What problem does it solve? Redis code fails silently in production: sessions lose their TTL on refresh, caches stampede the database when keys expire together, multi-key commands break on Cluster with CROSSSLOT errors, and queues lose messages when a worker dies. This Skill encodes the measured, version-gated rules for designing key spaces, caches, queues, search indexes, and cluster deployments that survive contact with production. ## Core Features & Use Cases - Structure and key-space design: Choose String, Hash, List, Set, Sorted Set, Stream, JSON, Bitmap, HyperLogLog, Bloom, or vector storage by access pattern, with measured listpack encoding thresholds and big-key splitting strategies. - Cache repair: Diagnose and fix synchronised-expiry avalanches, hot-key stampedes, cache penetration, and pattern-based invalidation using TTL jitter, single-flight locks, negative caching, Bloom filters, and versioned prefixes. - Cluster migration and operations: Inventory multi-key operations, apply entity-scoped hash tags, swap to cluster-aware clients, and verify with CLUSTER KEYSLOT, SLOWLOG, INFO, and --bigkeys. - Use Case: Given a worker.py that loses paid orders via RPOP and blocks the server with SMEMBERS on a 2M-member set, the Skill prescribes a Stream with consumer groups (XADD/XREADGROUP/XACK/XAUTOCLAIM) plus SSCAN or SINTERCARD, with a kill-a-consumer verification gate. ## Quick Start Review my Redis caching code and redis.conf, identify what is causing the OOM errors and the hourly database spike, and tell me exactly what to change.