The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted content from source documents (PDFs or pasted text), which represents an indirect prompt injection surface. Maliciously crafted content in these documents could attempt to influence the agent's behavior during the extraction process.\n
Ingestion points: PDF files and pasted text processed according to instructions in SKILL.md.\n
Boundary markers: Absent; the skill does not define specific delimiters for separating untrusted document content from agent instructions.\n
Capability inventory: The agent can execute local Python scripts (validate_refs.py), run image extraction tools (pdfimages, mutool), write files to the workspace, and make network requests to metadata providers.\n
Sanitization: The instructions focus on stripping personal data but do not include programmatic sanitization of the input text.\n- [COMMAND_EXECUTION]: The skill involves the execution of local scripts and system tools. Specifically, it calls python3 scripts/validate_refs.py to process markdown files and suggests using pdfimages or mutool for extracting images from PDF sources.\n- [EXTERNAL_DOWNLOADS]: The scripts/crossref.py and scripts/common.py files perform network requests to https://api.crossref.org to verify and enrich bibliographic data. This is an expected function using a well-known academic metadata service.

proposal-import