Create Article Skill

ℹ️ First, set expectations: before anything else, print one short line with this skill's version and its most recent change — the top entry of metadata.changelog above — e.g. create-article vX.Y — recent: <summary>. Then proceed.

Create publication-ready long-form articles by synthesizing insights from the user's knowledge base, following established tone of voice and structural patterns.

Quick Start

1. Search knowledge base for relevant permanent notes on the topic
2. Check Article Index to avoid duplicates and see related work
3. Scan thought leaders for their most recent takes on the topic; propose integrations
4. Create article folder with proper structure
5. Write article following tone of voice guidelines
6. Fact-check the draft - verify every claim, quote, and statistic
7. Update Article Index with new entry
8. Upload to Google Drive Articles/cornelius_drafts for review

State Dependencies

Core Workflow

Step 1: Research Phase

Search the knowledge base for relevant insights:

# read role: voice (contract: scope-mount) - an article is the user's voice, so the source pass is pinned to core explicitly
# Semantic search for topic
BRAIN_READ_SCOPE=core resources/local-brain-search/run_search.sh "your topic" --limit 15 --json

# Find connections to existing notes
BRAIN_READ_SCOPE=core resources/local-brain-search/run_connections.sh "Related Note Name" --json

Key locations to search:

• Brain/02-Permanent/ - Core insights and permanent notes
• Brain/AI Extracted Notes/ - AI-extracted insights
• Brain/Document Insights/ - Research extracts
• Brain/03-MOCs/ - Maps of Content for topic overviews

Enrich with BDG context (optional - reveals which notes are generative frameworks vs reflective):

resources/brain-graph/run_brain_graph.sh inspect "Key Source Note" --json

Step 2: Check Article Index

MANDATORY: Read the Article Index before creating any article:

Brain/04-Output/Articles/ARTICLE-INDEX.md

• Check for existing articles on similar topics
• Identify gaps in topic coverage
• Note related articles for cross-referencing

Step 3: Thought-Leader & Market Scan

Search the web for what recognized thought leaders in the article's domain have said MOST RECENTLY (prefer the last 3-6 months) on the article's topics:

• Use WebSearch for recent essays, posts, interviews, talks, and papers
• Apply resources/SOURCE-AUTHORITY.md (canonical source diet) to decide which voices and outlets to trust, demote, or reject
• For each relevant voice capture: who, their claim, where/when (link + date)

Then propose how to integrate each into the article:

• Agrees - supports the user's thesis: cite as market reinforcement
• Extends - adds a dimension the KB doesn't cover: integrate with attribution
• Contrasts - opposes the user's view: position the user's take as a deliberate contrarian stance against the named mainstream position

Record the scan and chosen integrations in _metadata.md under "External Voices" (see metadata-template.md). If no relevant recent voices are found, note that too - absence of market conversation is itself a signal.

Purpose: every article demonstrates awareness of the current market conversation - the user's perspective is positioned relative to it, never written in a vacuum.

Step 4: Create Article Structure

Create folder and files:

Brain/04-Output/Articles/[article-name]/
├── [article-name].md          # Main article
├── _metadata.md               # Creation record
└── [images if needed]

Naming: Use kebab-case (e.g., cognitive-aware-ai-systems)

Step 5: Write Article

Follow the tone of voice and structure guidelines in:

• tone-of-voice.md - Voice DNA and writing style
• article-structure.md - Templates and patterns

Weave in the integrations chosen in Step 3 - external voices appear with attribution (name, source, date), and the user's position is stated relative to them.

Step 6: Fact-Check the Draft

Verify every factual claim before finalizing:

• Internal claims: statements sourced from KB notes match what the notes actually say (spot-check each [[citation]])
• External claims: statistics, dates, quotes, and attributions verified via WebSearch against resources/SOURCE-AUTHORITY.md-tier sources
• Thought-leader quotes: exact wording, correct attribution, working link
• Recency: any "current state" claims checked against the latest available data

Fix or remove anything that fails verification. If a claim is load-bearing but unverifiable, flag it in _metadata.md and either hedge it explicitly in the text or cut it.

Step 7: Update Article Index

Add entry to ARTICLE-INDEX.md:

• Article name and link
• Created date
• Topic category
• Status (Draft)
• Notes

Step 8: Upload to Google Drive

Upload the article folder to Articles/cornelius_drafts on the shared Google Drive:

# Upload the entire article folder (with images, metadata, etc.)
python3 .claude/scripts/google/google_drive.py upload-folder "Brain/04-Output/Articles/[article-name]" 1fxxbEwS86gguYXA1hJsFesr13ruXx80S --recursive

Google Drive folder IDs: See .claude/resources/google_drive_folders.md

• Articles/cornelius_drafts folder ID: 1fxxbEwS86gguYXA1hJsFesr13ruXx80S
• This puts the article where the user and a content agent can access it for review and publishing
• The folder preserves the same structure as the local article folder

File References

For detailed guidance, see:

Key Principles

Voice Summary

the user's long-form voice transforms complex research into actionable business insights through methodical, evidence-based teaching. He challenges assumptions with data while maintaining accessibility, creating structured learning experiences that bridge academic rigor with practical application.

Essential Do's

• Open with clear learning objectives or discoveries
• Cite specific research, studies, and data points
• Use numbered lists and bullet points for complex info
• Include TL;DR or summary sections
• Define technical terms immediately after introducing them
• Build arguments systematically from evidence to conclusion
• Provide actionable frameworks readers can apply

Essential Don'ts

• Never make claims without supporting evidence
• Avoid emotional appeals without data backing
• Don't use jargon without clear explanations
• Never hedge with "maybe" or passive voice
• Avoid corporate buzzwords ("synergy", "leverage", "utilize")

Platform-Specific Guidelines

Adjust content based on target platform:

Platform Adjustments

Substack/Newsletter:

• More personal voice allowed
• Can use "I've found..." openings
• Encourage replies/engagement at end
• Include "Further reading" section

Medium:

• Strong SEO-friendly title
• Compelling subtitle
• Use images/diagrams
• End with clear takeaway

LinkedIn:

• Shorter paragraphs (2-3 sentences)
• More white space
• Lead with insight, not setup
• Business application focus
• End with question to drive engagement

Content-Agent Integration

This skill can be called headlessly by content agent for content production:

cd $PROJECT_ROOT
claude -p "/create-article <topic> for <platform>" --output-format json

Caller scope: this is the workstation invocation path - a content agent runs on the user's machine with an authenticated Claude Code CLI. A caller running inside a Trinity container cannot use claude -p (turn-scoped credentials, plus the ent#643 no-CLI-subprocess rule); it must call chat_with_agent("cornelius", "...") over Trinity MCP instead.

Response format for a content agent:

{
  "result": "[Full article text with [[note citations]]]",
  "metadata": {
    "word_count": 1847,
    "platform": "substack",
    "cited_notes": ["Note 1", "Note 2", "Note 3"]
  }
}

When called by a content agent:

• Return article text directly (don't save to file)
• Include cited notes list
• Focus on the user's unique/contrarian perspectives

Output Location

All articles go to: Brain/04-Output/Articles/[article-name]/

After creation, open the folder:

open "Brain/04-Output/Articles/[article-name]/"

Completion Checklist

• Knowledge base searched for relevant permanent notes
• Article Index checked for duplicates and related work
• Thought-leader scan completed; external voices integrated or their absence noted in _metadata.md
• Article folder created with kebab-case naming
• Main article written following tone of voice guidelines
• All factual claims fact-checked; quotes and statistics verified with working links
• _metadata.md created with source insights, external voices, and thinking process
• Article Index updated with new entry (date, topic, status)
• Article folder uploaded to Google Drive Articles/cornelius_drafts
• Article folder opened in Finder for user review