The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by processing data from external sources and user inputs to modify repository files.
Ingestion points: Data enters the system via the arxiv_id and citation arguments in scripts/paper_manager.py, as well as responses from the external arXiv API (export.arxiv.org).
Boundary markers: The script uses explicit HTML comments ( and ) to delimit generated content in repository README files, which helps mitigate but not eliminate injection risks.
Capability inventory: The skill utilizes the huggingface_hub library to perform repository writes, including direct commits and pull request creation.
Sanitization: The implementation includes _sanitize_text and _escape_yaml_value methods in scripts/paper_manager.py to neutralize markdown code fences and YAML document delimiters.
[COMMAND_EXECUTION]: The skill instructions frequently prompt the agent to execute the bundled scripts/paper_manager.py using uv run. This script performs network operations and local file system writes to generate research articles from templates.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to huggingface.co for repository management and export.arxiv.org to fetch research paper metadata. These are well-known services in the research and AI ecosystem and are used for their intended primary purposes.

hugging-face-paper-publisher