Security audit
Security checks for vulnerabilities and agentic risk
ClawScan rates this skill suspicious because it reports unreviewed external sharing of reference images, filter-evasion guidance, and incomplete analysis of part of the board template.
Consider the suspicious verdict before installing. Review whether you are comfortable sharing prompts and reference images with hosted services, and avoid sending confidential or sensitive material unless you have confirmed that sharing is acceptable. The scanner also reports filter-evasion guidance and an unpinned installer. Analysis was partial, so the available evidence does not establish that the skill is safe.
Referenced artifact was not completely inspected
6. **Build the board** from `templates/board.html`: copy it, fill only the JSON block (frames, spec, refs,
Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Storyboard Board</title>
<!--
varg-mockups storyboard board. One self-contained page: no build step, no network needed.
HOW TO FILL IT (for Claude):
The skill turns user shot lists into prompts and generates stills through varg and its image models. Although the README names the services and generation costs, it does not warn that prompts derived from user material are transmitted externally, which matters for confidential scripts or creative briefs.
The installation command runs an unpinned npm package through npx, allowing the code executed during installation to change without review. This is a supply-chain exposure rather than an unversioned MCP server, as the finding's label suggests; there is no evidence that the referenced package is malicious.
The workflow directs the agent to generate frames from reference anchors without checking whether user-supplied images are approved for external processing. Approval of generation costs does not necessarily authorize disclosure of confidential product images, identifiable people, or private locations to hosted services.
The text explicitly connects safety-filter rejection with rephrasing prompts, without requiring the underlying scene to become safe. Although tension and slapstick can be legitimate alternatives, this blanket guidance could lead an agent to conceal the same rejected intent rather than respect the refusal.
These instructions explicitly move local reference-file bytes to a hosted upload endpoint without a sensitivity check or separate authorization for disclosure. This can expose private reference images or unreleased assets, even when the user has approved the cost of generating a board.
The board rules explicitly require prompts and sheets to be in English, regardless of the current user's language preference. A historical reviewer's request does not establish opt-in for subsequent users, and no technical justification for the restriction is provided.