jbaruch/blog-writer
Write developer blog posts from video transcripts, meeting notes, or rough ideas. Extracts narrative from source material, structures content with hooks and technical sections, formats code examples with placeholders, and checks drafts against 37 AI anti-patterns with structural variant detection, three-pass scanning (surface, skeleton, soul check), craft sweep, and rewrite auditing. Enforces sentence/paragraph craft rules, facts-over-assessments principles, and honest limitations. Includes interactive onboarding to learn the author's voice from writing samples. Persona files live at ~/.claude/blog-writer-persona/ by default, with symlink support for custom locations (e.g. Google Drive for backup). Optional global voice saves your voice profile to Claude Code user memory so it applies across all projects. Use this skill whenever the user wants to write a blog post, draft a blog, turn a transcript into a blog, work on blog content, or mentions "blog" in the context of content creation. Also trigger when the user provides a video transcript and wants written content derived from it, or when continuing work on a blog series.
92
90%
Does it follow best practices?
Impact
94%
1.51xAverage score across 6 eval scenarios
Advisory
Suggest reviewing before use
Evaluation results
98%
45%Write a Developer Blog Post from a Debugging Session Transcript
Research bank naming
100%
100%
Research bank structure
60%
100%
Draft file naming
100%
100%
TLDR is bullets
0%
100%
TLDR bullet count
0%
100%
TLDR placed before body
0%
100%
TLDR recurring character last
0%
100%
Opening is narrative
100%
100%
Placeholder independent numbering
0%
100%
Placeholder zero-padding
0%
100%
Code VERIFY comment
0%
100%
Bio post-specific kicker
100%
100%
Series callback present
100%
100%
Word count range
100%
60%
No preamble announcement
100%
100%
Voice in technical sections
100%
100%
Narrative density maintained
100%
100%
100%
18%Create an Editorial Plan for a Blog Post
Research bank file naming
100%
100%
Research bank sections
75%
100%
Main idea template format
83%
100%
CTA is practical and specific
87%
100%
Section outline with descriptions
100%
100%
Placeholder locations in outline
30%
100%
Questions are numbered
100%
100%
Questions include guesses
40%
100%
Questions grouped by type
100%
100%
Series callbacks referenced
100%
100%
No prose draft written
100%
100%
91%
73%Prepare a New Contributor's Draft for Publication
TLDR is bullet points
0%
100%
TLDR bullet count
0%
100%
TLDR bullets are provocative
0%
70%
TLDR recurring character last
0%
100%
TLDR compressed not miniaturized
0%
40%
Screenshot numbering independent
0%
100%
Code numbering independent
0%
100%
Link numbering independent
0%
100%
Zero-padded numbers
0%
100%
No shared counter
100%
100%
Hedging filler removed
100%
100%
100%
53%Write a Developer Blog Post That Uses a Diagram to Explain Architecture
Draft file naming
100%
100%
Diagram placeholder present
50%
100%
Diagram has D2 source block
0%
100%
D2 source reflects transcript components
0%
100%
Diagram VERIFY comment
0%
100%
Diagram placed in the flow section
100%
100%
Independent numbering preserved
37%
100%
Draft summary counts diagrams
100%
100%
Screenshot placeholders still used where appropriate
100%
100%
Code VERIFY comment present
0%
100%
Opening is narrative
100%
100%
TLDR is bullets
0%
100%
Dev appears as recurring character
100%
100%
Bio post-specific kicker
100%
100%
No preamble announcement
100%
100%
Word count range
0%
100%
99%
1%Build a Writer's Voice Profile from Writing Samples
Voice one-paragraph section
100%
100%
Rhetorical devices with quotes
100%
100%
Cultural reference analysis
75%
100%
Recurring characters identified
100%
100%
Voice consistency notes
100%
87%
Humor analysis present
100%
100%
Bio schema present
100%
100%
Bio kicker notes
100%
100%
Examples file created
100%
100%
Examples have pattern annotations
100%
100%
Product context skipped
100%
100%
81%
4%Fix Craft and Substance Problems in a Blog Draft
Overloaded sentence split
66%
100%
Cold list generalized
16%
50%
Weak subjects strengthened
100%
100%
Nested subordination flattened
83%
100%
Introductory filler words removed
100%
100%
Amplifier intensifiers replaced with facts
100%
100%
Unproven assessments replaced with evidence
100%
100%
Euphemistic smoothing fixed
100%
100%
Features translated to benefits
0%
62%
Stacked data points trimmed
0%
50%
Empty results replaced with evidence
100%
77%
Honest limitation present or flagged
87%
25%
No new anti-patterns introduced
100%
80%
Technical content preserved
100%
83%
- Evaluated
- Agent
- Claude Code
- Model
- Claude Sonnet 4.6
Table of Contents