The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to execute shell commands for dependency installation (npm ci), sprite rendering (node scripts/render.cjs), and binary uploads via curl.exe or Invoke-WebRequest. These are essential for the skill's primary function.
[EXTERNAL_DOWNLOADS]: The rendering utility depends on the puppeteer Node.js package, which is fetched from the public registry during the initial setup. Puppeteer is a standard, well-known library for browser automation.
[DYNAMIC_EXECUTION]: The skill generates and executes SVG, Canvas, or HTML code within a headless Chromium instance. To mitigate the risks of executing potentially malicious code, the rendering script (scripts/render.cjs) includes several defensive layers:
Request Interception: Explicitly blocks all outgoing network requests from the rendering page, preventing data exfiltration or remote resource loading.
Sanitization: Implements a multi-stage sanitizer that strips dangerous tags (<script>, <foreignObject>, <iframe>), event handlers (on*), and non-data URLs.
Content Security Policy (CSP): Injects a strict CSP (default-src 'none') to further restrict the execution environment.
Sandbox Usage: Chromium is launched with the OS sandbox enabled by default, with an opt-out only via an environment variable for specific containerized environments.
[INDIRECT_PROMPT_INJECTION]: The skill processes code (SVG/HTML) that may be influenced by user prompts or external data. This creates a surface where instructions could be embedded to target the rendering environment or the agent. The skill addresses this through its sanitization and sandboxing logic.
[CREDENTIALS_SAFE]: The skill implements best practices for handling 'presigned URLs' used for asset uploads. It instructs the agent to treat these short-lived credentials as secrets and pass them via environment variables and stdin configurations (curl -K -) to ensure they do not appear in process logs or shell history.

msw-painter