huggingface-paper-publisher

Pass

Audited by Gen Agent Trust Hub on Sep 9, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [SAFE]: No malicious behaviors or security vulnerabilities were identified. The skill's code and documentation align with its stated purpose of managing research papers.\n- [EXTERNAL_DOWNLOADS]: The skill interacts with recognized services, specifically fetching paper metadata from the official arXiv API and managing repository files on the Hugging Face Hub. These operations use official service endpoints and are appropriate for the skill's domain.\n- [CREDENTIALS_UNSAFE]: Secret management for Hugging Face Hub authentication follows secure practices. The instructions and scripts correctly guide users to use environment variables or local configuration files, avoiding the insecure hardcoding of API tokens.\n- [INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface that is managed with appropriate security controls:\n
  • Ingestion points: Paper metadata is retrieved from the arXiv API in scripts/paper_manager.py.\n
  • Boundary markers: The skill uses explicit HTML comments (<!-- paper-manager:start -->) to delimit content added to repository README files.\n
  • Capability inventory: The manager script uses upload_file for repository writes and requests for network operations.\n
  • Sanitization: Robust sanitization logic is implemented in the management script (_sanitize_text and _escape_yaml_value) to neutralize Markdown code-fence breakouts and YAML document delimiters in content sourced from external APIs.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 9, 2026, 03:52 AM
Security Audit — agent-trust-hub — huggingface-paper-publisher