nemo-retriever

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

• Third-party content exposure detected (high risk: 0.85). At runtime, retriever ingest/retriever query ingest and return text extracted from outsider-authored PDFs and other files in the user-supplied folder (e.g., ./pdfs/...), which becomes readable text in the LLM context via the tool’s JSON hits (retriever query stdout → agent parses/uses hits[].text).

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

• Potentially malicious external URL detected (high risk: 0.90). The skill's install recipe can run at setup time and explicitly clones remote code ("git clone https://github.com/NVIDIA/NeMo-Retriever.git") and then pip-installs that checkout (and may pull binaries from "https://download.pytorch.org/whl/cu130"), which fetches and installs remote code during runtime, so this is a runtime external dependency that executes remote code.