The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external files (CSV, JSON, PDF) using scripts/import.py. This data is stored in the vector database and subsequently queried by AI agents in scripts/ask.py and scripts/query_search.py. This processing chain creates a surface for indirect prompt injection attacks where malicious instructions embedded in the data could influence the agent's behavior during retrieval tasks.
Ingestion points: scripts/import.py (reads CSV, JSON, and PDF files provided by the user).
Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are used when the Query Agent processes the retrieved data.
Capability inventory: Includes file reading and network writing to Weaviate in scripts/import.py, and LLM-powered natural language querying in scripts/ask.py and scripts/query_search.py.
Sanitization: The scripts perform basic data type conversion (e.g., casting strings to integers or dates) but do not perform semantic sanitization or filtering of the content for safety.
[EXTERNAL_DOWNLOADS]: The scripts/example_data.py script downloads datasets from the Hugging Face Hub (e.g., jamescalam/ai-arxiv2, pkghf/ecom-product-catalog). These downloads target a well-known service and are used to provide sample data for the user's collections as intended.

weaviate