Agent Skills: Managing API Cache

Managing API Cache

Overview

Implement intelligent API response caching using Redis, Memcached, or in-memory stores with cache key generation, TTL management, cache invalidation strategies, and HTTP cache headers. Support read-through, write-through, and cache-aside patterns with tag-based invalidation for related resources and stale-while-revalidate behavior.

Prerequisites

• Redis 6+ or Memcached for distributed caching (required for multi-instance deployments)
• Cache client library: ioredis (Node.js), redis-py (Python), or Lettuce (Java)
• CDN with cache control support for edge caching (CloudFront, Cloudflare, Fastly) -- optional
• Monitoring for cache hit/miss ratios and eviction rates
• Understanding of data freshness requirements per endpoint

Instructions

1. Analyze endpoint characteristics using Read and Grep to classify endpoints by cacheability: fully cacheable (static data), conditionally cacheable (user-specific data), and never cacheable (mutations, real-time data).
2. Implement cache key generation middleware that creates deterministic keys from method, path, query parameters (sorted), and relevant headers (Accept, Authorization hash for user-specific caches).
3. Build a cache-aside middleware that checks the cache before executing the handler, returning cached responses with X-Cache: HIT header, or executing the handler, caching the result, and returning with X-Cache: MISS.
4. Configure TTL per endpoint category: long TTL (1 hour) for reference data, medium TTL (5 minutes) for frequently changing lists, short TTL (30 seconds) for near-real-time data.
5. Implement cache invalidation on mutations: when a POST/PUT/DELETE modifies a resource, invalidate all cached responses containing that resource using tag-based invalidation.
6. Add Cache-Control, ETag, and Last-Modified response headers for HTTP-level caching, enabling CDN and browser cache participation.
7. Implement stale-while-revalidate behavior: serve stale cached responses immediately while asynchronously refreshing the cache in the background, reducing perceived latency.
8. Add cache warming for critical endpoints: pre-populate cache entries on application startup or schedule for frequently accessed resources.
9. Write tests verifying cache hits, misses, invalidation correctness, TTL expiration, and stale-while-revalidate behavior.

See ${CLAUDE_SKILL_DIR}/references/implementation.md for the full implementation guide.

Output

• ${CLAUDE_SKILL_DIR}/src/middleware/cache.js - Cache-aside middleware with hit/miss tracking
• ${CLAUDE_SKILL_DIR}/src/cache/key-generator.js - Deterministic cache key generation
• ${CLAUDE_SKILL_DIR}/src/cache/invalidator.js - Tag-based cache invalidation on mutations
• ${CLAUDE_SKILL_DIR}/src/cache/store.js - Redis/Memcached cache store abstraction
• ${CLAUDE_SKILL_DIR}/src/config/cache-policies.js - Per-endpoint TTL and caching policy configuration
• ${CLAUDE_SKILL_DIR}/tests/cache/ - Cache behavior verification tests

Error Handling

| Error | Cause | Solution | |-------|-------|----------| | Cache stampede | TTL expires simultaneously for popular key; many requests hit database | Use lock-based revalidation (only one request refreshes); apply jittered TTLs | | Stale data served | Cache invalidation missed a related resource after mutation | Implement tag-based invalidation covering all affected cache keys; add invalidation audit logging | | Redis connection failure | Cache store unavailable due to network or server issue | Fall through to database with degraded performance; log cache bypass; alert on sustained failures | | Cache key collision | Different requests generating identical cache keys | Include all varying parameters in key; hash the full normalized request for uniqueness | | Memory pressure | Cache grows unbounded consuming all available Redis memory | Configure Redis maxmemory-policy to allkeys-lru; set per-key size limits; monitor memory usage |

Refer to ${CLAUDE_SKILL_DIR}/references/errors.md for comprehensive error patterns.

Examples

Product catalog caching: Cache GET /products list for 5 minutes and GET /products/:id for 1 hour, invalidating both when any product is created, updated, or deleted via tag products.

User-specific dashboard: Cache dashboard data per user using cache:dashboard:{userId} keys with 30-second TTL, serving stale data during revalidation to keep perceived response time under 50ms.

CDN edge caching: Set Cache-Control: public, max-age=300, stale-while-revalidate=60 on public endpoints, enabling CloudFront to serve cached responses at the edge while revalidating asynchronously.

See ${CLAUDE_SKILL_DIR}/references/examples.md for additional examples.

Resources

• Redis caching patterns: https://redis.io/docs/manual/patterns/
• HTTP caching (MDN): https://developer.mozilla.org/en-US/docs/Web/HTTP/Caching
• Cache-aside vs. read-through vs. write-through patterns
• Stale-while-revalidate: RFC 5861