Config Guardian

**Quick judgment:** Config Guardian is a focused, high-safety helper for updating `openclaw.json` via agents. It’s well-suited for any environment where config mistakes could break the gateway and where agents are allowed to modify configuration. **What it does well** - Wraps config changes in a single, atomic script (`atomic_apply.sh`) that: backs up, applies, validates, and auto-rolls back on failure. - Uses `openclaw doctor --non-interactive` for validation, significantly reducing the risk of persisting invalid configs. - Maintains timestamped backups under `~/.openclaw/config-guardian-backups/`, enabling manual restore when needed. - Includes guardrails so agents don’t restart/apply config blindly and don’t bypass validation. **Key risks / limitations** - Requires the helper scripts (`atomic_apply.sh`, `validate_config.sh`, `restore_config.sh`) and a shell-like environment; not suitable where such script execution is restricted. - Only as strong as the underlying `openclaw doctor` checks—misconfigurations that pass validation could still be applied. - Narrow scope: it focuses on safe single-value updates rather than complex, multi-key transactional changes or policy-based config management. **Recommended scenarios** - Use whenever an agent needs to modify `openclaw.json` (e.g., switching models, adjusting defaults, toggling features) and reliability is important. - Particularly valuable in shared or production-like environments, where an invalid config could disrupt multiple agents or users. - Good fit for cautious rollouts of new models/backends or config experiments, where instant rollback and backups are required.