The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill retrieves publication metadata from well-known and reputable research services including PubMed, Google Scholar, ResearchGate, bioRxiv, medRxiv, and Consensus. These operations are essential to the skill's purpose of identifying research gaps and do not involve sensitive user data.
[COMMAND_EXECUTION]: It uses local bash and Python scripts (e.g., pubmed_eutils.sh and parse_pubmed.py) to interface with academic APIs and parse the resulting structured data. These scripts are used for legitimate research tasks such as MeSH term clustering and publication counting.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external academic databases, which represents a potential surface for indirect prompt injection.
Ingestion points: Article titles, abstracts, and MeSH terms are retrieved from PubMed, Consensus, Scholar Gateway, and bioRxiv/medRxiv APIs.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present when processing the retrieved text.
Capability inventory: The skill is granted Bash, Write, and Edit permissions to manage its project directory and artifacts.
Sanitization: The skill does not explicitly sanitize or filter the academic text it processes, relying on the trusted nature of the research repositories.

ma-scout