website-screenshot-notte

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

• Third-party content exposure detected (high risk: 0.75). The workflow uses runtime curl calls to $GOOSEWORKS_API_BASE/v1/proxy/orthogonal/run with a user-supplied url (Step 2), which causes the browser/proxy to fetch arbitrary public web page content; that fetched page text can be ingested into the agent’s LLM context via the proxy’s page execution/screenshot pipeline (outsider public web content).

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill issues runtime curl requests to the Gooseworks proxy API (e.g., https://api.gooseworks.ai/v1/proxy/orthogonal/run) which executes remote browser sessions/commands on the provider side, so this external endpoint runs code required by the skill.