The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted external data from document sources, including HTML and docx files, which presents a vulnerability surface for indirect instructions.
Ingestion points: Processes HTML strings in SKILL.md and extracts content from docx files in references/gotchas.md.
Boundary markers: No delimiters or ignore-instructions are specified to protect the agent from embedded commands in the source documents.
Capability inventory: Subprocess calls for PDF previewing (pdftoppm), file reading (zipfile), and PDF rendering (weasyprint).
Sanitization: No explicit sanitization or escaping of the input document content is described.
[COMMAND_EXECUTION]: The workflow includes the execution of the pdftoppm utility via shell to generate images for visual inspection of the generated PDF.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for installing external software such as weasyprint and poppler-utils. Additionally, the base CSS file fetches assets from Google Fonts' official API.

pretty-pdf