The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external academic databases and web pages, which is then presented to the agent's output. A malicious paper entry with a crafted title or metadata could attempt to influence the agent's behavior.
Ingestion points: Fetches paper titles, DOI values, and publisher link text from pubmed.ncbi.nlm.nih.gov and eutils.ncbi.nlm.nih.gov inside SKILL.md step 2.
Boundary markers: There are no specific delimiters or instructions used to separate the external data from the agent's instructions in the final report.
Capability inventory: The skill has browser navigation capabilities via mcp__chrome-devtools__navigate_page and mcp__chrome-devtools__new_page.
Sanitization: No sanitization is applied to the title or link text retrieved from the APIs before it is rendered in the markdown report.
[EXTERNAL_DOWNLOADS]: The skill constructs URLs for sci-hub.ru and doi.org. These are well-known services in the academic community for resolving and accessing research papers, and their inclusion is consistent with the primary purpose of the skill.
[COMMAND_EXECUTION]: The skill uses evaluate_script to run JavaScript in the browser context. While the script logic is static, it interpolates user-provided PMID arguments directly into the script body, creating a potential command injection surface if the input is not strictly validated by the agent platform.

pm-fulltext