If you're running LangChain or LangGraph apps and need to debug why an agent took a weird path or build a dataset from real production traces, this gets you set up fast. It covers both sides: adding the @traceable decorator to your code (Python or TypeScript) and then querying that data via the langsmith CLI. The trace vs. run distinction matters here because traces give you the full execution tree while runs are just individual nodes, and you'll want traces first for most debugging work. The CLI filters are solid for finding slow traces, errors, or specific time windows. Good for turning observed behavior into datasets without manually crafting examples.
npx skills add https://github.com/langchain-ai/langsmith-skills --skill langsmith-trace