alibabacloud-remote-skills-connector

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

• Third-party content exposure detected (medium risk: 0.65). High: the required runtime path fetches and processes outsider-authored free text from remote AgentExplorer-hosted discovery metadata and—more importantly—the remote Agent’s HTTP/SSE outputs (e.g., approval questions, task messages) which are surfaced as stdout and could be ingested into the agent/LLM context.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill documentation includes a runtime installation instruction that fetches and executes remote code via curl: /bin/bash -c "$(curl -fsSL https://aliyuncli.alicdn.com/setup.sh)", which is a high-confidence external command that would run remote code on the host.