sglang-sota-humanize-loop

SUSPICIOUS: the skill’s benchmarking and optimization purpose is coherent, but its footprint is broad. The main risks are autonomous code-changing behavior, execution of another skill’s setup script, and optional reliance on a separate public skill for kernel evidence. I found no direct credential theft or exfiltration, so this is not confirmed malware, but it is a medium-high risk automation skill.