The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates an indirect prompt injection surface by parsing untrusted documents and presenting the extracted text to the agent. 1. Ingestion points: The parse method in SKILL.md and the batch_parse_dir.py script process external files from the filesystem or URLs. 2. Boundary markers: The skill does not define delimiters or instruct the agent to treat the output as untrusted content, increasing the risk of the agent following instructions embedded in documents. 3. Capability inventory: The agent is configured with the Bash, Write, and Edit tools, which are high-impact capabilities that could be exploited via injection. 4. Sanitization: No sanitization or filtering is performed on the extracted text before it is returned to the agent.
[EXTERNAL_DOWNLOADS]: The skill depends on the installation of external packages and system utilities for its core functionality. It references the liteparse package on PyPI and @llamaindex/liteparse on npm, as well as system tools like LibreOffice, ImageMagick, and Tesseract. These references target well-known repositories and official installation paths consistent with the skill's purpose.

liteparse