LangChainLangSmith CLI and Skills Released
LangChain has released a new LangSmith CLI alongside a set of skills—curated instructions and scripts designed to enhance AI coding agents' capabilities within the LangSmith ecosystem. The CLI serves as an agent-native interface for performing core LangSmith operations directly from the terminal, including fetching traces, managing datasets, and running experiments.
What Are Skills?
Skills are portable, shareable packages consisting of markdown documentation and scripts that improve agent performance in specialized domains. They employ progressive disclosure—agents only load relevant skills when needed—preventing performance degradation that occurs when agents have access to too many tools simultaneously. These skills can be used across any coding agent that supports the skills framework.
The LangSmith Skill Set
LangChain is launching three initial skills covering the core LangSmith AI engineering workflow:
- trace: Add tracing to existing code and query execution traces
- dataset: Build curated datasets of examples for testing
- evaluator: Run evaluations against datasets to validate agent correctness
Dramatic Performance Improvements
On LangChain's evaluation benchmark, Claude (Sonnet 4.6) achieved a 92% pass rate with skills enabled, compared to just 17% without them—a 75 percentage point improvement. This demonstrates how targeted agent expertise can dramatically enhance performance on specialized tasks.
Installation and Usage
Install locally with npx skills add langchain-ai/langsmith-skills --skill '*' --yes, or globally for all projects. The skills enable a powerful development loop: agents can add tracing to code, generate and analyze traces for debugging, create test datasets from those traces, evaluate results, and iterate on agent architecture based on findings.
LangChain plans to expand the skills library as LangSmith gains new capabilities, and is also releasing parallel skills for interacting with LangChain's open source libraries.