design-taste-frontend
Pass
Audited by Gen Agent Trust Hub on Nov 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data that could potentially influence agent behavior or script execution.
- Ingestion points: The skill reads a user-supplied 'brief' (SKILL.md, Step 4.1) and uses files provided by the user (SKILL.md, Section 2) to populate website content and parameters.
- Boundary markers: The instructions lack explicit markers or 'ignore embedded instructions' warnings for content parsed from the external brief.
- Capability inventory: The agent has the capability to execute multiple shell commands via Node.js scripts (consistency.mjs, shoot.mjs, etc.) and write to the local filesystem (index.html, names.json).
- Sanitization: No explicit sanitization or strict schema validation is described for the brief's input before it is used to generate the site and run checker scripts.
- [COMMAND_EXECUTION]: The skill's primary workflow involves executing a series of local CLI tools.
- Evidence: Multiple steps in SKILL.md (Steps 4 and 6) instruct the agent to run local Node.js scripts like
node <skill>/scripts/start.mjsandnode <skill>/assets/layout/check.mjs. While these are part of the skill package, they represent a significant capability that is driven by external input from the user brief.
Audit Metadata