The Agent Skills Directory

[SAFE]: The skill provides boilerplate code and architectural patterns for building RAG systems. It correctly identifies and mitigates indirect prompt injection risks by recommending the use of boundary markers (e.g., '---') and explicit system instructions to ground the AI's responses in the provided context.
Ingestion points: HTMLLoader (fetching content from URLs) and PDFLoader / CodeLoader (reading local files).
Boundary markers: The build_rag_prompt function implements clear context delimiters and instructions to ignore external commands.
Capability inventory: The skill utilizes file system reads (pathlib), persistent vector storage (chromadb), and database operations (psycopg2).
Sanitization: Includes basic whitespace normalization (_clean) to prepare text for embedding.
[EXTERNAL_DOWNLOADS]: The skill demonstrates how to fetch remote documents using the requests library and interact with well-known AI services (OpenAI, Anthropic, Cohere) for generating embeddings and model responses, which is essential for RAG functionality.
[COMMAND_EXECUTION]: The provided Python examples use standard libraries for indexing and searching document corpora, including the use of psycopg2 for interacting with Postgres and chromadb for local vector storage, which are appropriate for the skill's purpose.

rag-implementation