The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The implementation guides provide instructions for installing development tools such as uv (from astral.sh) and ollama (from ollama.com). Both are well-known services for AI application development.
[DATA_EXPOSURE]: The skill includes a dedicated reference for environment requirements and strictly recommends the use of .gitignore and security configurations (e.g., .claude/settings.json) to prevent the accidental exposure of sensitive API keys and credentials.
[PROMPT_INJECTION]: The cookbooks describe Retrieval-Augmented Generation (RAG) architectures which ingest untrusted data from external sources such as user queries and PDF documents. This creates a surface for potential indirect prompt injection attacks.
Ingestion points: references/pdf_multimodal_rag.md (PDF processing), references/basic_rag.md (user input handling).
Boundary markers: The guides suggest using structured output frameworks like DSPy, which help separate instructions from data, though they do not provide specific delimiter implementations.
Capability inventory: Retrieval operations via weaviate-client and generation via ollama or other LLM providers.
Sanitization: The skill explicitly instructs developers to review generated applications for security, data privacy, and prompt injection exposure before production deployment.

weaviate-cookbooks