The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to generate executable React components (themes/<id>.demo.tsx) that are automatically loaded and rendered by the application's dev-UI Themes panel. This script generation involves building functional components like Title, Footer, and Eyebrow using data extracted from external sources.
[PROMPT_INJECTION]: The skill faces a risk of indirect prompt injection because it processes untrusted data from multiple sources to generate the theme bundle. An attacker could provide a malicious image, description, or an existing slide containing hidden instructions or code snippets that the agent might inadvertently include in the final .tsx file.
Ingestion points: Data is ingested via the Read tool from user-supplied image paths/URLs and existing slide source code (slides/<id>/index.tsx).
Boundary markers: The skill lacks explicit instructions or delimiters to isolate untrusted data or warn the agent to ignore instructions embedded within those sources.
Capability inventory: The skill has the capability to write new .tsx files to the local filesystem under the themes/ directory.
Sanitization: There are no documented sanitization steps to validate that extracted design tokens or code snippets do not contain malicious payloads before they are written to the executable demo file.
[EXTERNAL_DOWNLOADS]: The skill accepts image references as URLs, which are then processed by the Read tool. While the download itself is handled by the platform tool, the skill's workflow encourages the processing of remote, untrusted content.

create-theme