PlugWiki

langchain-middleware Skill: Human Approval and Custom Middleware for LangChain Agents

Teaches your agent to add human approval steps and custom middleware to LangChain agents, in Python and TypeScript.

Installs
16.6K
on skills.sh
Stars
1.3K
on GitHub
Updated
Oct 2026
last repo commit

The skill covers pausing an agent before a risky tool call (such as sending an email) so a person can approve, edit or reject it, and resuming the run afterwards. It also shows custom middleware hooks for retries, logging and guards. Every example comes in Python and TypeScript.

Every approval flow needs a checkpointer and a thread id, which the skill repeats in its fixes. The description also mentions structured output, but the SKILL.md has no section on it. The skill bundles no scripts and needs the LangChain packages in your own project.

Good for

  • Developers building LangChain agents that must ask before risky actions
  • Teams adding retries, logging or guards around tool calls
  • TypeScript and Python users who want the same patterns in both

Install langchain-middleware

Claude Code

  1. 1

    Copy this prompt.

    Install the langchain-middleware skill in Claude Code. Run this command:

npx skills add https://github.com/langchain-ai/langchain-skills --skill langchain-middleware

Then check that it loaded and tell me how to use it.
  2. 2

    Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js.

From its listing on skills.sh. Source.

What's in the skill

When it loads

INVOKE THIS SKILL when you need human-in-the-loop approval, custom middleware, or structured output. Covers HumanInTheLoopMiddleware for human approval of dangerous tool calls, creating custom middleware with hooks, Command resume patterns, and structured output with Pydantic/Zod.

Sections in SKILL.md

  1. Human-in-the-Loop
  2. What You CAN Configure
  3. Custom Middleware Hooks
  4. What You CANNOT Configure

Key facts

Author
LangChain
License
MIT (plugin.json, for the repo's plugin)
Installs on skills.sh
16,559 (Nov 2, 2026)
GitHub stars
1,320 (Nov 2, 2026), whole repo
Files
1 file including SKILL.md
Last commit
Oct 29, 2026 (repo)
Status
The README warns the project is in early development and content may change
Languages
Python and TypeScript examples
Needs in your project
A checkpointer and a thread_id for every human-in-the-loop flow

What it includes

  • HumanInTheLoopMiddleware with a checkpointer

    Pauses before chosen tools and allows approve, edit or reject.

    Basic HITL setup

  • Detect an interrupt, then resume with a Command

    Shows the resume syntax for approve decisions.

    Running with interrupts

  • Human edits the call before it runs

    The edited action must include the name and args.

    Editing tool arguments

  • Human rejects and explains

    Feedback goes back to the agent.

    Rejecting with feedback

  • interrupt_on rules per tool

    Each tool gets its own allowed decisions.

    Different policies per tool

  • wrap_tool_call, wrap_model_call, before and after hooks

    Six decorator hooks in two patterns: wrap hooks and before/after hooks. Warns not to use yield in wrap hooks.

    Custom middleware hooks

  • Missing checkpointer, missing thread id, wrong resume syntax

    WRONG versus CORRECT code for each.

    Fixes

From the official docs

Questions

What does the langchain-middleware skill do?

It teaches your agent LangChain's human-in-the-loop middleware and custom middleware hooks. That covers approving, editing or rejecting tool calls and resuming the run.

How do I install it?

Use the install box above. The folder is a single SKILL.md. The LangChain README also documents npx skills add langchain-ai/langchain-skills --skill '*' for every skill, and a Claude Code plugin (/plugin install langchain-skills@langchain-skills) whose plugin.json points at the repo's 22 skills and lists no hooks or MCP servers.

Does it need an API key?

The skill itself does not. The agents it helps you write need a model key, and the LangChain README says to set OPENAI_API_KEY or ANTHROPIC_API_KEY before running your agent.

Does it cover structured output?

The description says so, but the SKILL.md body has no structured output section. Use another source for Pydantic or Zod response formats.

Is it safe?

The folder bundles no scripts, only SKILL.md with Python and TypeScript examples. The code it helps you write can send emails or call other tools, which is why the skill adds an approval step; review that code before you run it. The README warns the project is in early development.

Why does my approval flow fail?

The skill lists the usual causes: no checkpointer, no thread_id, or the wrong resume syntax. Each has a corrected example.

Does it work in Codex or Cursor?

The LangChain README says the skills install through npx skills for any agent that supports the Agent Skills format, and names Claude Code, Cursor, Windsurf, Codex and GitHub Copilot.

Information

updated Nov 2, 2026

Repository
langchain-ai/langchain-skills
Folder
config/skills/langchain-middleware

langchain-ai/langchain-skillsskills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 2
  1. langchain-middleware SKILL.md · github.com
  2. langchain-skills README · github.com

Updated Nov 2, 2026.