The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains instructions that explicitly encourage the agent to override its internal configuration in favor of user requests. In Section 1, the directive "ALWAYS listen to the user: adapt these values dynamically based on what they explicitly request in their chat prompts" creates a vulnerability where a user can manipulate the agent to ignore its defined baseline and design constraints.
[INDIRECT_PROMPT_INJECTION]: The skill defines a process for ingesting untrusted data (user chat prompts) to generate code but lacks necessary security controls to mitigate potential injection attacks.
Ingestion points: User-provided chat prompts as specified in Section 1.
Boundary markers: Absent; there are no instructions to use delimiters or ignore embedded instructions within user content.
Capability inventory: The agent is authorized to generate frontend components and suggest terminal commands such as npm install.
Sanitization: Absent; no instructions are provided to sanitize or validate user content before it is processed into generated code.
[EXTERNAL_DOWNLOADS]: The skill directs the agent to provide instructions for downloading external software packages. Section 2 mandates that the agent "MUST output the installation command (e.g. npm install package-name)" for libraries including framer-motion, lucide-react, zustand, @phosphor-icons/react, and @radix-ui/react-icons if they are missing from the project environment.

design-taste-frontend