The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides templates for LLM integration and RAG pipelines that ingest untrusted external data without sufficient boundary markers or sanitization.
Ingestion points: references/llm_integration_guide.md contains a system prompt template that interpolates {product_context}. references/rag_system_architecture.md contains a RAG prompt template that interpolates {context} and {question}.
Boundary markers: The provided code templates lack delimiters (e.g., XML tags, triple quotes) or explicit instructions for the model to ignore embedded commands within the interpolated context.
Capability inventory: The skill documentation describes pipelines that utilize LLM provider abstractions (OpenAI, Anthropic) and vector store retrievals (Pinecone, Qdrant) to perform completions.
Sanitization: The example code snippets for chunking and retrieval do not include steps for sanitizing, escaping, or validating the content retrieved from external sources before interpolation.
[COMMAND_EXECUTION]: The skill includes multiple executable Python scripts in the scripts/ directory designed for MLOps tasks. Analysis of scripts/ml_monitoring_suite.py, scripts/model_deployment_pipeline.py, and scripts/rag_system_builder.py confirms they are standard boilerplate utilities that handle configuration and argument parsing safely without spawning dangerous subprocesses or executing arbitrary strings.

senior-ml-engineer