The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various system utilities including npm, unzip, tar, and ldd, as well as the Marp CLI tool. These executions are part of the standard setup and rendering workflow to manage dependencies and generate slide outputs.
[EXTERNAL_DOWNLOADS]: The setup command fetches the chrome-headless-shell binary from Google's official storage domains and installs dependencies via the NPM registry. Both types of downloads are from well-known services and are verified against hardcoded SHA256 checksums to prevent supply chain attacks.
[DYNAMIC_EXECUTION]: The skill uses import() and createRequire() to load the Marp core engine from its local tool cache. Additionally, it utilizes puppeteer.evaluate() during the check phase to execute measurement scripts within the browser to verify accessibility standards like text contrast and font size.
[INDIRECT_PROMPT_INJECTION]: As the skill processes untrusted Markdown and CSS files from a deck folder, it represents an indirect injection surface. However, it implements comprehensive mitigations including a custom parser that limits raw HTML to a safe allowlist (<span> and <br>) and a path validation engine that prevents the deck from referencing any files outside its own directory (path traversal and symlink escape protection).

rendering-slides