stitch-ui-shadcn-components
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external HTML content and screenshots retrieved from the Stitch MCP (referenced in
references/workflow.md) as well as source code from third-party component registries. Maliciously crafted content in these external sources could theoretically attempt to influence agent behavior. - Ingestion points: Retrieval of HTML, screenshots, and component code via MCP tools and external registries.
- Boundary markers: Instructions emphasize manual review and metadata inspection but do not specify explicit prompt delimiters to isolate untrusted data.
- Capability inventory: The agent performs local file writes to create components and executes CLI tools like
npx shadcn. - Sanitization: The skill requires replacing sensitive customer data (names, phone numbers) with demo values before processing, reducing data exposure risk.
- [COMMAND_EXECUTION]: The skill utilizes the
shadcnCLI (vianpx) for component installation and includes a local diagnostic scriptscripts/verify-setup.sh. While these are standard developer operations, they represent an execution capability that could be targeted if other layers are compromised.
Audit Metadata