The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to perform a global installation of an unverified npm package using the command npm i -g vplan@latest. This package does not originate from a trusted organization or follow the author's verified naming patterns.
[COMMAND_EXECUTION]: The skill makes extensive use of the vplan CLI to validate, render, and export files. It specifically invokes shell commands such as vplan check, vplan --static, and vplan export. It also starts a long-running local server for interactive reviews.
[REMOTE_CODE_EXECUTION]: The requirement to install and execute a global package from a public registry that is not on the trusted vendors list represents a supply chain attack surface. Furthermore, the tool's reliance on MDX (Markdown with JSX) introduces a dynamic execution risk, as MDX can execute embedded code within the rendering environment.
[DATA_EXFILTRATION]: The vplan tool reads local filesystem content and creates snapshots of plans keyed by file path. While no explicit remote exfiltration is identified in the instructions, the tool's behavior of processing local data and hosting a local server for browser-based interaction creates a data exposure surface for any local files it is directed to process.

visual-plan