browser-probe

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

• Third-party content exposure detected (high risk: 1.00). The skill's workflow (e.g., SKILL.md and references/stage-1-model.md) explicitly instructs the agent to open arbitrary TARGET_URL / $BASE_URL via agent-browser open "$TARGET_URL" and run eval/snapshot/click/fill on the live page DOM, meaning it fetches and interprets untrusted public third-party web content as part of its decision and action flow.