name: qdrant-file-exporter description: Extracts all chunks with three-tier contextual embeddings from a specific PDF file in Qdrant vector database and saves to plain text. Use when user wants to extract, export, dump, or view all chunks from a PDF document, inspect file content, save chunks for analysis, or review contextual embeddings.

Qdrant File Exporter

Extracts all text chunks with three-tier contextual embeddings from a specific PDF file stored in the Qdrant vector database and saves them to a plain text file.

When to Use This Skill

Activate this skill when the user:

• Wants to extract all chunks from a specific PDF
• Asks to export, dump, or save chunks from a document
• Needs to view all chunk content for a file
• Wants to analyze chunks outside of the vector database
• Needs to review or inspect the three-tier contextual embeddings
• Wants to see how chunks were contextualized during loading
• Says things like "export chunks from [filename]" or "dump all chunks for [file]"

How to Use

Step 1: Get the PDF Filename

Ask the user which PDF file they want to export chunks from. The filename should match exactly what's stored in the Qdrant doc field (e.g., "bcy-26-income-eligibility-and-maximum-psoc-twc.pdf").

If the user doesn't know the exact filename, suggest using the qdrant-chunk-retriever skill first to search for files.

Step 2: Run the Export Script

Execute the Python helper script:

python .claude/skills/qdrant-file-exporter/scripts/export_chunks.py "filename.pdf"

The script will:

1. Connect to Qdrant (tro-child-3-contextual collection)
2. Retrieve all chunks matching the filename
3. Save them to UTIL/[filename]_chunks.txt

Step 3: Report Results

After the script completes, inform the user:

• Total number of chunks extracted
• Output file location
• File size (if available)
• Mention that contexts are included

Example:

✅ Extracted 47 chunks from bcy-26-income-eligibility-and-maximum-psoc-twc.pdf
📄 Saved to: UTIL/bcy-26-income-eligibility-and-maximum-psoc-twc_chunks.txt
📊 Includes master context, document context, and chunk-specific contexts

Step 4: Offer Next Steps

Ask the user if they want to:

• View the exported file
• Export another PDF
• Analyze the chunk content

Examples

Example 1: Basic Export

User: "Export all chunks from the income eligibility PDF"
Assistant: "Which PDF would you like to export? Please provide the exact filename."
User: "bcy-26-income-eligibility-and-maximum-psoc-twc.pdf"
Assistant: *Runs export script*
          "✅ Extracted 47 chunks from bcy-26-income-eligibility-and-maximum-psoc-twc.pdf
           📄 Saved to: UTIL/bcy-26-income-eligibility-and-maximum-psoc-twc_chunks.txt"

Example 2: Export for Analysis

User: "I need to see all the chunks for the PSOC chart document"
Assistant: *Runs export script with bcy-26-psoc-chart-twc.pdf*
          "✅ Extracted 12 chunks from bcy-26-psoc-chart-twc.pdf
           📄 Saved to: UTIL/bcy-26-psoc-chart-twc_chunks.txt

           Would you like me to open the file or analyze the content?"

Error Handling

File Not Found in Qdrant

If the PDF filename doesn't match any documents:

• Suggest the user check the filename spelling
• Recommend using qdrant-chunk-retriever to search for available files
• List similar filenames if possible

Connection Errors

If Qdrant connection fails:

• Check QDRANT_API_URL and QDRANT_API_KEY environment variables
• Verify the collection name (tro-child-3-contextual) exists
• Suggest running python LOAD_DB/verify_qdrant.py to check connection

No Chunks Found

If the file exists but has 0 chunks:

• Verify the file was loaded correctly
• Suggest running the loader script if needed

Dependencies

• Qdrant client (qdrant-client)
• Environment variables: QDRANT_API_URL, QDRANT_API_KEY
• Collection: tro-child-3-contextual (must exist)

Output Format

The exported text file contains:

• Master Context (once at beginning): Domain-level context for all chunks
• Document Context (once at beginning): Document-specific summary
• Chunks in original document order (sorted by chunk_index from loading pipeline)
  • Header: --- Chunk N (Page X) ---
  • Chunk-specific context: [Chunk Context]: ...
  • Plain text content
• Starts from the beginning of the document