kernel-cute-writing

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

• Third-party content exposure detected (high risk: 0.90). Yes — Workflow 0 step 1 of SKILL.md explicitly instructs fetching example source files from the public GitHub raw URL (https://raw.githubusercontent.com/NVIDIA/cutlass/main/examples/python/CuTeDSL), so the agent is required to ingest and act on untrusted third-party content that can influence tool use and next actions.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly instructs fetching and using live example Python kernel code at runtime via web_fetch from https://raw.githubusercontent.com/NVIDIA/cutlass/main/examples/python/CuTeDSL (e.g., web_fetch https://raw.githubusercontent.com/NVIDIA/cutlass/main/examples/python/CuTeDSL/hopper/dense_gemm.py), which is a runtime fetch of executable code that the skill directs the user to run or adapt—constituting a high-confidence runtime external code dependency.