Civil Judgment Taiwan Vectorstore
AdvisoryAudited by VirusTotal on Mar 29, 2026.
Overview
Type: OpenClaw Skill Name: civil-judgment-taiwan-vectorstore Version: 1.0.0 The skill bundle is a legitimate tool for ingesting Taiwan civil court judgments into a Qdrant vector database using Ollama embeddings. The core logic in `scripts/ingest.py` and `scripts/build_reasoning_collection.py` uses standard libraries (BeautifulSoup, pypdf, requests) to parse documents and interact with local/configured service endpoints. The instructions in `SKILL.md` are well-structured, providing clear operational boundaries for the AI agent without any signs of prompt injection or malicious intent. No evidence of data exfiltration, unauthorized execution, or obfuscation was found.
Findings (0)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
Judgment content and local file paths can remain searchable in Qdrant after ingestion.
The skill stores judgment text chunks and local file paths in Qdrant payloads, making them persistent retrievable context.
`text` | string | Chunk plain text (for payload retrieval) ... `local_path` | string | Absolute path to archived HTML file
Only ingest intended folders, protect access to Qdrant, and delete or rebuild collections when you no longer want the data retained.
If the user points these URLs at remote or untrusted services, judgment text or embeddings could leave the local machine.
The skill sends embedding and database operations to configured Ollama and Qdrant HTTP endpoints; defaults are local, but the endpoints can be changed.
OLLAMA_URL=http://localhost:11434 QDRANT_URL=http://localhost:6333 ... --ollama http://localhost:11434 --qdrant http://localhost:6333
Use localhost or trusted private endpoints; if using remote services, ensure network access, authentication, and data-handling expectations are appropriate.
Using `--rebuild` can remove existing `civil_case_reasoning` data before recreating it, which matters on shared or production Qdrant instances.
The helper script includes an explicit user-controlled rebuild option that deletes and recreates a Qdrant collection.
`--rebuild` ... `Drop and recreate civil_case_reasoning` ... `client.delete_collection(REASONING_COLLECTION)`
Use rebuild only when intended, confirm the target Qdrant URL first, and avoid running it against shared collections unless you are authorized.
A future install may resolve different package versions than the author tested.
The setup command installs packages without pinned versions, which is a normal but notable dependency-provenance and reproducibility consideration.
pip install requests beautifulsoup4 pypdf qdrant-client
Install in an isolated virtual environment and consider pinning or reviewing dependency versions before production use.