← Back
LangChain releases LangSmith CLI and skills for AI agents, boosting coding performance to 92%
· releasefeatureapisdkplatform · blog.langchain.com ↗

LangSmith CLI Launch

LangChain has released a new LangSmith CLI designed specifically for agent-native workflows. The CLI provides developers and coding agents with building blocks to interact with the entire LangSmith ecosystem through the terminal, including fetching traces, curating datasets, and running experiments. This agent-first approach recognizes that future AI development loops will increasingly be driven by terminal-based agents iterating autonomously.

Skills: Intelligent Tool Loading

The release introduces Skills—curated instructions, scripts, and resources that dynamically enhance agent capabilities in specialized domains. Rather than flooding agents with all available tools (which historically degrades performance), skills are loaded on-demand through progressive disclosure. Skills are portable and shareable as markdown files and scripts, making them reusable across any coding agent that supports the skills framework.

Three Core LangSmith Skills

LangChain maintains three initial skills within the langsmith-skills repository:

  • Trace: Add tracing to existing code and query traces
  • Dataset: Build test sets of examples for evaluation
  • Evaluator: Assess agent performance against datasets

These represent the three pillars of LangSmith's AI engineering workflow, with plans to expand the skill set.

Measurable Performance Improvements

On internal benchmarks, Claude Code with LangSmith skills achieved a 92% pass rate on basic LangSmith tasks, up from 17% without skills—a 75-percentage-point improvement. This demonstrates the value of targeted, domain-specific tool guidance for AI agents.

Agent Development Loop

The skills enable a virtuous development cycle: agents can add tracing to their own code, generate execution traces for debugging, create test datasets from those traces, build evaluators, and iterate on agent architecture based on evaluation results. This self-improving loop is designed to accelerate agent development workflows.

Installation

Install the CLI via:

curl -sSL https://raw.githubusercontent.com/langchain-ai/langsmith-cli/main/scripts/install.sh | sh

Add skills locally or globally using:

npx skills add langchain-ai/langsmith-skills --skill '*' --yes

For Claude Code integration:

npx skills add langchain-ai/langsmith-skills --agent claude-code --skill '*' --yes --global