lurx/smb-ai-skills-hub/blog-style

Blog Style - Writing Style Learning

Learn an author voice profile from existing posts and use it as a durable baseline in the project's VOICE.md file. The profile captures measurable style signals so future drafts can preserve the author's cadence, vocabulary, and tone. Other tools (such as the blog-persona skill) can read VOICE.md to seed their settings.

Learn Workflow

Use 5 to 10 representative posts from the same author, brand, or editorial voice. Accept individual markdown files, MDX files, text files, or a directory containing posts. If fewer than 5 posts are supplied, warn that the profile may be less stable and continue.

Claude performs the analysis directly — no script is involved:

1. Read every sample post in full. Strip frontmatter, code blocks, raw

HTML/JSX, and image/link URLs before measuring; keep the visible prose.

1. Segment sentences and paragraphs. For Hebrew posts, compute all metrics

on the Hebrew text itself; sentence segmentation by Hebrew punctuation (. ! ? and the sof pasuq-style period usage) is fine.

1. Measure, per post and aggregated across the corpus:

• Sentence-length distribution: percentage of short (under 10 words),

medium (10-20 words), and long (over 20 words) sentences, plus mean and median words per sentence

• Average paragraph length in sentences and in words
• Vocabulary tier: consumer, professional, or technical, judged from word

choice, jargon density, and assumed reader knowledge

• Contraction and colloquialism frequency (rare / occasional / frequent,

with an approximate rate per 100 sentences; for Hebrew, count slang and spoken-register forms instead of contractions)

• First-person usage rate and heading-as-question ratio
• Passive-voice sentence rate

1. Place the voice on the NNGroup 4 tone dimensions, each as a 0.0-1.0

value with a one-line justification quoting evidence from the samples:

• funny_serious (0.0 funny, 1.0 serious)
• formal_casual (0.0 formal, 1.0 casual)
• respectful_irreverent (0.0 respectful, 1.0 irreverent)
• enthusiastic_matter_of_fact (0.0 enthusiastic, 1.0 matter-of-fact)

1. Collect signature elements: recurring phrases (2-3 word content

phrases, stopwords removed), characteristic openers (how posts and sections tend to begin), and characteristic closers (how posts tend to end).

1. Emit the profile block below and write it to the project's VOICE.md

(append or replace the existing profile block after asking the user).

VOICE.md Profile Block

# Voice Profile
 
Learned from: [N] posts ([list of files]), [date]
Language: [English / Hebrew / mixed]
 
## Sentence & Paragraph Rhythm
-Sentence length: [x]% short (<10w), [y]% medium (10-20w), [z]% long (>20w)
-Mean sentence length: [n] words (median [m])
-Average paragraph: [n] sentences / [m] words
 
## Vocabulary & Register
-Vocabulary tier: [consumer | professional | technical]
-Contractions/colloquialisms: [rare | occasional | frequent] (~[n] per 100 sentences)
-Passive voice: [n]% of sentences
-First person: [rate]; question headings: [ratio]
 
## Tone (NNGroup dimensions, 0.0-1.0)
-funny_serious: [v] — [evidence]
-formal_casual: [v] — [evidence]
-respectful_irreverent: [v] — [evidence]
-enthusiastic_matter_of_fact: [v] — [evidence]
 
## Signature Elements
-Phrases: [list]
-Openers: [description + examples]
-Closers: [description + examples]
 
## Drafting Guidance
-[3-6 concrete rules derived from the metrics above]

Consuming the Profile

When writing or editing posts with the profile available:

• Keep average sentence length near the learned mean and match the learned

short/medium/long mix unless the user asks for a tighter or looser cadence.

• Preserve signature phrases only when they fit the topic naturally.
• Use the first-person and heading-question rates to decide how personal and

question-led the draft should feel.

• The blog-persona skill can read VOICE.md to seed a structured persona:

map the learned values to persona sentence length, passive voice, vocabulary, and tone settings.

Error Handling

• Too few posts: Continue and warn that the profile may be less stable.
• Missing paths: Skip missing paths and include a warning in the profile.
• Unsupported files: Skip unsupported file types and include a warning.
• Empty samples: Note the empty file and exclude it from the metrics.