Security audit
Security checks for vulnerabilities and agentic risk
The skill does what it says, but it can send task contents to discovered external agents without clear trust, privacy, or retry controls.
Review this skill carefully before installing. Use it only for tasks that are safe to disclose to discovered peers, avoid secrets or regulated data in task bodies, and prefer verified or allowlisted agent addresses for sensitive or side-effecting work.
The skill demonstrates submitting task content to a network-discovered external agent without any explicit privacy, trust, or data-handling warning. Users may paste sensitive prompts, data, or operational details into the task body, and the routing logic optimizes for score rather than trust boundaries, which can lead to unintended disclosure to third-party peers.
The fallback logic may submit the same task to multiple candidate agents, increasing the number of external parties that receive the task content. This materially expands confidentiality risk and can also create duplicate execution side effects if the task is not idempotent.
The workflow example automatically discovers a peer and submits a task to it based on capability and polo score, but provides no warning that a discovered peer may be untrusted or outside the user's security boundary. In context, this makes accidental exfiltration of sensitive prompts or data more likely because the example appears to endorse automated routing as normal usage.
No suspicious patterns detected.
