The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides templates for building RAG (Retrieval-Augmented Generation) and agentic systems that process untrusted data from external Weaviate collections and user prompts.
Ingestion points: Untrusted data enters the agent context through the user_prompt field in the RouterAgent and retrieval results from Weaviate collections (e.g., references/basic_agent.md, references/agentic_rag.md).
Boundary markers: The provided templates generally lack explicit boundary markers or instructions to ignore embedded commands within retrieved context, though some signatures like Generator in advanced_rag.md include instructions to use only provided context.
Capability inventory: The generated applications utilize tool-calling capabilities (RouterAgent.get_response) where the LLM can trigger arbitrary Python functions defined by the user.
Sanitization: The code templates do not implement specific sanitization or escaping for external content before interpolation into prompts.
[REMOTE_CODE_EXECUTION]: The documentation includes setup instructions that involve piping remote scripts to a shell for environment preparation.
Evidence: references/pdf_multimodal_rag.md contains commands to install uv (curl -LsSf https://astral.sh/uv/install.sh | sh) and Ollama (curl -fsSL https://ollama.com/install.sh | sh). Both target official domains for well-known developer tools and are documented neutrally as part of the project setup process.

weaviate-cookbooks