The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external documents such as PDFs, Office files, and images which may contain malicious instructions designed to influence the agent's behavior.
Ingestion points: Untrusted data enters via LiteParse.parse() in Python, the lit parse CLI command, and the provided batch_parse_dir.py script.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to ignore or isolate instructions embedded within the processed document content.
Capability inventory: The skill allows writing extracted content to the file system (via CLI -o or the batch script) and executes subprocesses for document conversion (e.g., LibreOffice's soffice and ImageMagick's convert).
Sanitization: The skill does not perform sanitization or filtering of the extracted text before it is returned to the agent context.
[DATA_EXFILTRATION]: The skill supports a ocr_server_url configuration parameter in both its Python API and CLI. If configured with an external URL, document pages (as images) are transmitted to that endpoint for OCR processing, creating a potential egress path for sensitive information.
[COMMAND_EXECUTION]: To support multiple document formats, the skill relies on executing the lit CLI and system utilities like LibreOffice and ImageMagick. While these are required for its primary purpose, they involve the execution of local shell commands and subprocesses.

liteparse