earth2studio-data-fetch

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

• Third-party content exposure detected (high risk: 0.70). SKILL.md requires fetching “live docs and lexicon” from multiple public GitHub and docs URLs, so at runtime the LLM may ingest outsider-authored free text from those web pages to confirm API signatures and variable support.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly instructs the agent at runtime to fetch live docs and lexicon files (e.g., https://github.com/NVIDIA/earth2studio/blob/main/earth2studio/lexicon/.py and https://nvidia.github.io/earth2studio/modules/datasources_analysis.html), and those fetched files are used to determine which variables/sources are supported and to generate the fetch script — i.e., remote content directly controls the agent's decisions/output.