design
Pass
Audited by Gen Agent Trust Hub on Nov 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [SAFE]: No malicious patterns or security vulnerabilities were detected in the analyzed files. The skill demonstrates a strong security posture by explicitly acknowledging and mitigating risks related to untrusted input.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted design files and user comments, which represents a potential attack surface. However, the instructions mandate explicit defensive measures such as fencing external data with nonces and directing the agent to ignore instructions embedded within processed content.
- Ingestion points: Project files including 'canvas.json', '.dc.html' artboard files, and 'comments.json'.
- Boundary markers: Present (instructions explicitly require the use of nonce-delimited untrusted-data markers and specific 'data only' warnings).
- Capability inventory: File system read/write/delete operations and execution of the skill's management script.
- Sanitization: Present (relying on prompt-level fencing and strict instructional scoping).
- [EXTERNAL_DOWNLOADS]: The skill allows fetching fonts from Google Fonts (fonts.googleapis.com), which is documented neutrally as a well-known and trusted external service.
- [COMMAND_EXECUTION]: The skill utilizes a local Node.js script ('scripts/cli.mjs') to manage the design environment, perform file checks, and handle asset uploads. These operations are constrained to the local project folder and standard development workflows like git hooks.
Audit Metadata