The Agent Skills Directory

[SAFE]: The skill relies on the 'liteparse' library and its associated Node.js bindings, which are maintained by LlamaIndex, a well-known and recognized organization in the AI technology sector. Installations from official registries like PyPI and npm for these packages are considered safe and follow standard developer practices.\n- [SAFE]: The tool is architected for local document processing. It utilizes a Rust-based core to handle file conversions and text extraction on the user's host system without requiring cloud-based API calls or external processing, thereby protecting document privacy and reducing exfiltration surfaces.\n- [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface because it is designed to ingest and parse untrusted external data (PDFs, Word documents, and images). This is a standard risk for document-parsing skills.\n
Ingestion points: Untrusted data enters the agent's context through the LiteParse.parse() Python method and the lit parse CLI command, as documented in SKILL.md and scripts/batch_parse_dir.py.\n
Boundary markers: The instructions do not define specific boundary markers or 'ignore' instructions for the agent to use when interpreting the extracted text, which could allow malicious instructions inside a document to influence agent behavior.\n
Capability inventory: The skill allows the agent access to the Bash, Read, Write, and Edit tools, which provides a significant capability surface if an indirect injection were successful.\n
Sanitization: The parser extracts content as-is to preserve layout and metadata; it does not implement semantic filtering to identify or neutralize potential prompt injection payloads embedded within document text or metadata.

liteparse