The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on shell scripts (scripts/init-artifact.sh, scripts/bundle-artifact.sh) that execute complex command sequences to configure project environments and build artifacts. This includes using node -e to dynamically modify package.json and tsconfig.json files.
[EXTERNAL_DOWNLOADS]: The initialization and bundling scripts perform multiple downloads of external Node.js packages from the NPM registry and install the pnpm package manager.
[REMOTE_CODE_EXECUTION]: Automated execution of pnpm install and npm install involves running code from external package registries during the build and setup process. The skill also explicitly enables post-install scripts for specific native dependencies like @parcel/watcher and @swc/core by modifying the pnpm.onlyBuiltDependencies field in package.json.
[DATA_EXFILTRATION]: The skill provides instructions and scripts for uploading generated HTML artifacts to GitHub Gists and using the gistpreview.github.io service for rendering, which involves transmitting data to external third-party platforms.
[COMMAND_EXECUTION]: The init-artifact.sh script extracts content from a local tarball (shadcn-components.tar.gz) into the source directory. The content of this archive is not provided within the skill source, making it an opaque data injection point.
[PRIVILEGE_ESCALATION]: The setup script attempts to install pnpm globally (npm install -g pnpm), an action that typically requires elevated system permissions and modifies the host environment outside the immediate project scope.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it ingests 'source material' and 'notes' to generate HTML artifacts. Malicious instructions or scripts within the source data could be interpolated into the final HTML output and executed in a user's browser environment.

html-rich-communication