kage-bunshin

SUSPICIOUS: the skill’s core purpose is coherent, and its toolchain appears official, but it materially expands agent autonomy by spawning persistent peer Claude sessions that inherit full project capabilities and are launched with permission bypass enabled. The main risk is not malware or exfiltration, but disabled safeguards and high-impact autonomous operation.

No clear malware behavior is evident. The principal security issue is path traversal through unvalidated session_id/name values in registry reads, writes, and deletions. The permission-bypass option is a consequential feature of the session launcher and should be treated as an elevated-risk design choice.