The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to extract text and data from external files and URLs. This presents a risk where malicious instructions contained within a processed document could be interpreted by the agent as legitimate commands.
Ingestion points: Documents are processed via scripts/parse-pdf.py, scripts/parse-docx.py, scripts/parse-html.py, and templates/multi-format-parser.py.
Boundary markers: Extracted text is returned without delimiters or warnings to the agent to treat the content as data rather than instructions.
Capability inventory: The agent is granted access to high-privilege tools such as Bash, Write, and Edit.
Sanitization: There is no evidence of sanitization or filtering of the extracted content before it enters the agent's context.
[PRIVILEGE_ESCALATION]: The script scripts/setup-unstructured.sh utilizes sudo apt-get install to install system-level dependencies like poppler-utils and tesseract-ocr. While these are legitimate requirements for the parser, the use of sudo within a skill script requires administrative rights.
[COMMAND_EXECUTION]: The setup scripts (scripts/setup-llamaparse.sh and scripts/setup-unstructured.sh) execute several shell commands to manage environment variables, install packages via pip, and configure the local environment.
[DYNAMIC_EXECUTION]: The setup scripts dynamically generate and execute Python test scripts (test_llamaparse.py and test_unstructured.py) to verify the installation of dependencies. This is a low-risk pattern but involves generating executable code at runtime.

document-parsers