deepclaude-proxy

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

• Potentially malicious external URL detected (high risk: 0.90). The skill routes runtime model calls to external LLM endpoints (e.g., https://api.deepseek.com/v1 and https://openrouter.ai/api/v1) and uses Anthropic's bridge (wss://bridge.claudeusercontent.com), which are contacted during runtime and return model-generated outputs that directly determine the agent's prompts and are required for operation.

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

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The prompt explicitly instructs using sudo to create a symlink in /usr/local/bin (and repeats sudo use in troubleshooting), which requires elevated privileges and modifies system-level files/directories.