The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of data from untrusted external sources, such as web pages, GitHub repositories, and local files, into LLM prompts. This is a core feature of the LlamaIndex framework.
Ingestion points: The skill references SimpleDirectoryReader, SimpleWebPageReader, GithubRepositoryReader, DatabaseReader, and JSONReader in SKILL.md and references/data_connectors.md.
Capability inventory: The documented tools enable network operations (web scraping and API calls), file system access, and function-calling through agent tools.
Boundary markers: Examples include PromptTemplate and MetadataFilters, though precise boundary enforcement is left to the final implementation.
Sanitization: The provided snippets do not show explicit sanitization of external content before interpolation.
[COMMAND_EXECUTION]: The documentation includes standard commands to install the llama-index ecosystem via pip. These commands target official and well-known packages used for building AI applications.
[EXTERNAL_DOWNLOADS]: The skill describes legitimate methods for fetching documentation and data from external URLs, GitHub repositories (e.g., facebook/react), and cloud-based services like Pinecone and LlamaHub. These operations are intended functionalities of the framework.

llamaindex