cutlass-skill

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

• Third-party content exposure detected (high risk: 0.80). The skill's required workflow instructs running update scripts (e.g., update-cutlass.sh / update-repos.sh) to clone and then read/search the public CUTLASS GitHub repository (https://github.com/NVIDIA/cutlass.git -> ~/.cursor/skills/.../cutlass/), so the agent will ingest and act on open third-party content that could contain instructions affecting its behavior.

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

• Potentially malicious external URL detected (high risk: 0.70). The update-cutlass.sh script clones/updates the CUTLASS repository from https://github.com/NVIDIA/cutlass.git at runtime and the skill requires and runs code from that repo (pip install -e python/CuTeDSL/, python examples, or compiled examples), so this remote git URL is a runtime external dependency that can lead to executing fetched code.