The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/package-loader.mjs script utilizes node:child_process.spawnSync to execute npm install and relaunch the current Node.js process. This allows the skill to execute shell commands to manage its own environment.
[EXTERNAL_DOWNLOADS]: The skill's bootstrapping logic in scripts/package-loader.mjs downloads external Node.js packages from the NPM registry at runtime to resolve missing dependencies such as @hyperframes/producer and sharp.
[REMOTE_CODE_EXECUTION]: The skill implements a custom package loader in scripts/package-loader.mjs that dynamically installs and then import()s external code. While the installation uses the --ignore-scripts flag, the practice of downloading and executing unverified code from a remote registry poses a security risk.
[DATA_EXFILTRATION]: Documentation in references/design-picker.md instructs the user to start a local HTTP server using python3 -m http.server. This exposes the project's root directory to the local network, potentially allowing unauthorized access to local files.
[INDIRECT_PROMPT_INJECTION]: The templates/design-picker.html file uses innerHTML to render user-controllable text placeholders (such as {{prompt_headline}} and {{prompt_sub}}) into a preview pane. This creates a Cross-Site Scripting (XSS) surface where a malicious user prompt could inject script tags that execute in the context of the local design picker server.

hyperframes