The Agent Skills Directory

[PROMPT_INJECTION]: The skill's code generation logic (in assets/main.js.template) is susceptible to indirect prompt injection (Cross-Site Scripting). The generated frontend code reads records from the external HAP database and renders them directly into the web page using innerHTML without performing any HTML escaping or sanitization.
Ingestion points: Business data (products, news, etc.) fetched from the HAP platform at runtime (identified in SKILL.md Phase 7 and main.js.template).
Boundary markers: Absent; the generated application does not use techniques to differentiate between data and HTML instructions during rendering.
Capability inventory: The generated code possesses full access to the browser DOM and performs network requests via fetch.
Sanitization: Absent; the template literals in main.js.template directly insert raw field values into HTML strings.
[COMMAND_EXECUTION]: In SKILL.md Phase 8, the skill instructs the agent to execute shell commands to start a local development server, including python -m http.server 8000, npx serve, and php -S localhost:8000. These commands run in the background to provide a preview of the generated project. Although these are legitimate development tools, they represent direct shell command execution initiated by the agent instructions.

hap-api-website