dym-diagram-design-authoring-in-chat
Pass
Audited by Gen Agent Trust Hub on Oct 17, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input (natural language descriptions of diagrams) to generate HTML and SVG files.
- Ingestion points: User input provided via chat commands (e.g., "Make me an architecture diagram...").
- Boundary markers: The skill requires a mandatory confirmation step ("Confirm before drawing") before the agent proceeds with file generation.
- Capability inventory: The agent generates HTML/SVG files and can utilize local shell commands to copy templates (
cp). - Sanitization: The skill documentation references a linter (
lint-skin.py) and a self-check script (self_check.py) that explicitly reject inline scripts, remote assets (excluding approved Google Fonts), and executable HTML attributes. - [COMMAND_EXECUTION]: The skill documentation provides standard shell commands for managing local templates and verifying output integrity.
- Evidence: Examples include
cpfor copying local HTML templates andpython3for running a local verification script (self_check.py). These operations are scoped to local assets provided within the skill directory. - [EXTERNAL_DOWNLOADS]: The skill mentions the use of approved Google Fonts for styling.
- Evidence: The documentation specifies that no remote assets are allowed except for pre-approved Google Fonts stylesheets. Google Fonts is a well-known service, and this reference is documented for styling purposes without security escalation.
Audit Metadata