doc-classifier

Strategy:

1. Path pattern first (docs/api/ → api)
2. Content analysis fallback (read fractary_doc_type from frontmatter) </CONTEXT>

<CRITICAL_RULES>

• ALWAYS try path-based detection first
• ONLY read file if path detection fails
• NEVER modify files
• ALWAYS return confidence score (0-100) </CRITICAL_RULES>

2. Content-Based Detection

   • Read file (if file_path) or use content
   • Extract frontmatter
   • Read fractary_doc_type field
   • Return with confidence 90

3. Fallback

   • Return _untyped with confidence 50 </WORKFLOW>