brand-preview

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). In scripts/preview.mjs inspect, the skill reads free-text from the user’s local application source files (e.g., route component files and CSS) to build the harness (readOptional → inspect() token/selector classification), so the agent ingests untrusted outsider-authored text via the workspace files.