The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted content from external PubMed articles.
Ingestion points: PubMed search results (titles, URLs, and full article content) are retrieved from the Valyu API and presented to the agent.
Boundary markers: No delimiters or explicit instructions to ignore embedded commands within search results were found.
Capability inventory: The skill has the capability to execute local bash and Node.js scripts (scripts/search).
Sanitization: There is no evidence of sanitization or filtering of the retrieved literature content before it is processed by the agent.
[COMMAND_EXECUTION]: The skill uses a filesystem search command to dynamically resolve paths to its own components.
Evidence: It executes find ~/.claude/plugins/cache -name "search" ... to identify the location of the search script within the agent's plugin cache.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to an external API to fetch biomedical research data.
Evidence: Connects to api.valyu.ai/v1 using Node.js fetch to retrieve PubMed content and metadata.

pubmed-search