Run LLMs in production with structured outputs, observability, evals, and stateful agent memory — the toolkit for teams shipping AI features at scale
8 skills · Works with Claude Code, Codex, Cursor & more
Build typed, testable LLM functions with BAML — a DSL that compiles prompts into typesafe clients for TypeScript, Python, and Ruby. Handles schema-aligned parsing, function definitions, retries, and playground testing.
claude install-skill community/baml-llm Force LLMs to produce valid JSON, regex matches, or grammar-constrained outputs with Outlines. Works with open-weight models via Transformers, vLLM, llama.cpp, and remote APIs that support structured outputs.
claude install-skill community/outlines-generation Build stateful agents with persistent memory using Letta (formerly MemGPT). Configures memory blocks, archival memory, tool execution, and self-editing memory loops that survive across sessions and process restarts.
claude install-skill community/letta-agents Add LLM observability with Helicone — one-line proxy integration for OpenAI, Anthropic, and OpenRouter. Sets up request logging, cost tracking, prompt versioning, cache, rate limiting, and user-level analytics.
claude install-skill community/helicone-llm-ops Monitor and debug LLM applications with LangFuse. Instruments traces, tracks token usage and costs, creates evaluation datasets, runs prompt experiments, and builds quality dashboards for production AI systems.
claude install-skill nicholasoxford/claude-code-skills/langfuse Test and evaluate LLM outputs with PromptFoo. Generates eval configs, custom assertion functions, red-team security tests, model comparison matrices, and CI/CD integration for prompt regression testing.
claude install-skill nicholasoxford/claude-code-skills/promptfoo Run Python on serverless GPUs with Modal — define containers, GPUs, schedules, and web endpoints in pure Python. Covers function decorators, volumes, secrets, async, batching, and cold-start tuning.
claude install-skill community/modal-serverless-ml Deploy high-throughput LLM inference with vLLM. Generates server configurations, model loading scripts, PagedAttention tuning, LoRA adapter serving, structured output schemas, and GPU memory optimization.
claude install-skill nicholasoxford/claude-code-skills/vllm Build AI-powered applications with RAG, multi-agent systems, prompt engineering, and MCP integrations
Modern warehouse-native data engineering — dbt for transformations, Dagster for orchestration, DuckDB and Polars for fast local analytics, plus pipeline and SQL fundamentals
End-to-end deployment, monitoring, and infrastructure skills for shipping and operating production systems