The Agent Skills Directory

[DYNAMIC_EXECUTION]: The documentation includes an example of dynamic code execution.
Evidence: In references/code-highlight.md, a 'runCode' function is demonstrated that uses eval(content) to execute string content captured from a code block. While this is intended to show how to build a 'code runner' feature, executing arbitrary code via eval() is a significant security risk if the input is not strictly controlled and sanitized.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to render and process content generated by AI models, which serves as an entry point for untrusted data.
Ingestion points: The content property in components like ElABubble, ElAMarkdown, and ElACodeHighlight (found in references/bubble.md, references/markdown.md, and others).
Boundary markers: The skill instructions do not explicitly mandate the use of delimiters or 'ignore' instructions for the data being rendered.
Capability inventory: The components support rendering complex formats including Markdown, KaTeX, and Mermaid图表. The documentation also provides a template for a client-side eval() runner.
Sanitization: While the underlying library may perform sanitization, the skill documentation does not provide specific instructions to the agent on how to handle potentially malicious instructions embedded in the AI's output.
[EXTERNAL_DOWNLOADS]: The skill references external packages and resources necessary for its operation.
Evidence: SKILL.md and installation.md recommend installing the element-ai-vue package via npm, pnpm, or yarn. It also mentions CDN usage via unpkg.com for rapid prototyping. These are standard developer practices for using third-party libraries.

element-ai-vue-helper