The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill is designed to ingest and analyze untrusted CSS and HTML code provided by users for performance reviews.
Ingestion points: The skill reviews "stylesheets, component styles, and responsive behavior" as specified in the SKILL.md check and code review sections.
Boundary markers: There are no explicit instructions or delimiters defined to separate user-provided code from the agent's primary instructions, nor warnings to ignore embedded instructions in the analyzed data.
Capability inventory: The skill is primarily instructional and analytical; it does not request high-privilege tools such as arbitrary command execution or unrestricted network access in its configuration.
Sanitization: The instructions do not specify any filtering, escaping, or sanitization of the input CSS/HTML before the agent processes it.
[EXTERNAL_DOWNLOADS]: The skill references several well-known and reputable third-party tools and repositories including filamentgroup/loadCSS, addyosmani/critical, and purgecss. These references point to established open-source projects and industry-standard performance tools, presenting no security risk.
[SAFE]: The inclusion of a minified JavaScript snippet for the loadCSS polyfill in references/rule.md is consistent with standard web performance documentation practices. The script originates from a reputable source (Filament Group) and is used for its intended educational purpose.

css-non-blocking