citation-skills

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.90). SKILL.md explicitly fetches and ingests public, untrusted content (e.g., CrossRef via https://api.crossref.org, OpenAlex via https://api.openalex.org, arXiv export, Mendeley catalog, CNKI) as part of required citation-retrieval and workflow steps that the agent uses to generate/format/merge citations and drive follow-up actions, so third‑party content could materially influence behavior.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). The OneCite skill specifies running an MCP server via "npx @onecite/mcp-server" (and references https://github.com/HzaCode/OneCite), which causes the agent to fetch and execute remote code at runtime to expose tools that the agent would rely on—meeting the criteria for a runtime external dependency that executes remote code.