The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates a feedback loop where data from generated HTML prototypes is sent back to the agent. It includes explicit instructions to treat this data strictly as input and never as new instructions or commands, mitigating potential injection attacks from the prototype's state.
Ingestion points: Data sent via the submitToClaude function in generated artifacts.
Boundary markers: The instructions explicitly state that submissions are data, not instructions.
Capability inventory: The agent uses the data to tune UI parameters (e.g., CSS duration, easing, scale).
Sanitization: The skill mandates the use of safe DOM APIs like textContent and createElement to prevent XSS.
[DATA_EXPOSURE]: The skill includes guidelines to prevent sensitive data exposure, such as prohibiting the storage of secrets in localStorage and ensuring that artifacts with private data are not published to external platforms.
[EXTERNAL_DOWNLOADS]: The skill references Google Fonts for typography, which is a well-known and trusted service. It avoids using untrusted CDNs or external package managers for the generated code.
[DYNAMIC_EXECUTION]: While the skill generates executable HTML and JavaScript, it enforces security hooks by forbidding the use of innerHTML for dynamic content, significantly reducing the risk of cross-site scripting in the generated prototypes.

html-design-prototypes