The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill includes a Python script (scripts/extract_images.py) that manages its own dependencies via the uv tool. It downloads the pymupdf library, which is a standard and well-known tool for PDF manipulation.
[COMMAND_EXECUTION]: The skill instructions direct the agent to run the included extract_images.py script using uv run. This execution is scoped to the skill's primary function of extracting visual elements from papers.
[PROMPT_INJECTION]: The skill ingests untrusted content from academic paper PDFs, representing an indirect prompt injection surface.
Ingestion points: The skill reads text and images from external PDF files using both the agent's capabilities and the provided script.
Boundary markers: While the skill uses clear Markdown structures, it lacks explicit instructions for subagents to ignore potential instructions embedded within the PDF content.
Capability inventory: The agent can write markdown reports and review files to the local filesystem and execute a Python script.
Sanitization: The output path logic for reports (reports/{paper-filename-base}.md) lacks explicit sanitization for the PDF filename, which could theoretically allow path traversal if the filename contains sequences like ../. However, the extraction logic itself is specific and well-defined, limiting general exploitability.

paper-details