LangChain Skills: build LangChain / LangGraph Agents from the official skills

Who it is for

If you want a coding Agent to build LangChain, LangGraph, or Deep Agents from the official skills—instead of mixing old interfaces from memory—this skill set is a better fit. It installs with npx skills and targets any Agent that supports the Agent Skills spec. The spec entry is

https://skills.sh/

Installer notes are at

https://github.com/vercel-labs/skills

The page names several clients and writes and more, so do not treat that as a closed roster. People who only want LangSmith trace or dataset flows should go to another repository, not install this set. The project is in early development; APIs and skill content may change. You need to accept that the interfaces will still move.

First try in 3 minutes

The official install is written on the page as this one:

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

--skill '*' means install every skill; --yes skips confirmation. This changes the current project's skill settings; it is not a read-only look. Confirm you have npx locally and that the current directory is the project you want to write to, then run it. Do not put another installer beside it.

If you will next use OpenAI or Anthropic models, first set OPENAI_API_KEY and/or ANTHROPIC_API_KEY in the shell. The page gave export examples with placeholders; those placeholder values are not repeated here. Do not paste keys into the chat.

After install, do a read-only check first: have the Agent explain how it will use these skills to build a minimal LangChain or LangGraph Agent, and do not write files yet. Success signal: it talks about these three pieces (LangChain, LangGraph, Deep Agents) and can point at installed skills, instead of starting a LangSmith dataset. If it wants to generate a project skeleton, have it name the paths it will create, and allow disk writes only after you confirm.

Usage examples

Example 1: Build an Agent inside this repository's skills. A task can be "draft a LangChain or LangGraph Agent from the installed skills". Deep Agents is named, but the available page text does not explain how it differs from the first two, and it gives no scaffold command. So what you can ask is: read the skills first, then propose; you cannot ask for an unwritten deep-agents CLI. Still confirm the file set before generating code. Early development means an example that runs today may not match in a few days.

Example 2: LangSmith traces and datasets. Officially, those workflows should use

https://github.com/langchain-ai/langsmith-skills

not this repository. If the Agent starts installing LangSmith-only skills or changing tracing settings in this project, stop and move to that repository. This page does not write the langsmith-skills install command, so here you only "spot the wrong repository"; you do not stand in for that other getting-started.

Limits

Early development is a hard limit: APIs and skill content may change. Do not treat the skill body from one install as a long-term stable contract. The install command is the npx skills add line above; do not put another installer beside it.

Keys cover only these two variable names: OpenAI and Anthropic. The page does not write env vars for other vendors, and it does not say whether the skills can still do pure structure design without keys.

The first stop is a read-only plan after install. Any write to the repo, any submitting of keys, or any stuffing of a LangSmith flow into this project must first have the Agent repeat the files it will change and why; you confirm, then it continues.