qveris-cn

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

• Third-party content exposure detected (medium risk: 0.65). At runtime, the required discover/call workflow fetches tool candidates and execution results from the external QVeris API (https://qveris.cn/api/v1/...) and then prints/returns their JSON/text (including description, examples, and result/error_message) which can be treated as free-form outsider-authored content that may be ingested into the agent’s LLM context.

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

• Potentially malicious external URL detected (high risk: 0.90). The install instructions explicitly download remote executable skill files and then run them (curl/Invoke-WebRequest fetching https://qveris.cn/skill/scripts/qveris_tool.mjs and related .mjs files), which fetches remote code that would be executed locally.