civil-judgment-taiwan-vectorstore

What This Skill Does

The civil-judgment-taiwan-vectorstore skill is a specialized agent capability designed to process, vectorize, and store Taiwan civil court judgments (民事判決) into a Qdrant vector database. It functions as a robust bridge between raw legal document archives and searchable AI memory. By leveraging Ollama for embeddings and a centralized ingest.py script, this skill ensures that legal practitioners and researchers can perform semantic searches across massive datasets of civil rulings with high traceability and data integrity.

Installation

To integrate this skill into your environment, run the following command in your terminal: clawhub install openclaw/skills/skills/alex02131926/civil-judgment-taiwan-vectorstore

Ensure you have your virtual environment activated by running source {baseDir}/.venv/bin/activate before triggering the ingestion processes.

Use Cases

• Legal Research: Instantly surface relevant civil court precedents by querying natural language questions against thousands of vectorized PDF or HTML judgments.
• Trend Analysis: Automate the batch processing of new judicial filings to identify shifts in judicial reasoning or sentencing patterns over time.
• Archive Management: Transform unstructured, heterogeneous collections of court documents into a structured, queryable knowledge base with built-in deduplication.

Example Prompts

1. "OpenClaw, ingest the files in /data/judgments/20260305_142030 and provide a summary of the ingestion report once finished."
2. "Can you check if the files in /mnt/cases/archive have been processed, and if not, run the ingestion script for me?"
3. "Summarize the results of the latest ingestion for the folder /outputs/judicialyuan/batch_01, focusing on any files that returned errors."

Tips & Limitations

• Data Integrity: The skill treats the archive/ folder as an immutable source of truth. Do not modify or delete files within this directory; the script relies on this structure for idempotent re-runs.
• Safety: Always verify that your run folder contains valid HTML or PDF files before initiating the process. If no valid files are detected, the system will prevent execution to save resources.
• Workflow: For high-volume batches, always inspect the generated ingest_report.md immediately following execution to identify potential parsing edge cases or connectivity issues with the vector database.