The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill frequently instructs the user to run commands such as npx ai-elements@latest add [component-name]. This pattern involves downloading and executing a script from the npm registry at runtime. Since the ai-elements package is not part of the trusted vendors list and does not match the author's (sstobo) verified resource patterns, this represents a high-risk remote code execution vector from an unverified source.- [METADATA_POISONING]: Although authored by 'sstobo', the skill is titled 'vercel-ai-elements' and claims to provide documentation for 'Vercel AI Elements'. This use of official branding for a tool provided via an unverified npm package ('ai-elements') may mislead users into granting higher trust to the provided installation commands than is warranted.- [INDIRECT_PROMPT_INJECTION]: Several components, specifically Response and Artifact, are designed to render output generated by AI models. This introduces a surface for indirect prompt injection if the model generates malicious markdown or scripts intended to deceive the user or exploit the rendering environment.
Ingestion points: Markdown content processed by the Response component (detailed in references/vercel-ai-elements-content-display.md).
Boundary markers: The documentation does not specify the use of delimiters or 'ignore' instructions to prevent the agent from following instructions embedded in the generated data.
Capability inventory: The skill supports rendering markdown, code, and images, and provides interactive 'Artifact' containers.
Sanitization: While the skill mentions the use of streamdown for markdown rendering, there is no evidence of strict input validation or sanitization for the generated content.- [DYNAMIC_EXECUTION]: The WebPreview component allows for rendering web content and components within an iframe, which is a surface for dynamic execution of potentially untrusted code generated by an AI or loaded from a URL. Additionally, the instruction to add @source '../node_modules/streamdown/dist/index.js' to a global CSS file represents a configuration-time dependency on external code.- [COMMAND_EXECUTION]: The skill facilitates and encourages the execution of shell commands (npx) to manage the component library, which could be exploited if the agent executes these commands without sufficient user oversight.

vercel-ai-elements