Browse reusable Agent Skills, each with a clear purpose and practical guidance.
Create browser automation agents that navigate websites, extract data, fill forms, and perform multi-step web tasks using natural language instructions. Use when user asks to automate browsing, scrape websites with AI, build a web agent, or perform complex browser interactions.
Configure AI model providers (OpenAI, Anthropic, Gemini, Ollama, HuggingFace, FAL, Replicate), set up API keys, choose models by task, run local inference with llama.cpp/MLX. Use when user asks about models, providers, API keys, which model to use, or configure any AI provider.
Set up RAG pipelines, vector indexing, document ingestion, vector search, and knowledge base creation in NodeTool. Use when user asks about RAG, document indexing, vector search, chat with documents, knowledge base, embeddings, or collection management.
Use the Agent class for multi-step tool loops, conversation history, and reusable LLM agents. Use when building agents with skills, structured output, or stateful conversations.
Pick the right LLM model for an Output SDK prompt file. Use when writing a new .prompt file, reviewing a model choice, or upgrading a stale model. Walks through priority (reasoning/balance/speed/cost), provider selection, and a live lookup against the Vercel AI Gateway model index.
Create .prompt files for LLM operations in Output SDK workflows. Use when designing prompts, configuring LLM providers, or using Liquid.js templating.
Create .md skill files for Output framework's lazy-loaded instruction system. Use when adding skills to prompts, configuring skill loading, or debugging skill resolution.
Bulk-upgrade the model field across .prompt files to the latest version of each prompt's existing family. Use when prompt models have drifted (eg sonnet-4 → sonnet-4-6), after a long pause between framework updates, or as part of a periodic model-freshness pass. Within-family only — never changes provider or tier.
Design diverse eval datasets using dimension-based variation. Use when bootstrapping eval datasets, when real traces are sparse, or when existing datasets miss edge cases.
Design effective LLM judge .prompt files for evaluators. Use when creating judgeVerdict/judgeScore/judgeLabel prompts, or when existing judges produce unreliable results.