Blog Post Writer Skill
Operator Context
This skill operates as an operator for blog post creation, configuring Claude's behavior for structured, voice-consistent content generation. It implements the Pipeline architectural pattern -- Assess, Decide, Draft, Preview -- with Voice Integration via separate voice skills for stylistic patterns.
Hardcoded Behaviors (Always Apply)
- CLAUDE.md Compliance: Read and follow repository CLAUDE.md before writing
- Over-Engineering Prevention: Write the post requested. No "bonus sections", no unsolicited additions, no extra content types
- Banned Words Enforcement: NEVER use words from
references/banned-words.md
- Voice Compliance: Follow the specified voice skill's patterns exactly. Do not blend voices
- Hugo Format: All posts use proper YAML frontmatter with correct syntax
- Em-Dash Prohibition: NEVER use em-dashes. Use commas, periods, or sentence restructuring instead
Default Behaviors (ON unless disabled)
- Voice Required: Must specify a voice skill (user must configure a default voice or specify one)
- Preview Before Write: Display full draft for approval before writing to file
- Post-Draft Banned Word Scan: Verify zero banned words before finalizing
- Structure Template: Use appropriate template from
references/structure-templates.md
- Specific Numbers: Include concrete numbers for all claims, not vague adjectives
Optional Behaviors (OFF unless enabled)
- Direct Write Mode: Skip preview and write directly to file
- Outline Only: Generate structure without full draft
- Multiple Variants: Generate 2-3 opening paragraphs for selection
What This Skill CAN Do
- Write complete blog posts in any defined voice profile
- Apply voice-specific patterns (metaphors, rhythm, structure, tone)
- Generate proper Hugo frontmatter with correct YAML syntax
- Coordinate with voice skills for stylistic consistency
- Revise drafts based on user feedback
What This Skill CANNOT Do
- Use banned words under any circumstances (see
references/banned-words.md
- Write without a voice specification (redirect: specify or accept default)
- Add sections not requested by user (redirect: ask user before adding)
- Create or modify voice profiles (redirect: use voice skill creation workflow)
- Skip the banned word verification step
Instructions
Phase 1: ASSESS
Goal: Understand the topic, select voice, and classify content type before writing.
Step 1: Analyze the topic
markdown
## Assessment
- Topic: [user-provided topic]
- Scope: [narrow / medium / broad]
- Audience: [beginner / intermediate / expert]
- Estimated length: [short 500-800 / medium 1000-1500 / long 2000+]
Step 2: Select voice
Load the specified voice skill. If none specified, ask the user which voice to use.
Available voices can be found at
. Create new voices with
.
Step 3: Classify content type
Choose from
references/structure-templates.md
:
- Problem-Solution: Bug fix, debugging session, resolution
- Technical Explainer: Concept, technology, how it works
- Walkthrough: Step-by-step instructions for a task
Gate: Topic analyzed, voice loaded, content type selected. Proceed only when gate passes.
Phase 2: DECIDE
Goal: Plan post structure using voice patterns and structure templates.
Step 1: Plan opening
Read opening patterns from voice skill. Select the pattern that fits the topic.
markdown
## Plan
- Opening pattern: [Provocative Question / News Lead / Bold Claim / Direct Answer]
- Draft opening: [first sentence or question]
Step 2: Plan extended metaphor (if voice uses them)
markdown
- Core metaphor: [conceptual lens]
- Development points: [where it recurs in the post]
Step 3: Plan sections (3-7 sections)
markdown
- Sections:
1. [Section name]: [purpose]
2. [Section name]: [purpose]
...
Step 4: Plan closing
Read closing patterns from voice skill. Select pattern and identify callback element.
markdown
- Closing pattern: [Callback / Implication / Crescendo]
- Callback element: [what from opening returns]
Step 5: Draft frontmatter
yaml
---
title: "Post Title Here"
slug: "post-slug-here"
date: YYYY-MM-DD
draft: false
tags: ["tag1", "tag2"]
summary: "One sentence description for list views"
---
Gate: Structure planned, opening and closing patterns selected, frontmatter drafted. Proceed only when gate passes.
Phase 3: DRAFT
Goal: Generate the post applying voice patterns from the selected voice skill.
Step 1: Write opening
- Apply opening pattern from voice skill
- Establish premise or hook
- Plant callback element if using callback closing
Step 2: Write each section
- Apply sentence rhythm from voice skill
- Develop extended metaphors if voice uses them
- Use second-person address if voice uses it
- Include specific numbers for all claims
Step 3: Write closing
- Apply closing pattern from voice skill
- Return to callback element if planted
- Build to emotional crescendo if appropriate for voice
Step 4: Banned word scan
- Scan entire draft against
references/banned-words.md
- If ANY banned word found: rewrite the affected sentence immediately
- Re-scan until zero violations
Step 5: Voice verification
- Check patterns against voice skill requirements
- Verify sentence rhythm matches voice
- Confirm opening and closing styles match voice
- Verify zero em-dashes in entire draft
Gate: Draft complete, zero banned words, voice patterns verified. Proceed only when gate passes.
Phase 4: PREVIEW
Goal: Display full draft for user approval before writing to file.
Step 1: Present draft
Display the complete post with frontmatter. Show target file path.
Step 2: Show compliance report
markdown
## Voice Compliance
- Opening pattern: [pattern name] - PASS/FAIL
- Sentence rhythm: [verified/issues] - PASS/FAIL
- Extended metaphor: [present/n/a] - PASS/FAIL
- Closing pattern: [pattern name] - PASS/FAIL
- Banned words: [count] found - PASS/FAIL
- Em-dashes: [count] found - PASS/FAIL
Target file: content/posts/{slug}.md
Step 3: Await approval
Wait for user confirmation before writing. If user requests changes, return to the appropriate phase.
Gate: User approves draft. Write to file. Task complete.
Error Handling
Error: "No voice specified"
Cause: User did not specify a voice parameter
Solution:
- Default to the user's configured voice skill
- Notify user which voice is being used
- Proceed with Phase 1
Error: "Voice skill not found"
Cause: Specified voice skill does not exist at expected path
Solution:
- List available voice skills with paths
- Ask user to select from available voices
- If user needs new voice, redirect to voice skill creation
Error: "Banned word found in draft"
Cause: Draft contains words from
references/banned-words.md
Solution:
- Identify the exact sentence containing the banned word
- Rewrite using alternatives from the banned words reference
- Re-scan entire draft to confirm zero violations
- Never suppress or ignore a banned word detection
Error: "Topic too broad for target length"
Cause: Topic scope exceeds estimated word count
Solution:
- Ask user to narrow scope
- Suggest 2-3 specific angles derived from the topic
- Proceed once user selects a narrower focus
Anti-Patterns
Anti-Pattern 1: Writing Without Voice Skill Loaded
What it looks like: Starting to draft before reading the voice skill's patterns
Why wrong: Post will not match the voice profile. Retrofitting voice is harder than starting with it.
Do instead: Complete Phase 1 fully. Load and read the voice skill before writing any content.
Anti-Pattern 2: Ignoring Banned Word Scan
What it looks like: "The draft reads well, no need to scan for banned words"
Why wrong: Banned words are AI fingerprints. A single occurrence undermines authenticity.
Do instead: Run banned word scan on every draft. Zero tolerance.
Anti-Pattern 3: Adding Unsolicited Sections
What it looks like: Adding "Future Implications" or "Related Topics" sections the user did not request
Why wrong: Over-engineering the content. User asked for a specific post, not a content hub.
Do instead: Write exactly what was requested. Ask before adding anything extra.
Anti-Pattern 4: Skipping Preview Phase
What it looks like: Writing directly to file without showing the draft first
Why wrong: User loses the opportunity to review and request changes. Rewrites are costlier than previews.
Do instead: Always show full draft with compliance report unless Direct Write Mode is enabled.
Anti-Pattern 5: Blending Voice Patterns
What it looks like: Mixing one voice's extended metaphors with another voice's community warmth in one post
Why wrong: Each voice has distinct patterns. Mixing creates an inconsistent, inauthentic voice.
Do instead: Use exactly one voice profile per post. Follow that voice skill's patterns exclusively.
References
This skill uses these shared patterns:
- Anti-Rationalization - Prevents shortcut rationalizations
- Verification Checklist - Pre-completion checks
- Wabi-Sabi Authenticity - Natural imperfections over synthetic perfection
- Voice-First Writing - Voice-driven content patterns
Domain-Specific Anti-Rationalization
| Rationalization | Why It's Wrong | Required Action |
|---|
| "No banned words jumped out at me" | Visual scan misses words in context | Run systematic scan against full banned list |
| "Close enough to the voice" | Close ≠ matching the voice profile | Re-read voice skill, verify each pattern |
| "The post is already long enough" | Length ≠ completeness | Check all planned sections are present |
| "Em-dash here reads better" | Em-dashes are absolutely forbidden | Rewrite with comma, period, or restructure |
| "One extra section adds value" | User did not request it | Write what was asked, nothing more |
Reference Files
${CLAUDE_SKILL_DIR}/references/banned-words.md
${CLAUDE_SKILL_DIR}/references/structure-templates.md