The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to run npx getdesign@latest add <brand>, which downloads and executes a package from the npm registry. While the instructions require explicit user approval and forbid automatic execution, this pattern allows for the execution of remote code from an unverified third-party source.
[COMMAND_EXECUTION]: The agent is instructed to use shell commands like grep to scan the local repository for component names and class definitions to extract design information.
[EXTERNAL_DOWNLOADS]: The skill references multiple external GitHub repositories as sources for its design rules, benchmarks, and brand presets, including VoltAgent/awesome-design-md, pbakaus/impeccable, and Leonxlnx/taste-skill. These are not from verified or trusted organizations.
[PROMPT_INJECTION]: The skill's workflow involves processing untrusted data from user-provided screenshots and existing repository source code, creating a surface for indirect prompt injection.
Ingestion points: User-uploaded images (screenshots) and local source files (e.g., theme.ts, colors.ts).
Boundary markers: Absent. There are no instructions provided to the agent to differentiate between legitimate design information and malicious directives potentially embedded in the processed data.
Capability inventory: The agent has the ability to read local files, execute shell commands (grep), and perform remote code execution via npx.
Sanitization: Absent. The skill does not specify any validation or sanitization steps for data extracted from visual or code-based inputs before it is used to influence the agent's output.

design