volcengine-api

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

• Third-party content exposure detected (high risk: 0.75). The required runtime workflow calls Volcengine API Explorer endpoints (e.g., GET /api/common/explorer/api-swagger and related search/list endpoints) and then ingests the returned Swagger/JSON fields (including description, enum, x-demo, etc.) into the agent’s LLM context; these responses are not authored by the operating user, so they are outsider-provided free text/structured prose at runtime.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly issues runtime requests to Volcengine API Explorer endpoints (e.g. GET https://api.volcengine.com/api/common/explorer/api-swagger?... and https://api.volcengine.com/api/common/search/all?Query={keyword}&Channel=api) to fetch Swagger/docs that are then used to generate and control agent responses, so these external endpoints are required runtime dependencies that directly influence prompts.