concept-to-image
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill takes user-supplied concepts to generate HTML/CSS, creating a potential surface where malicious instructions in a prompt could lead to the generation of scripts within the rendered artifact. * Ingestion points: User concepts and design instructions (SKILL.md). * Boundary markers: None used to delimit user data from template code. * Capability inventory: The
render_to_image.pyscript uses Playwright to render the generated HTML, providing a browser environment capable of executing JavaScript. * Sanitization: No explicit sanitization or Content Security Policy (CSP) is applied during rendering; the skill relies on the agent following instructions for self-contained, zero-dependency HTML. - [DYNAMIC_EXECUTION]: The skill dynamically generates HTML and CSS at runtime and executes it in a headless Chromium instance to produce visual output. This is the intended primary behavior for converting concepts to images.
- [COMMAND_EXECUTION]: The
scripts/render_to_image.pyscript executes shell commands to launch a Chromium browser instance. The script utilizes the--no-sandboxflag, which is standard for many headless or containerized environments but reduces the security boundaries typically provided by the browser sandbox.
Audit Metadata