The Agent Skills Directory

[OBFUSCATION]: The template templates/full-decks/testing-safety-alert/index.html uses HTML entities (e.g., m for 'm') to encode potentially dangerous shell commands like rm -rf / and SQL commands like drop table. This is used defensively to display these commands as text examples in a security-themed presentation slide without triggering simple keyword filters or accidental execution.
[DYNAMIC_EXECUTION]: The assets/animations/fx-runtime.js script implements a dynamic plugin loader that identifies elements with data-fx attributes and injects corresponding script tags from the assets/animations/fx/ directory. While this involves runtime code loading, it is restricted to a predefined list of local animation modules.
[COMMAND_EXECUTION]: The scripts/render.sh script facilitates the conversion of HTML slides to PNG images by executing Google Chrome in headless mode. It uses the --no-sandbox flag, which is a common requirement for headless environments but expands the browser's attack surface if rendering untrusted remote content; however, the script is designed to operate on local file:// URLs generated by the skill.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to interpolate user-provided data into HTML templates. This creates a surface where malicious instructions could be embedded in the processed data. The skill lacks explicit sanitization of this content, relying on the agent's internal guardrails and the user's review of the generated HTML files.
[EXTERNAL_DOWNLOADS]: Several templates (e.g., chart-bar.html, code.html) reference external JavaScript libraries including Chart.js and highlight.js from the established cdn.jsdelivr.net CDN. These are well-known services used for their intended purposes of data visualization and code highlighting.

html-ppt