What problem does it solve? When an AgentBox sandbox will not start, a pool sits below its target size, or commands inside a sandbox fail, it is hard to know where to look first. This Skill provides a structured troubleshooting workflow that starts from Kubernetes events and pool status, then moves to sandbox logs and E2B-native metrics, so you find the root cause instead of guessing. ## Core Features & Use Cases - Event-first diagnosis: Inspect environment readiness, pool desired/idle/running counts, and Kubernetes events with abx envs and abx sandboxes commands. - E2B-native metrics: Read CPU, memory, and disk usage through the E2B SDK's get_metrics() and the /sandboxes/{id}/metrics endpoints, with correct unit interpretation (percentage of sandbox cores vs. console cores). - Failure pattern matching: A lookup table maps common symptoms (commands failing to connect, pools stuck at zero, empty lists) to their usual causes. - Use Case: A pool reports Ready but nothing is claimable. Run abx envs demo pools demo-1c2gi --json and check idleReplicas to see the real claimable count, then check events for the reservation or quota warning blocking growth. ## Quick Start Ask the assistant to investigate why your sandbox or pool is failing, for example: "My AgentBox pool is stuck at zero available sandboxes, find out why."
