hugging-face-paper-publisher

Pass

Audited by Gen Agent Trust Hub on Mar 10, 2026

Risk Level: SAFENO_CODECOMMAND_EXECUTIONCREDENTIALS_UNSAFEDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [NO_CODE]: The skill's functionality relies entirely on an external Python script located at scripts/paper_manager.py. This script is not included in the provided file list, making it impossible to verify its actual behavior or security.
  • [COMMAND_EXECUTION]: Usage instructions heavily involve executing shell commands via uv run scripts/paper_manager.py. These commands perform sensitive operations including repository metadata modification and authorship claims.
  • [CREDENTIALS_UNSAFE]: The skill requires the HF_TOKEN environment variable to be configured with write-access permissions. While necessary for the stated functionality, the exposure of high-privilege tokens to an unverified script is a security risk.
  • [DATA_EXFILTRATION]: The skill uses requests and huggingface_hub to interact with external APIs (arXiv and Hugging Face). Without the source code for the manager script, it cannot be confirmed that the sensitive HF_TOKEN or other local data is not being exfiltrated to unauthorized domains.
  • [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface (Category 8). It ingests untrusted data from external sources (arXiv metadata) and user-supplied files (citation.txt, abstract.txt) and interpolates this content into repository READMEs and YAML metadata.
  • Ingestion points: arXiv ID metadata fetching, --citation and --abstract command arguments.
  • Boundary markers: No boundary markers or 'ignore embedded instructions' warnings are present in the provided templates.
  • Capability inventory: The paper_manager.py script possesses file-write and network communication capabilities.
  • Sanitization: No sanitization or validation of the external content is documented or verifiable.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 10, 2026, 01:50 PM