The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends and provides implementation examples for numerous industry-standard AI and data processing libraries, including langchain, llama-index, sentence-transformers, and transformers. It also references official models from Hugging Face and API services from OpenAI, Pinecone, and Weaviate. All referenced sources are established organizations or well-known services within the development community.
[INDIRECT_PROMPT_INJECTION]: The core functionality involves ingesting and processing external document content for retrieval systems. While the skill includes advanced methods like 'Contextual Retrieval' (which uses an LLM to summarize chunks), it does not implement specific sanitization to prevent instructions embedded within source documents from influencing the agent. However, this is a standard risk inherent to RAG processing and is handled through typical prompt engineering patterns.
[DATA_EXPOSURE_AND_EXFILTRATION]: Several code examples demonstrate how to interact with third-party APIs (e.g., OpenAI). These snippets correctly use variables and placeholders for sensitive information like API keys, advising users to manage these credentials securely rather than hardcoding them.
[COMMAND_EXECUTION]: The skill provides Python snippets and bash commands for metric evaluation and library installation. These are intended for user-driven optimization and follow standard development workflows without including hidden or malicious commands.

chunking-strategy