autopilot

SUSPICIOUS. The skill’s core behavior is coherent with its stated purpose as an autonomous builder, and the visible secret-handling rules are protective rather than exfiltrative. However, it deliberately grants the agent broad end-to-end autonomy over planning and code changes, depends on unseen local helper tooling, and references a third-party skills CLI update path. No confirmed malware or credential theft is visible, but the execution scope is high enough to rate as medium-to-high security risk.