Security audit
Security checks for vulnerabilities and agentic risk
The skill is a HashMob API reference, but it under-warns about account-changing and destructive actions that could affect user data or account value.
Review this skill before installing if you intend to provide a HashMob API key or let an agent run curl commands. Use it only with explicit approval for paid, destructive, withdrawal, purchase, submission, preference, notification, or hashlist-management actions, and verify HashMob's real authorization requirements independently before relying on endpoints marked no API key.
The documentation presents destructive DELETE endpoints for hashlists, including a hard-delete variant, as requiring no API key. If an agent trusts this documentation, it may invoke deletion operations without treating them as highly sensitive or requiring user approval, and if the backend actually permits such calls, this would enable unauthorized destruction of data.
The manifest description says the skill merely lets an agent 'interact with Hashmob.net', but the body documents authenticated endpoints that can spend balance, purchase items, withdraw Gold, submit data, and modify account state. This understatement can mislead users or orchestration systems into granting the skill broader trust than warranted, increasing the chance of unsafe autonomous use.
The skill includes destructive delete and hard-delete examples without adjacent warnings, approval requirements, or explicit guidance that these operations are dangerous. In an agent context, omission of such guardrails materially increases the risk of autonomous or accidental data deletion, especially because the same section also claims no API key is needed.
The 'Get user submissions' endpoint is described as requiring no API key even though it appears to expose user-associated submission data. This inconsistency can mislead agents about the sensitivity of the endpoint and, if the backend is similarly exposed, may allow unauthorized access to account-related information.
The notifications section labels all endpoints as unauthenticated, including an endpoint that marks unread notifications as read, which is an account-state-changing action. This contradicts the earlier security model and may cause agents to under-classify a state-changing request, leading to unintended modification of user account data or unsafe assumptions about endpoint access control.
No suspicious patterns detected.
