The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The file scripts/research.py hardcodes the author's email address (eddyhu@gmail.com) within the BIB_PATH configuration, leaking personal identifiable information (PII).\n- [EXTERNAL_DOWNLOADS]: The script fetches academic metadata from the CrossRef API (api.crossref.org) to resolve journal names from DOIs.\n- [COMMAND_EXECUTION]: The skill uses the Python subprocess module to execute external binaries scholar and consensus to retrieve search results.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted academic paper titles and takeaways from multiple sources (Scholar, Consensus, and BibTeX files) without sanitization.\n
Ingestion points: Data enters via run_scholar, run_consensus, and search_bib in scripts/research.py.\n
Boundary markers: The skill lacks explicit markers or instructions to treat external paper metadata as untrusted content.\n
Capability inventory: The skill has command execution capabilities via subprocess.\n
Sanitization: No validation or escaping is performed on external strings before they are returned to the agent context.

research