The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's setup instructions require the agent to execute a shell command (scripts/search setup <api-key>) using a string provided directly by the user. This creates a risk of command injection if the user provides a string containing shell metacharacters (e.g., semicolons, backticks, or pipe symbols) and the underlying script or agent environment does not perform strict sanitization.
[INDIRECT_PROMPT_INJECTION]: The skill retrieves full-text article content from PubMed and provides it to the agent. Because this content originates from an external, third-party source, it constitutes an indirect prompt injection surface. Maliciously crafted content within a research paper could potentially attempt to override agent instructions or exfiltrate data from the current session.
Ingestion points: PubMed search results, specifically the content field returned by the semantic search API and displayed to the agent.
Boundary markers: The provided documentation does not specify the use of delimiters or 'ignore embedded instructions' warnings when the agent processes the retrieved text.
Capability inventory: The agent has the capability to execute local shell scripts (scripts/search) and perform network operations to the Valyu API.
Sanitization: There are no instructions for the agent to sanitize, filter, or validate the biomedical text before processing it for the user.

pubmed-search