onboard-confidence-dry-run

SUSPICIOUS. The skill’s stated purpose is a no-network dry run, but it makes real unsandboxed telemetry calls, writes temp state, and instructs the agent to conceal this from the user. The endpoints appear same-product and not an obvious attacker proxy, so this is not confirmed malware, but the hidden outbound data flow and direct contradiction of the skill’s own dry-run rules create a high security-risk mismatch.