The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill makes legitimate requests to well-known scientific databases and APIs for citation retrieval and verification. These include PubMed (NCBI E-utilities), Semantic Scholar, bioRxiv/medRxiv, and CrossRef. All are trusted scientific resources.
[COMMAND_EXECUTION]: The skill uses bundled Bash and Python scripts (pubmed_eutils.sh, parse_pubmed.py, snowball.py, check_doi_record_match.py) to process API responses and manage bibliographic data. These scripts are deterministic, operate on local data, and perform scoped network requests to trusted domains.
[REMOTE_CODE_EXECUTION]: There is no evidence of remote script execution or arbitrary command execution from untrusted sources. The skill uses urllib for data fetching from fixed, trusted API endpoints.
[DATA_EXFILTRATION]: No evidence of sensitive local file access or exfiltration. The skill reads from the user's BibTeX library files and writes candidate citations to references/library.bib within the project workspace.
[PROMPT_INJECTION]: The skill instructions emphasize anti-hallucination and verification protocols. There are no attempts to override safety filters or bypass system instructions.
[DYNAMC_EXECUTION]: The scripts use standard Python libraries and subprocesses for intended functionality (e.g., calling the parser) without unsafe dynamic code generation or deserialization.

search-lit