stitch-ui-uview2-components
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads design assets including HTML and screenshots from Google Cloud Storage URLs retrieved via the Stitch MCP. These fetches are managed by a local script to handle potential network reliability issues.
- [COMMAND_EXECUTION]: The skill utilizes a local bash script (
scripts/fetch-stitch.sh) to perform its download operations. The agent is instructed to invoke this script via a shell command, which is a significant capability requiring careful handling of arguments. - [INDIRECT_PROMPT_INJECTION]: The skill ingests and parses external HTML data from a design tool to produce Vue components, creating a potential surface for indirect prompt injection.
- Ingestion points: Design metadata and HTML content fetched from the
stitch-mcp-get-screentool (SKILL.md). - Boundary markers: The instructions do not define specific delimiters or security warnings to isolate the external HTML content from the agent's instructions.
- Capability inventory: The skill can execute shell scripts and write multiple file types (Vue, JavaScript, JSON) to the local workspace.
- Sanitization: There is no evidence of automated sanitization or instruction-filtering applied to the retrieved HTML before it is processed by the agent.
Audit Metadata