Model Failover Guard

Automatically monitors model health and switches between primary and fallback models to maintain stability and recover when possible.

This skill appears to do what it claims, but review and take these precautions before enabling it: 1) Backup your OpenClaw config (~/.openclaw/openclaw.json) so you can restore the primary model if needed. 2) Verify the 'openclaw' CLI is the expected binary on PATH (the script calls it to test models and restart the gateway). 3) Inspect config.json (copy of config.example.json) to confirm preferred/excluded providers and candidate models — ensure no unexpected provider IDs appear that could point to untrusted endpoints. 4) Run the script once in a non-production environment (python3 .../failover.py once) to observe behavior and logs (~/.openclaw/failover.log) before enabling a loop or systemd service. 5) If you have strict process-control policies, note the script restarts the gateway; plan for maintenance windows. Overall the code is coherent and limited to the stated purpose; use the normal operational precautions above.

Type: OpenClaw Skill Name: model-failover-guard Version: 1.0.0 The skill bundle provides an automatic model failover and failback guard for OpenClaw. The core Python script (`failover.py`) uses `subprocess.run` to execute `openclaw` commands with fixed arguments, preventing shell injection. It modifies the `~/.openclaw/openclaw.json` configuration and logs to `~/.openclaw/failover.log`, which is consistent with its stated purpose. While paths for configuration, state, and logs are configurable via `config.json`, this is a design feature for flexibility rather than an inherent vulnerability, as it requires prior compromise of the local `config.json`. The markdown files (`SKILL.md`, `README.md`) clearly describe the skill's functionality and do not contain any prompt injection attempts or instructions for malicious behavior. The '安全边界' section in `skills/model-failover-guard/SKILL.md` explicitly states benign actions.