Security audit
Security checks for vulnerabilities and agentic risk
The skill is marked suspicious because it includes instructions to rephrase image requests and retry or switch providers after a safety block; the review also flagged an unpinned installer command.
Review the request-retry and provider-switching advice before installing, and consider whether you are comfortable with an unpinned installer. The scan was partial: it inspected the two reported Markdown components, but two local references could not be resolved, so evidence about those references is incomplete.
The installation command invokes an unpinned npm package through npx, allowing the executed installer to change without changes to this document. A compromised package release could execute code with the installing user's permissions; however, this is an installation-time supply-chain risk, not an unpinned MCP server as the finding describes.
The skill generates images from user-provided descriptions through varg and the listed image-generation services, but the README does not explain that those descriptions are transmitted for external processing. Listing the APIs and account requirements does not disclose which user data leaves the agent environment.
The section explicitly teaches preserving a blocked image request while changing moderation-triggering terminology, reinforced by instructions to retry blocks and switch providers. The adult-only, clothed, nonsexual scope reduces the immediate harm, but categorically treating refusals as false positives encourages an agent to circumvent safeguards without first checking why the request was rejected.
