• Incremental Maintenance Mode (default): When the user informs "added a file", append index entries for a single file
• Health Check Mode: When the user says "index health check", scan the entire library ↔ INDEX for comparison, batch fill in missing entries, and clean up invalid entries

• Core Topic: A one-sentence summary of what this content is about
• Applicable Scenarios: What types of topics/content will use it (specific to scenarios, e.g., "talking about procrastination" "talking about product storytelling")
• Keywords: 5-8 terms, prioritizing counterintuitive concepts, those with specific operations, or unique perspectives

• What is the table column structure?
• Is the partitioning by topic, workflow stage, or content type?
• How is the file path written? Are there special characters?
• Is there a decision rule section?

• The new content covers topic types not included in the rules → Add a new rule
• The new content is a valid supplement in a certain direction → Append "or see 'file name'"
• Highly overlapping with existing entries → Skip

Content Material Library/Reading Notes Library/INDEX.md

• Partition Logic: Partitioned by topic (Personal Growth/Psychology/Communication/Entrepreneurship/Philosophy/Technology/Creation)
• Special Requirements: Book files are usually very long; you must read the entire text, not just the first few paragraphs; the core viewpoints of books are often in the middle to later sections
• Path Notes: Book titles may contain multiple spaces (e.g.,

  《Hackers & Painters》  Paul Graham.md

  has double spaces), the path must match the actual file name exactly

Content Material Library/Core Concepts Library/INDEX.md

• Partition Logic: Partitioned by workflow stage (Topic Selection Stage/Script Writing Stage/Title & Cover Stage), each file represents a set of methodologies or templates
• Keyword Direction: Focus on method framework names (e.g., "Viral Formula" "Cognitive Hijacking"), rather than specific viewpoints in books
• Decision Rules: Rules should be linked to specific creation steps (which one to read when deepening topics, which one to read when generating scripts)

Content Material Library/Viral Script Library/Viral Index.md

• Different Trigger Timing: This index is only updated when post-publishing data meets viral standards, not when new files are added
  • Balanced Viral: Both likes and saves ≥ 2000+
  • Save-Focused Viral: Save ratio > 1.5
  • Follower-Growth Viral: Follower growth rate > 2%
• Completely Different Format: Each entry is a structured item containing data/type/file path/reusable structure/golden sentence; there is no table or decision rules
• Writing Format (refer to existing entries):
  markdown

  #### [Content Title]
  - **Data**: X likes, X saves (save ratio X), X new followers (follower growth rate X%)
  - **Type**: [Content type, e.g., Methodology/Tool Tutorial/Experience]
  - **Path**: `[File path]`
  - **Reusable Structure**:
    - Opening: [Opening method]
    - Main Body: [Main body structure]
    - Ending: [Ending method]
  - **Golden Sentence**: "[Most representative sentence]"

1. Use Glob to find all files under the library
2. Read each file completely one by one (cannot skip; batch parallel reading is allowed)
3. Refer to the format of

   Content Material Library/Core Concepts Library/INDEX.md

   to create a new file, and add AI usage rules at the beginning
4. After creation, update li-topic/SKILL.md, li-writer/SKILL.md, and CLAUDE.md so that the AI creation process is aware of the new index

• Use Glob to list actual files under the library
• Read the corresponding INDEX.md, extract recorded relative paths from table rows/links
• Calculate two sets:

  • Missing

    = Actual existing files - Recorded in INDEX

  • Invalid

    = Recorded in INDEX - Actual existing files

📋 Index Health Check Report

📚 Reading Notes Library
  Missing (X entries):
    - [File path]
  Invalid (X entries):
    - [Entries existing in INDEX but files no longer exist]

📚 Core Concepts Library
  ...

📚 04-Methodology Precipitation
  ...

Total: X missing entries, X invalid entries

"Above is the diff. Do you want to fill in all of them? Or select a few? Should invalid entries be deleted or marked first?"

• Missing entries usually come from

  02-Published Content/

  rather than the material library itself
• When writing to

  Viral Index.md

  , use the existing structured format (data/type/path/reusable structure/golden sentence); do not apply the general INDEX table format
• The original text path should be the real path of the published content, not the path within the script library

• If user selects "Delete" → Remove the corresponding row from the INDEX table/links
• If user selects "Mark" → Append

  ⚠️ File no longer exists, pending confirmation

  at the end of the entry

✅ Health Check Completed
  Filled in: X entries → [List new entries]
  Cleaned up: X entries
  Retained for manual processing: X entries

1. Path writing must match INDEX: Some INDEX use

   relative-path.md

   , some use

   [name](path)

   ; you must identify the existing format
2. Same name different path is not considered missing: For example, INDEX writes

   《X》.md

   but the actual file is

   《X》  Author.md

   (with double spaces), this is a path error rather than missing; you should Ask the user first
3. Do not read the Viral Script Library: Explained above
4. Prioritize parallel Read for large libraries: Methodology Precipitation may have dozens of files; read all in parallel before starting to write