The Agent Skills Directory

[SAFE]: The skill does not exhibit any malicious patterns such as prompt injection, obfuscation, or unauthorized data exfiltration. All operations are consistent with its stated purpose of managing research papers.
[CREDENTIALS_UNSAFE]: The skill requires a HF_TOKEN for write access to the Hugging Face Hub. It correctly retrieves this from environment variables or a .env file, adhering to standard security practices for secret management. No hardcoded credentials were found.
[DATA_EXFILTRATION]: Network operations are restricted to huggingface.co and arxiv.org. These are legitimate domains for the skill's functionality. The skill does not send sensitive local data or credentials to unauthorized third-party domains.
[INDIRECT_PROMPT_INJECTION]: The skill processes paper metadata from the external arXiv API, which is an ingestion point for untrusted data. However, the scripts/paper_manager.py file implements robust sanitization routines, including _sanitize_text and _escape_yaml_value. These methods effectively neutralize potential injection attacks by escaping markdown code fences and YAML document delimiters before the data is written to README files or research templates.
[COMMAND_EXECUTION]: The provided commands are well-defined CLI operations for the paper manager script. There is no evidence of arbitrary command execution or unsafe interpolation of user-supplied strings into system shell calls.

huggingface-paper-publisher