The Agent Skills Directory

[SAFE]: The skill provides high-quality reference architectures and implementation scripts for RAG systems. It is informative and does not contain malicious code.
[EXTERNAL_DOWNLOADS]: The skill references and downloads dependencies for well-known and trusted services including OpenAI, Anthropic, Voyage AI, Pinecone, and Tavily. These are standard components for the skill's stated purpose and are documented neutrally.
[PROMPT_INJECTION]: Prompt templates included in the skill (such as those for Self-RAG and basic RAG) incorporate grounding instructions like 'Answer using ONLY the provided context'. These are best-practice defensive measures to minimize hallucinations and do not attempt to bypass safety filters.
[DATA_EXFILTRATION]: The skill does not access sensitive local files or exfiltrate data. Credentials in templates are managed via environment variables, and network operations are confined to the intended AI service APIs.
[COMMAND_EXECUTION]: No dangerous shell command execution or subprocess spawning was found. File system interactions are limited to standard PDF and image processing logic required for multimodal RAG.

rag-retrieval