The Agent Skills Directory

[SAFE]: The skill serves as a legitimate technical guide and template library for RAG system design, providing production-grade code examples and architectural trade-off analyses.
[EXTERNAL_DOWNLOADS]: Demonstrates how to fetch necessary embedding models and linguistic data from reputable sources such as Hugging Face and NLTK data servers.
[COMMAND_EXECUTION]: Provides implementation patterns for interacting with various vector databases (Qdrant, Pinecone, Weaviate, Chroma, pgvector) and LLM providers via their official Python SDKs.
[PROMPT_INJECTION]: Identifies a common indirect prompt injection surface inherent in RAG applications where untrusted document content is processed. 1. Ingestion points: Found in SKILL.md and references/chunking-strategies.md via document splitting functions. 2. Boundary markers: The current architectural examples do not include explicit context delimiters (e.g., XML tags) to encapsulate retrieved data. 3. Capability inventory: The pipeline includes capabilities for external API requests (LLM providers) and vector database modifications. 4. Sanitization: Basic regex cleaning logic is suggested in the references/embedding-models.md guidance.

rag-architect