material-3-expressive
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- SAFE (SAFE): The skill files consist of static markdown documentation and examples. No hardcoded credentials or data exfiltration patterns were detected.
- EXTERNAL_DOWNLOADS (LOW): The maintenance section in SKILL.md directs users to install Playwright and Chromium to run the reference update script. These are trusted developer tools.
- COMMAND_EXECUTION (LOW): The skill includes a python script (
update_m3_expressive_refs.py) to automate data collection from design websites. This is a local administrative task. - Indirect Prompt Injection (LOW): The skill possesses a vulnerability surface for indirect prompt injection. 1. Ingestion points: External URLs (Material Design docs, GitHub) via the update script. 2. Boundary markers: Absent in generated reference files. 3. Capability inventory: Agent processes reference data to provide design decisions. 4. Sanitization: Not verifiable from provided documentation.
Audit Metadata