The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains specific behavioral instructions to interrupt the agent's workflow and ask for user clarification when UI components are requested. These are standard task-specific guidelines and do not attempt to bypass safety filters.
[DATA_EXFILTRATION]: The skill establishes a submission pipeline where data from generated HTML artifacts is POSTed to a local ephemeral port at 127.0.0.1. This is intended for local session communication and is documented as the standard mechanism for the platform's HTML skills.
[EXTERNAL_DOWNLOADS]: The skill references Google Fonts as a permitted external resource for typography in the generated HTML files.
[COMMAND_EXECUTION]: The instructions require the invocation of the html-skills:html-skills-listen tool to initialize a local receiver for user submissions.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it ingests user design requirements to generate executable HTML/JS.
Ingestion points: User-provided UI descriptions and parameters in SKILL.md.
Boundary markers: None explicitly defined in the prompt interpolation.
Capability inventory: File system writes (creating .html files), local network operations (POST to localhost), and tool execution (html-skills-listen).
Sanitization: The skill includes explicit instructions to avoid innerHTML when handling variables or user input to prevent XSS vulnerabilities.

html-design-prototypes