The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill retrieves and processes external biomedical literature from the PubMed database, creating a surface for potential injection attacks embedded in third-party content.
Ingestion points: Untrusted data enters the agent context via the NCBI E-utilities API responses (PMIDs, abstracts, and full-text metadata) as described in SKILL.md (Section 4) and references/api_reference.md.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to treat retrieved search results as untrusted or to ignore any instructions embedded within them.
Capability inventory: The skill relies on network operations via the requests library to fetch data and performs parsing of external XML, JSON, and text responses.
Sanitization: The documentation does not specify any sanitization, filtering, or validation steps for the content retrieved from PubMed before it is displayed or used by the agent.
[OBFUSCATION]: The skill uses string reversal to hide a support email address in multiple files.
Evidence: The string vog.hin.mln.ibcn@seitilitue in SKILL.md and references/api_reference.md is a reversed representation of the email utilities@ncbi.nlm.nih.gov.
[EXTERNAL_DOWNLOADS]: The skill initiates network requests to retrieve medical research data from a well-known government-hosted API.
Source: https://eutils.ncbi.nlm.nih.gov/entrez/eutils/
Context: These requests target the official National Center for Biotechnology Information (NCBI) infrastructure and are consistent with the documented purpose of biomedical literature retrieval.

pubmed-database