The Agent Skills Directory

[Privilege Escalation] (HIGH): The script scripts/setup-unstructured.sh executes sudo apt-get install to install system dependencies (poppler-utils, libmagic, tesseract-ocr). Running commands with elevated privileges is a high-risk activity for AI agent skills.
[Credential Exposure] (MEDIUM): scripts/setup-llamaparse.sh prompts the user for an API key and writes it directly to a .env file. While standard for many applications, this handles sensitive credentials in plain text on the local filesystem.
[Dynamic Code Execution] (MEDIUM): Both scripts/setup-llamaparse.sh and scripts/setup-unstructured.sh use 'heredocs' to generate and then execute new Python scripts (test_llamaparse.py and test_unstructured.py) at runtime to verify the installation.
[Data Exposure & Exfiltration] (LOW): scripts/parse-html.py uses the requests library to fetch content from arbitrary URLs provided by the user. This constitutes a network operation to non-whitelisted domains.
[Indirect Prompt Injection] (LOW): The skill possesses a significant attack surface for indirect prompt injection.
Ingestion points: scripts/parse-docx.py (DOCX files) and scripts/parse-html.py (external URLs and HTML files).
Boundary markers: Absent; the scripts extract and concatenate text directly into output strings.
Capability inventory: File writing (--output flag), network requests (requests.get), and system command execution (via the associated setup scripts).
Sanitization: No sanitization or escaping of the extracted text is performed before it is formatted for output (JSON/Markdown/Text).

document-parsers