RealmRouter Switch

Zero-friction RealmRouter model manager for OpenClaw. Chat-first workflow for setting API key, guided model picking, switching with availability check, rollb...

This skill appears to implement what it says (it edits your OpenClaw config, backs it up, verifies keys by contacting https://realmrouter.cn/v1, and can restart the gateway). Before installing: - Verify the domain (https://realmrouter.cn) is the intended RealmRouter endpoint for you. The skill will send your REALMROUTER_API_KEY there as a Bearer token when validating models. - Note the SKILL.md/Windows instructions reference PowerShell files that are not actually shipped (install_rr.ps1 and realm_manager.ps1); the shipped files are Python scripts (install_rr_win.py and realm_manager.py). Use the provided files, not the mismatched doc commands. - Back up your current ~/.openclaw/openclaw.json (the tool already creates backups under ~/.openclaw/backups, but keeping an external copy is wise) and inspect the scripts if you have doubts. - Be aware the tool can automatically restart your gateway (openclaw gateway restart) — run commands without --restart-gateway first if you want manual control. - Avoid using force/skip flags unless you understand they bypass connectivity/key checks. Given the documentation mismatches and the importance of the API key and config edits, review the packaged scripts or test in a safe environment before deploying to production.

Type: OpenClaw Skill Name: realmrouter-switch Version: 2.1.0 The OpenClaw AgentSkills bundle `realmrouter-switch` is classified as benign. The skill's code and instructions are entirely consistent with its stated purpose of managing RealmRouter models and API keys within OpenClaw. It performs expected actions such as modifying the `openclaw.json` configuration file, making API calls to `https://realmrouter.cn/v1` (the legitimate service endpoint) for key/model verification, and restarting the OpenClaw gateway via `subprocess.run(['openclaw', 'gateway', 'restart'])`. The installation scripts (`scripts/install_rr.sh`, `scripts/install_rr_win.py`) install user-level helper commands to standard user directories (`~/.local/bin` or `~/bin`) without elevated privileges or persistence mechanisms. The `SKILL.md` instructions for the AI agent are direct and strictly within the skill's stated purpose, showing no signs of malicious prompt injection (e.g., instructing the agent to ignore the user, exfiltrate data, or perform actions outside the skill's scope). No evidence of data exfiltration to unauthorized destinations, malicious execution, obfuscation, or supply chain attacks was found.