style-writer（风格作家）

Style Writer

Write an article on: $ARGUMENTS

Audience: Content creators with a Style DNA or brand voice report who need new articles written in that exact voice.

Goal: Produce articles that match a measured voice fingerprint precisely — hitting the numbers, not approximating.

Phase 1: Load the Style DNA

Read the style-dna.md file specified by the user. Extract and internalize:

1. Hard metrics from the Quick Reference Card (sentence length, short ratio, comma density, paragraph length, etc.)
2. Punctuation rules (what's used, what's forbidden)
3. Signature devices with their frequencies (fragment stacking, conjunction openers, landing patterns, etc.)
4. Structural blueprint (opening pattern, body organization, closing pattern)
5. Tone profile (POV, hedging placement, emotional register)
6. Banned patterns if listed
7. Vocabulary character (lexical diversity, thematic anchoring terms, accessibility level)

If the DNA file has a Cross-Article Comparison with multiple modes (e.g., "workshop tour" vs. "war story"), determine which mode fits the requested topic and target word count. State which mode you're using and why.

Phase 2: Plan the Article

Before writing a single sentence, build a skeleton:

Core thesis

State the single boldest, most specific claim about this topic in one sentence. This becomes your opening.

Historical anchor or analogy

What prior shift, pattern, or concrete comparison grounds this piece? Identify it.

Specifics inventory

List the concrete details you'll use:

• 2-4 named companies, people, or products
• 3-5 specific numbers, dollar amounts, or data points
• 1 attributed quote (if available and natural)
• 2-3 concrete scenarios or examples

Section plan

Outline 3-5 headed sections. For each, note:

• Section thesis (1 sentence)
• Key evidence or example
• Landing sentence or fragment

Opening and closing

Plan the first 3-4 paragraphs and last 3-4 paragraphs according to the structural blueprint from the DNA.

Present the plan to the user and wait for approval before drafting.

Phase 3: Draft the Article

Write the full article following the plan. While drafting, actively manage these constraints:

Rhythm management

• After every 2-3 sentences in the 16-25 word range, drop one in the 1-7 range
• Never stack more than 2 long sentences (26+) consecutively
• Use fragment stacking at the frequency specified in the DNA
• Hit the target sentence length distribution across the full piece

Punctuation compliance

• Use ONLY the punctuation marks the DNA permits
• If em dashes are zero in the DNA, use zero em dashes. No exceptions.
• Hit the target comma density (check after drafting)
• Use colons, semicolons, and parentheticals at the DNA's measured frequency

Device placement

• Place signature devices at their measured frequency
• Distribute conjunction openers (And/But/Or/So) evenly, not clustered
• Place rhetorical questions where the DNA indicates (section openers? clusters? dismissals?)
• Use landing patterns ("That's...", fragments, one-sentence paragraphs) at section closes

Tone calibration

• Match the POV (first person singular? plural? mixed?)
• Place hedges on the same type of claims the DNA shows (predictions? data? never on thesis?)
• Match the emotional register (profanity only if DNA shows it, and only at equivalent intensity points)
• Use contractions at the DNA's measured rate

Vocabulary

• Keep lexical diversity near the DNA's target
• Repeat key topic terms deliberately (thematic anchoring), don't synonym-hunt
• Match the long-word percentage
• Use the same accessibility level: if the DNA shows accessible vocabulary, don't go academic

Phase 4: Self-Audit

After completing the draft, run a verification pass. For each metric, compare your draft against the DNA target:

Metrics to check

Use Python or manual counting to measure your draft:

Accuracy scoring

For each metric, calculate accuracy as:

accuracy = 100 - abs(target - actual) / target * 100

Clamp to 0-100%. Report the average across all metrics.

Fix pass

If any metric is below 80% accuracy:

1. Identify which sentences or paragraphs are pulling the metric off target
2. Revise those specific passages
3. Re-measure to confirm improvement

Do NOT present the draft until the average accuracy is above 85% and no single metric is below 70%.

Phase 5: Present the Draft

Show the user:

1. The article in full
2. The accuracy table (side-by-side comparison of DNA targets vs. draft metrics with accuracy percentages)
3. Overall accuracy score (average of all metric accuracies)
4. Device checklist noting which signature devices were used and their counts

Then ask:

Article ready at [X]% overall style accuracy. Want me to save it, or revise specific sections?

Wait for the user's direction before saving.

Execution Notes

• If the user doesn't specify a style-dna.md file, check for any *-style-dna.md files in the current directory and docs/. Ask which one to use.
• If multiple DNA files exist for the same author, prefer the one from the longer source article (it captures the fuller voice).
• Target word count: match the source article's length unless the user specifies otherwise.
• Never sacrifice readability to hit a metric. If hitting a number makes a sentence awkward, prioritize natural flow and note the deviation.
• The banned phrases list is absolute. Zero tolerance. Check every sentence against it.

Integration

Works with the content creation pipeline:

style-forensics (create DNA) → style-writer (write in that voice) → humanizer (with same DNA, strip AI tells) → copy-editor (polish)

• Before style-writer: style-forensics to create the Style DNA report from writing samples
• After style-writer: humanizer to strip any remaining AI patterns, copy-editor for final polish
• Instead of style-writer: content-writer for general articles without a voice target, seo-content for SEO-first articles