randroid

SUSPICIOUS: the skill’s capabilities mostly fit its stated developer-workflow purpose, but it combines autonomous code-change behavior, public GitHub posting, and execution of a third-party local-built CLI from non-registry paths. Data flows are mostly local/official, so this is not confirmed malware, but the execution trust and autonomy scope are broader than ideal.

This is a small hook configuration that delegates all meaningful behavior to an external stop-hook.sh executed on “Stop” events. The description’s claim of intercepting session exit and re-feeding prompts to create a self-sustaining loop is atypical and could indicate non-terminating or repeatedly reinforcing behavior. No direct malicious actions are visible in this fragment, but the command execution sink makes the package’s risk primarily dependent on the contents and trustworthiness of stop-hook.sh, which is not provided here; it should be reviewed end-to-end for data handling, persistence, and any unintended looping behavior.