The Agent Skills Directory

Indirect Prompt Injection (HIGH): The skill is designed to build systems that ingest untrusted external data, creating a significant attack surface for indirect prompt injection.
Ingestion points: The logic in SKILL.md uses document.text as the primary data source for chunking and vector storage.
Boundary markers: Absent. The templates do not demonstrate the use of XML tags, delimiters, or system instructions to encapsulate retrieved context.
Capability inventory: The resulting pipeline is intended to drive the generation phase of an AI assistant, meaning malicious instructions in a document could hijack the assistant's behavior.
Sanitization: Absent. There is no validation or filtering logic provided for data entering the vector store or being retrieved for the LLM context.
External Dependencies (LOW): The skill's code examples rely on several external Python libraries which would need to be installed in a runtime environment.
Evidence: Code snippets reference langchain, pinecone-client, and sentence-transformers.
Credentials Safety (INFO): The code includes placeholders for API keys.
Evidence: pinecone.init(api_key="...", environment="...") in SKILL.md. While safe as a placeholder, it highlights the requirement for sensitive credentials in the resulting implementation.

rag-pipeline-builder