benchmark-sandbox

The skill is internally coherent for remote Vercel benchmark automation and mostly uses official tooling and endpoints, so it is not malware. However, it is a high-impact automation skill: it forwards real credentials into remote sandboxes, runs Claude with dangerously skipped permissions, ingests untrusted prompts, and can autonomously deploy public apps. Overall classification: SUSPICIOUS due to elevated operational risk and autonomous external actions, not because of clear malicious intent.