glm-review

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

• Potentially malicious external URL detected (high risk: 0.90). The skill sets ANTHROPIC_BASE_URL to "https://api.z.ai/api/anthropic" and passes it to the foreground claude -p process at runtime, so responses fetched from that remote model endpoint directly control agent behavior (the model is expected to call the Write tool to produce the review), making this a required runtime dependency.