image-to-code
Pass
Audited by Gen Agent Trust Hub on Jul 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- Local Script and Command Execution: The workflow involves executing local scripts and running applications within the local environment. Specifically, it references a 'preflight script' triggered when shell access is available and instructions to 'Run the local app.' While necessary for prototyping, this provides the agent with the ability to execute commands on the local system.
- Indirect Prompt Injection Surface: The skill is designed to ingest and interpret external data, including images, screenshots, Figma files, and codebase paths. If these sources contain embedded text instructions designed to manipulate the agent, they could influence the agent's behavior during the code generation process. The skill manages this by instructing the agent to focus only on necessary task-specific materials.
- File System Interaction: The agent is instructed to create and save files locally, such as 'design-qa.md' in the project root. This file-write capability is a core part of the automated QA workflow but involves standard interactions with the local storage.
- External Asset Generation: The skill utilizes an 'Image Gen' tool to create assets based on a reference design. This involves an automated chain where the agent evaluates visual targets and generates new image content to be used in the final code output.
Audit Metadata