setup
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill scans untrusted local project files, which establishes a potential ingestion surface for malicious instructions that could influence the agent's synthesized output.
- Ingestion points: In Step 1, the agent is directed to scan the README, package.json, components.json, and other project documentation to discover tech stack and brand assets.
- Boundary markers: The instructions do not provide explicit delimiters or warnings to treat codebase content as untrusted data.
- Capability inventory: The skill allows writing the synthesized design context to .better-web-ui.md and AGENTS.md (Step 3).
- Sanitization: No instructions for sanitizing or escaping the content read from the codebase are included.
Audit Metadata