Claude Skills

Retrieval-Augmented Generation patterns for grounded LLM responses. Use when building RAG pipelines, constructing context from retrieved documents, adding citations, or implementing hybrid search.

LangGraph supervisor-worker pattern. Use when building central coordinator agents that route to specialized workers, implementing round-robin or priority-based agent dispatch.

Multi-PR development for large features. Stack dependent PRs, manage rebases, and get faster reviews on smaller changes. Use when creating stacked PRs.

SQLAlchemy 2.0 async patterns with AsyncSession, async_sessionmaker, and FastAPI integration. Use when implementing async database operations, connection pooling, or async ORM queries.

Radix UI unstyled accessible primitives for dialogs, popovers, dropdowns, and more. Use when building custom accessible components, understanding shadcn internals, or needing polymorphic composition.

Query decomposition for multi-concept retrieval. Use when handling complex queries spanning multiple topics, implementing multi-hop retrieval, or improving coverage for compound questions.

LangGraph human-in-the-loop patterns. Use when implementing approval workflows, manual review gates, user feedback integration, or interactive agent supervision.

Provider-native prompt caching for Claude and OpenAI. Use when optimizing LLM costs with cache breakpoints, caching system prompts, or reducing token costs for repeated prefixes.

Multi-agent coordination and synthesis patterns. Use when orchestrating multiple specialized agents, implementing fan-out/fan-in workflows, or synthesizing outputs from parallel agents.

Temporal.io workflow orchestration for durable, fault-tolerant distributed applications. Use when implementing long-running workflows, saga patterns, microservice orchestration, or systems requiring exactly-once execution guarantees.

LLM function calling and tool use patterns. Use when enabling LLMs to call external tools, defining tool schemas, implementing tool execution loops, or getting structured output from LLMs.

Use when assessing task complexity, before starting complex tasks, or when stuck after multiple attempts. Provides quality-gates scoring (1-5) and escalation workflows.

Property-based testing with Hypothesis for discovering edge cases automatically. Use when testing invariants, finding boundary conditions, implementing stateful testing, or validating data transformations.

Comprehensive prompt engineering with Chain-of-Thought, few-shot learning, prompt versioning, and optimization. Use when designing prompts, improving accuracy, managing prompt lifecycle.

Security patterns for LLM integrations including prompt injection defense and hallucination prevention. Use when implementing context separation, validating LLM outputs, or protecting against prompt injection attacks.

LLM streaming response patterns. Use when implementing real-time token streaming, Server-Sent Events for AI responses, or streaming with tool calls.

Testing patterns for LLM-based applications. Use when testing AI/ML integrations, mocking LLM responses, testing async timeouts, or validating structured outputs from LLMs.

Comprehensive prompt engineering with Chain-of-Thought, few-shot learning, prompt versioning, and optimization. Use when designing prompts, improving accuracy, managing prompt lifecycle.

Real-time data streaming with SSE, WebSockets, and ReadableStream. Use when implementing streaming responses, real-time data updates, Server-Sent Events, WebSocket setup, live notifications, push updates, or chat server backends.