writer
Generate content in your authentic voice across emails, blogs, social media, and reports
by krishagel|Open Source
Writer Skill
Generate content in your authentic voice across emails, blogs, social media, and reports.
Overview
The Writer skill learns from your writing samples and generates content that authentically sounds like you. Unlike generic AI content generation, this skill:
- Maintains 6 distinct voice profiles (work email, personal email, blog, LinkedIn, Twitter, reports)
- Interviews you like a Pulitzer-winning journalist to understand your message
- Validates output against your authentic patterns
- Saves versioned drafts to Obsidian
Key Innovation: Uses Claude's native LLM capabilities for voice analysis - no complex NLP libraries needed. Scripts just fetch samples, Geoffrey analyzes them.
Quick Start
1. Initial Setup (Before First Use)
Gather writing samples for voice profile creation:
cd ~/non-ic-code/geoffrey/skills/writer/scripts
# Extract email samples (3 accounts)
bun extract-email-samples.js --account psd --date-range "2024-06-01:2025-12-06" --max-samples 100 --output "/tmp/email-psd.json"
bun extract-email-samples.js --account kh --date-range "2024-06-01:2025-12-06" --max-samples 50 --output "/tmp/email-kh.json"
bun extract-email-samples.js --account hrg --date-range "2024-06-01:2025-12-06" --max-samples 50 --output "/tmp/email-hrg.json"
# Extract blog samples (requires Geoffrey Chrome)
cd ~/non-ic-code/geoffrey/skills/browser-control && ./scripts/launch-chrome.sh
cd ~/non-ic-code/geoffrey/skills/writer/scripts
bun extract-blog-samples.js --urls "https://psd401.ai/blog/post1,https://blog.krishagel.com/post2,..." --output "/tmp/blog-samples.json"
# Extract social samples (requires login to LinkedIn/X in Geoffrey Chrome)
bun extract-social-samples.js --platform linkedin --profile-url "https://linkedin.com/in/krishagel" --max-posts 50 --output "/tmp/social-linkedin.json"
bun extract-social-samples.js --platform twitter --profile-url "https://x.com/KrisHagel" --max-posts 50 --output "/tmp/social-twitter.json"
Then ask Geoffrey to analyze samples and create voice profiles:
Analyze my writing samples and create voice profiles.
Email: /tmp/email-psd.json, /tmp/email-kh.json, /tmp/email-hrg.json
Blog: /tmp/blog-samples.json
Social: /tmp/social-linkedin.json, /tmp/social-twitter.json
Estimated time: 30-45 minutes
2. Using the Skill
Once voice profiles are created:
write blog post about our AI implementation journey
or
write email to my team about next week's meeting
or
write LinkedIn post about Stanford AI Tinkery visit
Geoffrey will:
- Load your voice profile
- Interview you about the content
- Create an outline for approval
- Generate a draft in your voice
- Validate against your patterns
- Save to Obsidian
File Structure
skills/writer/
├── SKILL.md # Main skill definition with 6-phase workflow
├── scripts/
│ ├── extract-email-samples.js # Fetch Gmail emails
│ ├── extract-blog-samples.js # Fetch blog posts
│ ├── extract-social-samples.js # Fetch LinkedIn/Twitter posts
│ └── README.md # Script documentation
├── templates/
│ ├── interview-questions.json # Question sets per content type
│ └── voice-profile-template.json # Empty profile structure
└── README.md # This file
Data Storage:
- Voice profiles:
~/Library/Mobile Documents/com~apple~CloudDocs/Geoffrey/knowledge/writing-voice.json
Obsidian Drafts:
~/Library/Mobile Documents/iCloud~md~obsidian/Documents/Personal_Notes/Geoffrey/Writing/Blog/Email/Social/Reports/
Voice Profiles
Six distinct profiles capture different writing contexts:
| Profile | Source | Target Samples | Confidence |
|---|---|---|---|
email_work | PSD Gmail (sent, 6mo) | 50+ emails | Check writing-voice.json |
email_personal | Personal/HRG Gmail | 50+ emails | Check writing-voice.json |
blog_technical | psd401.ai + blog.krishagel.com | 10+ posts | Check writing-voice.json |
social_linkedin | LinkedIn posts | 30+ posts | Check writing-voice.json |
social_twitter | X/Twitter posts | 30+ posts | Check writing-voice.json |
report_formal | User-provided samples | 5-10 reports | Check writing-voice.json |
Confidence Levels:
- High (0.85+): 50+ samples - voice well established
- Moderate (0.70-0.84): 20-49 samples - patterns emerging
- Low (<0.70): <20 samples - insufficient, warn user
6-Phase Workflow
-
Voice Profile Selection & Validation
- Detect content type
- Load profile from writing-voice.json
- Check confidence, warn if low
-
Deep Interview (Adaptive)
- 3-5 questions for tweets/social
- 4-8 questions for emails
- 8-12 questions for blogs
- 15+ questions for reports
-
Content Planning & Outline
- Create structure based on interview + content type
- Apply voice preferences
- User approves outline
-
Draft Generation
- Load identity from identity-core.json
- Apply voice characteristics (tone, structure, vocabulary)
- Write naturally with voice context
-
Validation & Refinement
- Geoffrey self-validates against voice profile
- Score 0-100 on tone, structure, vocab, formatting
- Refine if score <85
-
Storage & Versioning
- Save to Obsidian with frontmatter
- Return text for copy/paste
- Offer to open in Obsidian
Integration Points
- knowledge-manager: Load identity context for personality alignment
- google-workspace: Fetch email samples from Gmail
- browser-control: Scrape blog/social posts
- obsidian-manager: Save versioned drafts
Example Usage
Quick Work Email
You: write email to my team about next week's leadership meeting
Geoffrey:
- Uses email_work profile (confidence: 0.92)
- Quick interview (4 questions)
- Generates ~200 word email
- Direct, bullet-point style
- Validation score: 89/100
Blog Post
You: write blog about our AI implementation journey
Geoffrey:
- Uses blog_technical profile (confidence: 0.78)
- Standard interview (11 questions)
- Generates ~850 word post
- Accessible tone, staff quotes, chronological
- Validation score: 87/100
- Saves to Geoffrey/Writing/Blog/2025-12-06-ai-implementation.md
LinkedIn Post
You: LinkedIn post about Stanford AI Tinkery visit
Geoffrey:
- Uses social_linkedin profile (confidence: 0.91)
- Quick interview (5 questions)
- Generates ~180 word post
- Thought leadership tone
- Validation score: 92/100
Maintenance
Quarterly Voice Refresh
Voice profiles should be refreshed every 3 months as writing styles evolve:
# Re-extract recent samples
bun scripts/extract-email-samples.js --account psd --date-range "2025-09-01:2025-12-06" --output "/tmp/email-refresh.json"
# Ask Geoffrey to update profile
"Update my email_work profile with /tmp/email-refresh.json"
Geoffrey will auto-warn if profiles are >12 months old.
Troubleshooting
See scripts/README.md for detailed troubleshooting.
Common issues:
- "No voice profile found": Run initial setup to create profiles
- "Low confidence warning": Gather more samples (target 50+ for emails, 10+ for blogs)
- "Geoffrey Chrome not running":
cd ~/non-ic-code/geoffrey/skills/browser-control && ./scripts/launch-chrome.sh - "Failed to extract social posts": Log into LinkedIn/X in Geoffrey Chrome first
Development
Architecture:
- Scripts fetch raw writing samples (email, blog, social)
- Geoffrey analyzes samples natively using LLM capabilities
- No NLP libraries - cleaner, more maintainable
- Voice profiles store samples + Geoffrey's analysis
Adding New Content Types:
- Create extraction script if needed
- Add profile to
voice-profile-template.json - Add interview questions to
interview-questions.json - Update SKILL.md workflow
- Extract samples and ask Geoffrey to analyze
Future Enhancements
- Real-time learning (auto-analyze new emails monthly)
- A/B testing (generate 2 versions, learn from choices)
- Collaborative voice (blend user + co-author)
- Multi-language (Spanish voice profiles)
- Performance tracking (voice drift detection)
- Audience-aware sub-profiles
Version
v0.1.0 - Initial implementation
- 6 voice profiles
- Sample extraction scripts
- 6-phase workflow with checkpoints
- Native LLM validation
- Obsidian integration