Content Cleanup Loop Skill

Purpose

Systematically clean up a folder of articles by linting, creating a prioritized todo list, fixing issues, and repeating until all content passes ContentLint checks.

When to Use

• •User says "clean up my articles"
• •User wants to lint and fix a directory of content
• •User asks to "run contentlint and fix issues"
• •User wants automated content cleanup

Instructions

You are implementing a systematic content cleanup workflow. Your goal is to take a directory of content files, lint them, identify issues, fix them, and repeat until all files pass ContentLint checks.

Step 1: Initial Lint and Assessment

First, run ContentLint on the specified directory:

cd /path/to/contentlint
python3 -m contentlint lint /path/to/content-folder --format json --out /tmp/lint-report.json

Step 2: Analyze Results and Create Todo List

Read the lint report and create a prioritized todo list using the TodoWrite tool:

Priority order:

1. •FAIL items - Must be fixed (highest priority)
2. •AI detection patterns - Shows AI-generated content
3. •WARN items - Should be reviewed/fixed

Organize by:

• •File with most issues first
• •Within each file: FAIL → AI patterns → WARN

Create todos like:

- Fix 'article1.md': 5 FAIL items (ai-vocabulary, banned-words)
- Fix 'article2.md': 3 FAIL items (knowledge-cutoff, promotional-language)
- Review 'article3.md': 8 WARN items (conversational-hooks, meta-commentary)

Step 3: Work Through Files Systematically

For each file in your todo list:

1. •

   Read the file

2. •

   Identify the specific issues from the lint report

3. •

   Fix the issues:

   • •

     AI vocabulary - Replace AI words with natural alternatives

     • •"delve into" → "explore" or "examine"
     • •"underscore" → "emphasize" or "show"
     • •"tapestry" → "collection" or specific term
     • •"pivotal" → "important" or "key"
     • •"landscape" → specific term (market, field, etc.)
   • •

     Banned words (filler) - Remove or rephrase

     • •"really", "very", "just" → often can be deleted
     • •"that" → rewrite sentence structure
   • •

     AI patterns - Rewrite to sound human

     • •"Here's the thing:" → Delete or rephrase naturally
     • •"I've compiled" → Delete meta-commentary
     • •Parallel emphasis → Vary sentence structure
     • •"Not only...but" → Rewrite more naturally
   • •

     Promotional language - Neutralize

     • •"boasts" → "has" or "includes"
     • •"nestled" → "located"
     • •"vibrant" → Delete or be specific
   • •

     Knowledge cutoff disclaimers - Delete entirely

     • •"Based on available information" → Delete
     • •"as of my last update" → Delete
4. •

   Save the fixes using the Edit tool

5. •

   Mark the todo as completed

6. •

   Re-lint just that file to verify fixes:

   bash

   python3 -m contentlint lint /path/to/fixed-file.md
   

Step 4: Loop Until Clean

After fixing each file:

1. •Re-run ContentLint on the entire directory
2. •Check if any FAIL items remain
3. •If issues remain:
   • •Update todo list with remaining issues
   • •Continue to next file
4. •If all files pass:
   • •Report completion
   • •Show final summary

Step 5: Final Report

When all files are clean (or only acceptable WARN items remain):

✅ Content cleanup complete!

Summary:
- Files processed: 15
- FAIL items fixed: 47
- WARN items fixed: 23
- AI patterns removed: 31

Remaining WARN items: 8 (acceptable - natural repetition in lists)

All files now pass ContentLint FAIL threshold.

Important Guidelines

DO:

• •✅ Work on ONE file at a time
• •✅ Re-lint after each file to verify fixes
• •✅ Preserve the author's voice and meaning
• •✅ Mark todos completed as you go
• •✅ Show progress after each file

DON'T:

• •❌ Try to fix all files at once
• •❌ Remove personality or brand voice
• •❌ Change technical terms or product names
• •❌ Fix acceptable WARN items (like repetition in question lists)
• •❌ Make the content generic by over-editing

When to Stop:

Stop when:

1. •All FAIL items are fixed
2. •All AI detection patterns are removed or justified
3. •Only acceptable WARN items remain (repetition in lists, intentional conversational hooks, etc.)

Ask the user if they want to continue with WARN items, or if FAIL-only cleanup is sufficient.

Example Workflow

User: "Clean up my articles in /content/blog"

Assistant:

1. •Runs ContentLint on /content/blog
2. •Creates todo list:
   code

   - Fix article-1.md: 8 FAIL (ai-vocabulary, significance-language)
   - Fix article-2.md: 5 FAIL (knowledge-cutoff, promotional-language)
   - Review article-3.md: 12 WARN (conversational-hooks, repetition)
   

3. •Works on article-1.md:
   • •Reads file
   • •Identifies: "delve into" (3x), "underscore the importance" (2x), "stands as testament"
   • •Fixes issues
   • •Saves file
   • •Re-lints to verify
   • •Marks todo complete
4. •Works on article-2.md:
   • •Same process
5. •Continues until all FAIL items fixed
6. •Reports completion:
   code

   ✅ All FAIL items fixed! 3 files cleaned.
   Remaining: 12 WARN items (mostly natural repetition)
   
   Continue with WARN items? [y/n]
   

Monitoring Progress

After each file:

Progress: 2/5 files cleaned
Current: article-3.md (4 FAIL items remaining)

Edge Cases

If a file has too many issues (20+):

• •Ask user: "article-1.md has 23 issues. This might be AI-generated. Rewrite from scratch or try to fix?"

If fixes break content:

• •Show diff
• •Ask for approval before saving

If stuck on an issue:

• •Skip that specific issue
• •Note it in final report
• •Continue with other issues