This is a guarded, user-selected agent configuration capability that persists across launches, so it warrants a warning. The inspected source does not show an install-time trigger or covert credential exfiltration.
Static reason
High-risk behavior combination matched malicious policy.
Trigger
A user selects a supported model while an OpenRouter key is configured.
Impact
Claude prompts are routed to OpenRouter, and the local override can persist across launches. The inspected source does not establish covert credential theft or an install-triggered attack.
Mechanism
The CLI writes a loopback API URL and managed token into Claude’s local settings, then forwards requests to OpenRouter with the configured OpenRouter key.
Evidence
dist/index.js.claude/settings.local.json.claude/.cc-switcher.json
Network endpoints2
127.0.0.1:${t}openrouter.ai/api/v1/messages