The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted metadata from external research databases, creating a risk of indirect prompt injection. \n
Ingestion points: Metadata is fetched from external APIs via scripts/search_google_scholar.py, scripts/search_pubmed.py, and scripts/extract_metadata.py. \n
Boundary markers: The skill lacks explicit separators or warnings to prevent the agent from following instructions embedded in paper titles or abstracts. \n
Capability inventory: The skill uses Bash, Read, Write, and Edit tools to run scripts and modify bibliography files. \n
Sanitization: The Python scripts perform structural formatting for BibTeX but do not sanitize text for potential LLM-targeted instructions. \n- [METADATA_POISONING]: The SKILL.md file refers to a script scripts/generate_schematic.py that is missing from the provided package. Furthermore, there is a mismatch between the provided author context and the skill-author field in the frontmatter.

citation-management