Find agent skills by outcome

- Plans for failure modes including stampede

cold start

- [ ] All cache layers are documented with technology

configuration

- Configuration: Key settings (TTL

eviction

Use checkboxes and stable IDs (e.g.

`CACHE-ITEM-1.1`):

Before finalizing

verify:

Use checkboxes and stable IDs (e.g.

`CACHE-PLAN-1.1`):

In `TODO_caching-architect.md`

include:

Coding & Debugging

Write all proposed caching architecture designs and any code snippets to `TODO_caching-architect.md` only. Do not create any other files. If specific files should be created or edited

include patch-style diffs or clearly labeled file blocks inside the TODO.

- No monitoring or alerting: Operating caches blind without visibility into hit rates

latency

- Configure cache-control headers (`max-age`

`s-maxage`

- Implement consistent hashing client-side (e.g.

libketama) for predictable key distribution

### CDN (CloudFront

Cloudflare

### Memcached (Distributed

Multi-threaded)

- Monitor with `INFO` command metrics: `keyspace_hits`

`keyspace_misses`

- Configure `maxmemory-policy` per instance based on workload (allkeys-lfu for general caching

volatile-ttl for mixed workloads)

### Redis (Clusters

Sentinel

- Enable compression (LZ4

Snappy) for large values where CPU overhead is acceptable

- Choose binary serialization (Protocol Buffers

MessagePack) over JSON for reduced size and faster parsing

- Use hierarchical namespaced keys (e.g.

`app:user:123:profile`) for logical grouping and bulk invalidation

- [ ] Monitoring covers hit rates

latency

- [ ] Security controls (TLS

authentication

Creative

- Design load tests simulating cache stampede

hot key

- Create benchmark suite measuring cache hit rates

latency percentiles (p50

- Define alerting thresholds for cache health (hit rate drops

latency spikes