The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external PDF files (paper.pdf) as its primary input. This creates an attack surface where malicious instructions could be embedded in the PDF to influence the agent's behavior during the conversion process.
Ingestion points: External PDF files are read and processed by tools like pymupdf and Unlimited-OCR in SKILL.md and references/pdf-type-detection.md.
Boundary markers: The workflow does not utilize explicit boundary markers or safety instructions to prevent the agent from obeying instructions that might be contained within the extracted text.
Capability inventory: The skill has access to Bash, Python (via uv run), and Node.js (via node) to execute tasks.
Sanitization: There is no explicit sanitization or filtering mechanism for the extracted content before it is processed or written to markdown files.
[DYNAMIC_EXECUTION]: The skill utilizes inline scripts and local utility files for complex processing tasks like PDF rendering and math validation.
Evidence: SKILL.md contains an inline Python script using the pymupdf library, executed via a shell heredoc (uv run --python 3.14 --with pymupdf python3 << 'EOF').
Evidence: The skill executes a local Node.js validation script (references/validate-math.mjs) to check LaTeX syntax and GFM structure.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for downloading content and extensions from well-known external sources.
Evidence: references/pdf-type-detection.md suggests downloading source files from ArXiv (https://arxiv.org/src/{arxiv-id}).
Evidence: SKILL.md recommends installing a GitHub CLI extension for local previewing (gh extension install thiagokokada/gh-gfm-preview).

academic-pdf-to-gfm