The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from document files (PDF, DOCX, images), which presents a surface for indirect prompt injection.
Ingestion points: Document data enters the agent context through the lit parse output and search.py results read from local temporary files (SKILL.md, scripts/search.py).
Boundary markers: The instructions do not provide specific delimiters or "ignore instructions" warnings for the parsed content presented to the agent.
Capability inventory: The skill has the ability to execute shell commands, read/write local files, and run a bundled Python script.
Sanitization: There is no evidence of sanitization or filtering of the document content before it is processed by the agent.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill involves the installation and use of external tools and packages.
Findings: The skill requires the @llamaindex/liteparse Node.js package. This is a vendor-owned resource originating from LlamaIndex. The bundled search.py script also utilizes uv to manage Python dependencies (bm25s, aiofiles) from standard registries.
[DYNAMIC_EXECUTION]: The skill uses a bundled Python script to perform local data processing.
Findings: The search.py helper script is executed at runtime to provide ranked search results from parsed text files. It performs standard file I/O and text processing using well-known libraries.

liteparse