agent-loop-browser

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

• Third-party content exposure detected (high risk: 0.75). The skill’s runtime workflow navigates to arbitrary URLs (e.g., PR/preview pages) and then takes a “text snapshot”/page source via chrome-devtools.take_snapshot (or browse goto), which can ingest outsider-authored free text from public web pages into the agent’s LLM context.