suede-interface-explain
Pass
Audited by Gen Agent Trust Hub on Nov 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data including source code, HTML comments, and PR bodies. It proactively mitigates this risk through a dedicated 'Boundaries' section that explicitly instructs the AI to treat these inputs as task evidence only and ignore any embedded requests or instructions.
- [DATA_EXFILTRATION]: The instructions mandate a read-only approach and strictly forbid the exfiltration of secrets, credentials, or telemetry. It specifically bars the AI from submitting forms, signing in, or calling production actions during browser inspection.
- [REMOTE_CODE_EXECUTION]: The skill contains explicit prohibitions against running arbitrary repository scripts or pasted page code. It defines its environment as requiring no bundled executables or network dependencies for its core logic.
Audit Metadata