The Agent Skills Directory

[SAFE]: The skill serves as a documentation resource for the LlamaIndex framework. It includes standard installation instructions via pip and provides educational code snippets for loading data, creating indices, and building agents.
[INDIRECT_PROMPT_INJECTION]: The skill documents numerous data ingestion surfaces which are inherent to RAG applications. These include readers for local directories (SimpleDirectoryReader), web pages (SimpleWebPageReader, BeautifulSoupWebReader), GitHub repositories (GithubRepositoryReader), databases (DatabaseReader), and JSON APIs (JSONReader). While these represent a standard attack surface for indirect prompt injection (ingesting untrusted data into an LLM context), the documentation is consistent with the primary purpose of the framework and does not include malicious implementation patterns.
[EXTERNAL_DOWNLOADS]: The documentation mentions downloading models via SentenceTransformerRerank from Hugging Face (a well-known technology service) and installing official packages from the Python Package Index (PyPI). These operations are part of the standard usage of the framework.
[CREDENTIALS_SAFE]: Code examples use clearly labeled placeholders for sensitive information, such as "your-key" for Pinecone API keys, "your-token" for Notion integration, and "user:pass" in database connection strings. No hardcoded secrets are present.

llamaindex