The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The error recovery section in resources/execution-protocol.md instructs the agent to guide users to execute curl -LsSf https://astral.sh/uv/install.sh | sh. This is a classic remote code execution pattern that pipes a remote script directly into a shell, which is a high-risk practice.
[COMMAND_EXECUTION]: The skill makes extensive use of uvx to download and execute arbitrary Python packages from PyPI at runtime (opendataloader-pdf, mdformat, and opendataloader-pdf-hybrid). While uvx is a standard tool, its use to execute packages defined in the skill's instructions constitutes dynamic command execution.
[DATA_EXFILTRATION]: The skill is designed to process PDF files which often contain highly sensitive information (PII, financial data, legal documents). While a --sanitize flag exists for masking PII, it is optional and must be requested. The core purpose of extracting PDF content into Markdown for 'LLM context' or 'RAG' creates a significant risk of sensitive data exposure to the LLM or downstream systems.
[DYNAMIC_EXECUTION]: The skill implements a 'hybrid OCR server' architecture (opendataloader-pdf-hybrid) that starts a local network service on port 5002 to process document data. This involves downloading large ML stacks (torch, easyocr) and performing complex runtime processing of document images.
[PRIVILEGE_ESCALATION]: The instruction to install uv via shell piping often leads to users running the command with sudo if permissions fail, creating a path for privilege escalation through untrusted remote scripts.

oma-pdf