The Agent Skills Directory

[Command Execution]: The skill uses node to execute a local version check script and system utilities like open, xdg-open, or powershell.exe to display the generated preview. These actions are standard for the skill's purpose and follow common command-line patterns for opening local files.
[Potential Indirect Prompt Injection Surface]: The skill processes content from external project files, which introduces a surface for potential indirect prompt injection where untrusted data could influence agent behavior or the generated output:
Ingestion points: Data is read from project files including .tsx screens, package.json, and memory-bank.md (SKILL.md).
Boundary markers: No specific boundary delimiters are defined for the external content during the HTML generation process.
Capability inventory: The skill is capable of writing the preview.html file and executing shell commands to open the browser (SKILL.md).
Sanitization: The instructions mitigate risk by directing the agent to create a static visual approximation only and explicitly forbidding the inclusion of functional click handlers, scripts, or native API calls in the generated HTML.

preview-screens