Explore AI Agent Skills & Claude Prompts

Use this skill for any work involving Redis or Redis-compatible stores (Valkey, KeyDB, Dragonfly, Upstash, ElastiCache, MemoryDB, Azure Cache for Redis) — caching strategies (cache-aside, write-through, write-behind), TTLs, eviction policies (`allkeys-lru`, `volatile-ttl`), key design, data structures (strings, hashes, sets, sorted sets, streams, HyperLogLog, bitmaps, geo), pub/sub, Redis Streams, distributed locks, rate limiting (token bucket, sliding window), session stores, idempotency keys, leaderboards, queues, Lua scripts, pipelining, transactions (`MULTI`/`EXEC`/`WATCH`), persistence (RDB/AOF), replication, Sentinel, Cluster, sharding, hot keys, big keys, memory tuning, cache stampedes, and the licensing landscape (Redis 7.4+ source-available vs Valkey/KeyDB BSD forks).

llm-cost-optimization

Use this skill for reducing the token cost and latency of LLM-powered systems — model selection (Opus/Sonnet/Haiku, GPT-5/4o/4-mini, Gemini Pro/Flash, open-source), prompt caching (Anthropic, OpenAI, Gemini implementations), context compression, output length control, batch APIs, model routing/cascades, semantic caching, RAG vs fine-tuning trade-offs, structured outputs to cut retries, streaming, parallel tool calls, distillation, the "smaller model after a big model proves the prompt works" pattern, and reading token bills to understand where costs are coming from. Trigger when a user mentions LLM costs, "the bill is too high," budget caps, latency reduction, "this is too expensive to run," "we need to use a cheaper model," prompt caching, batching, or any time a system is being moved from prototype (where cost was acceptable) to production (where it isn't).

vanderbilt-data-classification

Apply Vanderbilt University's L1–L4 data classification scheme to any code, data flow, prompt, log, or vendor integration that touches institutional data. Trigger when the work involves student/HR/finance/research/health/donor/payment data, when deciding which generative-AI tool may receive a given dataset (ChatGPT Edu vs Amplify vs Microsoft Copilot vs none), when masking/redacting before an LLM call, when labeling files in OneDrive/SharePoint/Teams, when planning a vendor integration that exports data, or when reviewing a guardrail. Trigger even if the user does not say "L3" or "classification" — phrases like "send roster to OpenAI," "log this prompt," "export to a SaaS tool," "auto-label this file," or "prompt the model with the donor list" all need this skill.

amplify-platform

Integrate with Vanderbilt's Amplify GenAI platform (GitHub org `gaiin-platform`) — its serverless AWS backend, Next.js frontend, Terraform IaC, agent loop, MCP server registration model, Cognito auth, and the public `/chat`, `/files/*`, `/user-data/*`, `/assistants/*` endpoints. Use this skill when work touches Amplify, gaiin-platform repos (`amplify-genai-backend`, `amplify-genai-frontend`, `amplify-genai-iac`, `pycommon`, `amplify-mcp-servers-examples`, `fastify-mcp-server`, `claudia`/`opcode`, `open-notebook`, `skills` fork), the CCC social-media drafting pipeline integration, or any task that says "Amplify chat," "amplify-lambda-*," "amplify-agent-loop," "amplify MCP server," "register MCP in DynamoDB," "Cognito JWT for Amplify," or "Vanderbilt GenAI platform." Trigger on `amplify@vanderbilt.edu`, the maintainers (Jules White, Allen Karns, Karely Rodriguez, Max Moundas), or v0.9.0/v0.8.1 release references.

resilience-patterns

Use this skill any time the user is designing, implementing, or debugging behavior under failure — including timeouts, retries, backoff, jitter, circuit breakers, bulkheads, rate limiting, load shedding, idempotency, graceful degradation, fallbacks, queues and back-pressure, deadlines, the saga pattern, dead-letter queues, or chaos engineering. Trigger on phrases like "what if X is down", "should I retry", "this is timing out", "we got a thundering herd", "cascading failure", "exponential backoff", "circuit breaker", "is this idempotent", or when reviewing code that calls a remote system without obvious failure handling. Also trigger when the user asks "how do I make this reliable" — that's almost always a resilience-patterns question.

infrastructure-fundamentals

Use this skill for cloud-agnostic infrastructure and networking questions — DNS, TLS, load balancing, CDNs, reverse proxies, firewalls, VPNs, private connectivity, service meshes, certificate management, IP addressing, and HTTP-layer behavior. Trigger on anything involving traffic flow, latency, connectivity issues, "why is this slow," TLS errors, certificate renewal, choosing between L4 and L7 load balancers, designing public/private network topology, or understanding what sits between a user and a service. This skill complements aws and azure (which cover provider-specific implementations) by focusing on the underlying concepts.

cybersecurity-expert

In-depth security engineering companion that works macro-to-micro: defensive architecture and threat modeling first, then secure code review and concrete fixes. Use whenever security is in play — reviewing a system's security posture, threat modeling a feature, auditing code (especially AI-generated code, which ships ~2.7x more vulnerabilities) for injection / authn / authz / secrets / supply-chain flaws, hardening an API or deployment, or responding to a CVE. Trigger even when the user doesn't say "security" but the stakes are clear: "review this auth flow", "is this endpoint safe", "Copilot wrote this, can you check it", "we're storing passwords", "add a file upload", "harden our infra", "what could go wrong here". Also supports an explicit "update news" mode — on "update news", "refresh the threat landscape", or "pull the latest CVEs", it pulls current attacks and CVEs from a hardcoded whitelist of trusted sources and appends them to its own threat-landscape reference.