterns (matching variance?)
• Vocabulary fingerprint (right formality level?)
• Tone markers (humor, directness, reader relationship?)
• Structural habits (lists vs prose, formatting?)
• Opinion expression (appropriate confidence level?)

Actions:

• Replace formal language with conversational equivalents
• Vary sentence lengths to match profile
• Add characteristic phrases from the voice profile
• Adjust opinion strength to match profile
• Fix paragraph rhythm

Scan for:

• Tier 1 phrases (remove immediately)
• Tier 2 patterns (check context)
• Tier 3 clusters (break up if grouped)
• Structural tells (uniformity, over-balance)
• Staccato fragment spam (3+ consecutive short declarative sentences)
• Comparator sentences ("This isn't X. It's Y." patterns)

Actions:

• Delete or replace all Tier 1 phrases
• Evaluate Tier 2 in context, remove if generic
• Redistribute Tier 3 to avoid clustering
• Vary sentence lengths if too uniform
• Rebalance sections if all equal weight
• Combine staccato fragments into flowing prose with commas, em-dashes, or conjunctions
• Rewrite comparators directly — just say what something IS, don't waste words on what it isn't