external-gitcode-ascend-vllm-ascend-deploy

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.80). The prompt explicitly instructs the agent to ask for the user's SSH password and to use it (e.g., via paramiko) for remote setup, which requires the LLM to receive and potentially include plaintext credentials in actions or generated code, creating a secret-exfiltration risk.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly requires searching and fetching external documentation at runtime (e.g. https://gitcode.com/org/vLLM_Ascend/, https://docs.vllm.ai/projects/vllm-ascend-cn/zh-cn/latest/tutorials/models/, https://modelers.cn/user/vLLM_Ascend) to extract image names, TP values and startup parameters which directly determine the agent's deployment commands and behavior.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 0.90). The skill instructs collecting SSH passwords and configuring passwordless SSH, killing or managing remote processes, loading Docker images, creating containers and system cron jobs—operations that modify remote/local system state, can require elevated privileges or alter security-sensitive files (e.g., SSH keys, cron), and thus pose a high risk of compromising the host.

MEDIUM W021: Hidden or invisible Unicode characters detected (potential obfuscation or prompt injection).

• Hidden Unicode characters detected (1 type(s) found)