design-md
Warn
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill uses the
npx getdesign@latestcommand to fetch code from the NPM registry. The packagegetdesignis maintained by a third-party organization that is not on the trusted vendor list.\n- [REMOTE_CODE_EXECUTION]: The use ofnpxto execute an unverified package (getdesign) at its latest version involves downloading and running external code. This presents a risk of executing malicious content if the package or its registry entry is compromised.\n- [COMMAND_EXECUTION]: The instructions direct the agent to execute shell commands (specificallynpx) to add files to the local project environment.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingestDESIGN.mdfiles containing "AI agent prompts" for UI generation. This external content is read directly into the session without sanitization or boundary markers, allowing third-party instructions to influence the agent's behavior.\n - Ingestion points: The
DESIGN.mdfile generated in the project root by thenpxtool.\n - Boundary markers: None identified; the instructions direct the agent to read the file into the session context as plain text.\n
- Capability inventory: The skill utilizes shell command execution (
npx) and file reading capabilities.\n - Sanitization: There is no evidence of filtering or escaping the prompts contained within the fetched markdown files before they are processed by the agent.
Audit Metadata