LangSmith Releases CLI and Skills to Boost AI Coding Agents
New tools from LangChain give agents deep expertise in tracing, evaluation and LangSmith workflows, lifting Claude Code’s performance on internal benchmarks from 17% to 92%.
LangChain announced Thursday the release of the LangSmith CLI along with its first official set of “skills” designed to equip AI coding agents with specialized knowledge of the LangSmith observability and evaluation platform. The skills enable agents to add tracing, inspect execution traces, build test datasets and run evaluations — core tasks for developing reliable LLM applications.
The company said the new capabilities are available immediately through a skills installation system compatible with multiple agent frameworks, including Claude Code, Cursor, Windsurf, Goose and others supported by skills.sh. The GitHub repository langchain-ai/langsmith-skills provides the official skill definitions, which agents can install globally or for specific tools using a simple npx command.
Performance Leap on LangSmith Tasks
According to LangChain’s internal evaluation set, the skills dramatically improved Claude Code’s success rate on LangSmith-related coding tasks, increasing from 17% to 92%. The announcement highlights four primary capabilities now available to agents:
- Automatically adding LangSmith tracing to new agents
- Understanding and debugging agent execution flows
- Building and managing evaluation test sets
- Creating and running performance evaluations
These features address a common pain point for developers using AI coding assistants: the assistants often lack deep, up-to-date knowledge of specialized observability platforms like LangSmith, leading to incorrect or incomplete implementations.
How the Skills System Works
The skills are distributed as a package that agents can discover and utilize through the new LangSmith CLI and the skills.sh ecosystem. Once installed, an agent gains the ability to query traces, construct evaluation datasets, and implement custom evaluators directly from natural language instructions.
LangChain positioned the release as the first in what is expected to be a growing library of official skills covering the broader LangChain and LangSmith ecosystem. The company also published a YouTube demonstration showing Claude Code successfully navigating complex agent-development workflows using the new skills.
Industry Context and Competitive Landscape
LangSmith, LangChain’s observability and evaluation platform, has become a standard tool for teams building production LLM applications. The platform provides tracing, debugging, dataset management and automated evaluation capabilities that are critical for moving agents from prototype to production.
By packaging this domain expertise into a skill format consumable by frontier coding agents, LangChain is effectively turning its platform documentation and best practices into machine-readable knowledge. This approach mirrors broader industry efforts to give AI coding tools deeper context about specific frameworks and developer tooling.
Impact for Developers and Agents
For developers, the release means AI coding assistants can now provide more accurate guidance when integrating LangSmith into new projects. Tasks that previously required extensive manual prompting and correction can now be handled more autonomously by agents that “understand” the LangSmith ecosystem.
The skills system also creates a standardized way for the community and LangChain to distribute specialized knowledge. Additional skill packages for other parts of the LangChain stack are expected to follow.
What’s Next
LangChain has not announced a specific timeline for additional skill sets, but the company indicated this release represents the beginning of a broader effort to make its entire platform more accessible to AI agents. The skills are available now via the GitHub repository and can be installed using the documented npx commands.
The company encouraged developers to test the skills with their preferred coding agents and provide feedback on additional capabilities that should be added to future skill packages.
This article is based on LangChain’s official announcement and supporting documentation published March 2026.