autoresearch-ml

SUSPICIOUS. The skill’s ML-training purpose generally matches its local code-editing, benchmarking, git, and GPU capabilities, and there is no clear credential harvesting or third-party proxying. Risk comes mainly from the strong autonomy directives: indefinite looping, automatic resume, and repeated code execution/modification without per-run approval. Install trust is moderate rather than severe because commands use mainstream tools and no bundled downloader is present, but dependency contents/pinning are not visible. Overall this is a coherent but high-autonomy local experimentation skill, not confirmed malware.