← Back
LangChain
LangChain releases AI coding agent skills, improving Claude Code performance from 29% to 95%
· releasefeaturesdkopen-source · blog.langchain.com ↗

What Are Skills?

Skills are curated instructions, scripts, and resources designed to improve coding agent performance in specialized domains. Unlike traditional tool approaches that can degrade performance when agents have too many options, skills are dynamically loaded through progressive disclosure—agents only retrieve skills when relevant to the task at hand. This approach balances capability enhancement with focused decision-making.

New LangChain Skills Release

LangChain is releasing an initial set of 11 skills split across three categories:

  • LangChain Skills: Guidance on create_agent(), middleware, and tool patterns for classic tool-calling agent loops
  • LangGraph Skills: Instructions for working with LangGraph primitives, including native support for Human-in-the-Loop and durable execution
  • Deep Agents Skills: Documentation for the open-source Deep Agents package, covering prebuilt middleware and FileSystem capabilities

Skills are portable and shareable, consisting of markdown files and scripts that can be retrieved on demand and ported to any coding agent supporting skill functionality.

Performance Improvements

Testing with Claude Code (Sonnet 4.6) showed significant gains:

  • Without Skills: 25% pass rate on LangChain/LangGraph/Deep Agents tasks
  • With Skills: 95% pass rate on the same tasks

This 70 percentage point improvement was measured using LangSmith evaluations. The team plans to open-source the testing benchmark used for reproducibility.

Getting Started

Skills can be installed using the npx skills command-line tool:

  • Local installation: npx skills add langchain-ai/langchain-skills --skill '*' --yes
  • Global installation: npx skills add langchain-ai/langchain-skills --skill '*' --yes --global
  • Agent-specific linking: npx skills add langchain-ai/langchain-skills --agent claude-code --skill '*' --yes --global

The team plans to continue expanding skills content as new capabilities are added to both open-source offerings and LangSmith, and welcomes community feedback on additional skills and improvements.