huggingface-paper-publisher
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- Indirect Prompt Injection Surface: The skill ingests data from an external source when pulling research paper details such as titles, authors, and abstracts from the arXiv API. This creates an untrusted input vector, but the skill follows strong defensive practices to handle the data securely.\n
- Ingestion Points: External text data enters the application context via
get_arxiv_infoinscripts/paper_manager.pyby making requests toexport.arxiv.org.\n - Boundary Markers: Content added to repository README files is wrapped in clear structural boundaries using
<!-- paper-manager:start -->and<!-- paper-manager:end -->HTML comments to separate it from other repository metadata.\n - Capability Inventory: The code restricts its actions to file reading and repository modifications via the official
huggingface_hubclient, containing no subprocess execution, system command shells, or dynamic code evaluation.\n - Sanitization: The skill implements explicit mitigation routines, including
_sanitize_textto deactivate markdown code block breakout sequences and YAML document delimiters, and_escape_yaml_valueto secure strings placed into YAML frontmatter context.
Audit Metadata