The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion and processing of external documents using SimpleDirectoryReader, which creates a surface for potential indirect prompt injection attacks where untrusted data could influence agent responses.
Ingestion points: External files are loaded into the agent context in templates/basic-rag.py, templates/agentic-rag.py, and templates/custom-workflow.py.
Boundary markers: The templates rely on the framework's default prompt orchestration; they do not explicitly demonstrate the use of XML delimiters or specialized sanitization to isolate untrusted content.
Capability inventory: The skill provides capabilities for file system access, network communication with LLM providers, and complex logic execution via Workflows.
Sanitization: No explicit filtering or sanitization of document content is implemented in the reference templates.
[DYNAMIC_EXECUTION]: The verification utility scripts/check-setup.py uses dynamic importing to confirm the presence of LlamaIndex-related dependencies.
Evidence: The script uses importlib.import_module() to check for packages listed in the REQUIRED_PACKAGES and OPTIONAL_PACKAGES arrays.
Context: This is a standard and safe diagnostic practice as the package names are hardcoded and not derived from untrusted user input.

llamaindex