chart-craft
Pass
Audited by Gen Agent Trust Hub on Apr 16, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands such as
mkdir -pto create temporary directories,cdfor navigation, andopento launch the generated HTML files in a browser for user preview. - [EXTERNAL_DOWNLOADS]: The skill uses
npx getdesign@latestto fetch thegetdesignpackage from the NPM registry during the style selection workflow. This introduces a runtime dependency on external code. - [REMOTE_CODE_EXECUTION]: The execution of
npx getdesigninvolves downloading and running code from a remote source (NPM) on the local environment to process brand design tokens. - [PROMPT_INJECTION]: The skill processes user-provided chart descriptions and interpolates them directly into HTML templates, creating a surface for indirect injection.
- Ingestion points: User-supplied descriptions and chart data in the 'Step 2: 收集内容' workflow within
SKILL.md. - Boundary markers: None identified to separate user data from the HTML/CSS structure.
- Capability inventory: Shell command execution (
npx,mkdir,open) and local file system write access. - Sanitization: There are no instructions for sanitizing or escaping user input before it is embedded into the generated HTML files.
Audit Metadata