What problem does it solve? When running LLM training on remote NPU servers, you need to confirm those servers can actually reach the local AutoResearch services (W&B, Prometheus, Grafana, Pushgateway) before launching experiments. This Skill automates that reachability probing so connectivity failures are caught before training starts. ## Core Features & Use Cases - Remote Endpoint Probing: Runs curl-based checks from remote servers against local wandb, Prometheus, and Grafana endpoints. - SSH Reverse Proxy Validation: Confirms that SSH reverse tunneling is correctly exposing local services to remote hosts. - Reachability Matrix Output: Produces a reachability table covering 4 endpoints across N servers for quick diagnosis. - Use Case: Before launching a verl training run on server A2-AK-225, run the reachability check to verify all monitoring and logging endpoints are reachable, avoiding silent metric loss mid-training. ## Quick Start Ask the AI to run the service reachability check against server nvidia-01 using the config file at ./config/config.yaml and show the reachability table.