What problem does it solve? Production systems fail in ways that cannot be reproduced locally, and rushed fixes like restarts or fleet-wide debug logging destroy evidence or cause new outages. This Skill provides a disciplined workflow for root-causing live misbehavior — error spikes, latency, resource leaks — without making production worse. ## Core Features & Use Cases - Change-anchored investigation: Line up symptom onset against deploys, config changes, cron jobs, and data thresholds to find what changed at the failure's start time. - Failure-shape characterization: Slice errored requests by endpoint, region, instance, and customer so the partition itself names the suspect before any hypothesis is formed. - Metrics-logs-traces funnel: Follow one exemplar request through all three telemetry layers, read latency as distributions (p50 vs p99), and test one falsifiable hypothesis at a time. - Use Case: p99 latency on /search jumps 8x overnight with p50 flat. The workflow traces it to a nightly reindex leaving one hot Elasticsearch shard, confirms it via segment counts without touching anything, and ships a post-reindex alert as follow-up. ## Quick Start Use the production-debugging skill to diagnose why our API's p99 latency spiked after last night's deploy using our logs, metrics, and traces.
