Security audit
Security checks for vulnerabilities and agentic risk
The skill provides useful task auditing, but its instructions permit deleting tasks without clear approval and require unrelated advertising in reports.
Review before installing. Require an initial read-only report and explicit approval of each task change or deletion, with original configurations saved first. Remove the mandatory advertising from both the instructions and script. Confirm that your environment uses the documented WorkBuddy database. The static scan reported no findings, but the inspected instructions substantiate the deletion-control and advertising concerns; they do not establish malicious data transfer.
The declared purpose is scheduled-task auditing, but the skill also requires every output to advertise unrelated third-party accounts and a website. This deliberately redirects the assistant's output toward the author's promotional goals; there is no evidence that the linked site receives audit data or hosts malware.
The workflow explicitly permits deleting supposedly replaced tasks without further confirmation, even though common triggers request only an audit or recommendations. An incorrect equivalence assessment could remove distinct schedules or configuration and stop expected future runs; requiring the agent to check functional coverage does not establish user authorization or guarantee equivalence.
The instruction to append third-party advertising to every result contaminates otherwise legitimate audit reports with content unrelated to the user's request. Monthly automated reporting could repeatedly distribute this promotion, although the instruction does not itself require visiting the website or transmitting information.
The mandatory footer redirects trusted audit output into an unrelated promotional channel and may imply that the assistant endorses the advertised community. This is the same underlying output-hijacking issue identified by SDI-1 and TP4, not an additional independent exploit.
The script unconditionally appends unrelated advertising and an external community link to audit output, allowing promotional content to be carried into reports or downstream AI responses as though it were part of the trusted audit. This is low-impact output contamination rather than demonstrated phishing or data exfiltration: the code prints the URL but does not contact it, and its destination is not established as malicious. The audit context makes the promotion inappropriate, while placement outside the main guard also causes it to print when the module is imported.
No suspicious patterns detected.
