The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes building RAG pipelines that ingest data from external, potentially untrusted sources such as web pages, GitHub repositories, and local file systems. This introduces a surface for indirect prompt injection, where data retrieved at runtime could contain malicious instructions designed to influence the agent's subsequent actions.
Ingestion points: SimpleDirectoryReader, SimpleWebPageReader, GithubRepositoryReader, and JSONReader in SKILL.md and references/data_connectors.md.
Boundary markers: The prompt templates in SKILL.md and references/query_engines.md use basic delimiters but do not explicitly implement advanced sanitization for external content.
Capability inventory: The skill demonstrates capabilities for file system reads, network requests, and tool execution via FunctionAgent.
Sanitization: No explicit sanitization or filtering of ingested text is shown in the examples, which is standard for documentation but constitutes a risk surface in production.
[EXTERNAL_DOWNLOADS]: The instructions and scripts reference the installation of official LlamaIndex packages (e.g., llama-index, llama-index-core) and external documentation from official project domains such as llamaindex.ai, llamahub.ai, and github.com/run-llama. These are recognized as standard components and resources for this framework.

llamaindex