The Agent Skills Directory

[SAFE]: The skill functions as an educational and reference guide for AI/ML engineering, providing boilerplate code for embeddings, vector search, and monitoring.
[EXTERNAL_DOWNLOADS]: The skill utilizes several industry-standard Python libraries for AI development, including openai, lancedb, polars, and sentence-transformers.
[PROMPT_INJECTION]: The skill implements a Retrieval-Augmented Generation (RAG) pipeline, which creates a surface for indirect prompt injection where untrusted data could influence model behavior.
Ingestion points: Untrusted document text is ingested through the retrieve method in RAGPipeline (references/rag-pipelines.md).
Boundary markers: The build_rag_prompt function employs delimiters such as <|system|> and "Context:" headers to structure the prompt.
Capability inventory: The skill possesses capabilities to perform LLM inference and embedding generation via the OpenAI API.
Sanitization: The code samples use simple string concatenation to assemble prompts from retrieved context without specific sanitization or escaping mechanisms.
[COMMAND_EXECUTION]: All data processing and storage operations are performed using standard library interfaces (LanceDB, DuckDB) without the use of dangerous system commands or shell execution.
[DATA_EXFILTRATION]: Network operations are confined to legitimate interactions with well-known services (OpenAI, AWS S3) for standard pipeline tasks.

engineering-ai-pipelines