filesystem-agents

SUSPICIOUS: the skill's main behavior is coherent with a Vercel Academy course companion, and its remote content/data flows stay on official Vercel domains. However, the install instruction creates transitive trust and is inconsistent with Vercel Labs' documented Academy skill source, and the skill tells the agent to follow instructions embedded in fetched remote markdown. No clear credential theft or exfiltration is present, so this is not malicious, but the install-path inconsistency and remote-instruction trust make it moderately risky.

The code implements an intentional bash-execution tool for a sandbox and contains no clear malware or supply-chain attack behavior. Its main security risk is broad command execution: safety depends on strong sandbox isolation and trusted tool callers. Restricting commands and arguments, enforcing working-directory and network limits, and applying least-privilege sandbox permissions would reduce risk.