Agent Skills for LangChain, LangGraph, LangSmith & Deep Agents

A comprehensive collection of agent-optimized skills for AI coding assistants working in the LangChain ecosystem. These skills cover the complete development lifecycle from project setup to production deployment, monitoring, debugging, and Deep Agents setup/configuration.

Each skill is a self-contained package with a SKILL.md entry point plus optional scripts, references, and templates.

Available Skills

Coverage

9 production-ready skills covering the complete LangChain/LangGraph/Deep Agents development lifecycle:

• 🚀 Project Setup - Initialize projects with proper structure and configuration
• 🤝 Multi-Agent Patterns - Supervisor, router, orchestrator, and handoff patterns
• 💾 State Management - Schemas, reducers, persistence, and checkpointing
• 🛡️ Error Handling - Retry policies, LLM recovery, and human-in-the-loop
• ✅ Testing & Evaluation - Unit/integration tests, trajectory evaluation, LangSmith evaluation, A/B testing
• 🔍 Trace Analysis - Debug with LangSmith traces and pattern detection
• 🌐 Production Deployment - Cloud, Hybrid, and Standalone deployment with monitoring
• 🧠 Deep Agents Setup - Initialize and configure Deep Agents with validated project scaffolding and compatibility guidance
• 📝 Deep Agents Planning - Master write_todos for effective task decomposition with patterns, best practices, and trace visualization

Plus the skill-creator meta-skill for extending this collection.

Quick Start

Optional (recommended for contributors): create and sync a local environment.

uv venv --python=3.12
uv sync

1. Pick a skill folder in skills/ and open its SKILL.md.
2. Use it directly in your assistant by pasting relevant sections or pointing the assistant to the file path.
3. For development tasks (creating, validating, packaging skills), follow the commands in AGENTS.md.

Example: initialize a new skill using the repo tooling.

uv run skills/skill-creator/scripts/init_skill.py  --path skills/

Install as Claude Code Plugin

1. Add the marketplace:

/plugin marketplace add soba-labs/langchain-agent-skills

2. Install a plugin bundle:

/plugin install langgraph-skills@soba-labs-langchain-agent-skills

Three bundles are available: langgraph-skills (project setup, agent patterns, state management, error handling), langsmith-skills (trace analyzer, deployment) and deepagents-skills (setup and configuration, planning and todos). Each bundle also ships skill-creator.

Or use the interactive menu:

/plugin menu

For local development:

claude --plugin-dir ./path/to/langchain-agent-skills

Once installed, Claude Code will automatically use these skills when relevant.

Use in Other Repositories

These skills can be shared by copying a skill folder (for example skills/langgraph-agent-patterns/) into another repository or a supported assistant skills directory.

OpenAI Codex

Codex reads skills from .agents/skills in a repo and ~/.agents/skills globally (Codex customization docs). The SKILL.md format is identical to Claude Code’s, so no conversion is needed.

git clone https://github.com/soba-labs/langchain-agent-skills.git

# For one project
mkdir -p .agents/skills && cp -r langchain-agent-skills/skills/* .agents/skills/

# Or for every project
mkdir -p ~/.agents/skills && cp -r langchain-agent-skills/skills/* ~/.agents/skills/

This repository also ships a Codex plugin manifest at .codex-plugin/plugin.json, so it can be installed as a Codex plugin from a local marketplace rather than copied by hand.

OpenCode

OpenCode searches the widest set of locations and will find these skills in any of them. Per project it reads .opencode/skills/, .claude/skills/ and .agents/skills/; globally it reads ~/.config/opencode/skills/, ~/.claude/skills/ and ~/.agents/skills/. Either placement above therefore works unchanged, or:

mkdir -p .opencode/skills && cp -r langchain-agent-skills/skills/* .opencode/skills/

Cursor

Option 1: Remote rule (GitHub)

• Cursor Settings → Rules → Add Rule → Remote Rule (GitHub)
• Use: https://github.com/soba-labs/langchain-agent-skills.git

Option 2: Local installation

# Project-level
git clone https://github.com/soba-labs/langchain-agent-skills.git .cursor/skills/agent-skills

# User-level
git clone https://github.com/soba-labs/langchain-agent-skills.git ~/.cursor/skills/agent-skills

Usage: type / in Agent chat to search and select skills by name.

Other Assistants

If your assistant does not support skills directly, point it at the skill file:

Read skills/langsmith-deployment/SKILL.md for production deployment guidance
Read skills/langgraph-agent-patterns/SKILL.md for multi-agent patterns

Harness Compatibility

The skills themselves are harness-agnostic: one SKILL.md with name and description frontmatter, optional scripts/, references/ and assets/. Claude Code, Codex and OpenCode all read that same format and all use progressive disclosure, so nothing in skills/ is specific to one assistant. Only discovery and distribution differ, and this repository ships all three:

• Claude Code reads .claude/skills/ (project) and ~/.claude/skills/ (global); distribution via .claude-plugin/marketplace.json.
• Codex reads .agents/skills/ (project) and ~/.agents/skills/ (global); distribution via .codex-plugin/plugin.json.
• OpenCode reads .opencode/skills/, .claude/skills/ and .agents/skills/, plus the global equivalents, so it is satisfied by either of the above.

The repo root carries .claude/skills and .agents/skills as symlinks to skills/, which means all three assistants discover the skills when this repository is itself the working project (contributors get them for free). Windows contributors may need git config core.symlinks true.

Repository Structure

• skills/ - 10 skill packages (9 production skills + skill-creator)
• .claude-plugin/marketplace.json - Marketplace manifest for Claude Code
• Optional *.skill exports - Generated distribution artifacts
• PLAN.md - Roadmap and implementation ordering notes
• AGENTS.md - Complete development guidelines and workflow

Contributing

Community contributions are welcome. We want to build great, comprehensive, and widely used LangChain agent skills together.

1. Open an issue for bugs, feature ideas, new skills, or improvements. Include context, expected behavior, and reproducible steps when possible.
2. Fork the repo and create a focused branch for your change.
3. Follow AGENTS.md while implementing:
   • Use uv run for Python scripts.
   • Keep skill instructions concise and move deep details to references/.
   • Validate skill structure with uv run skills/skill-creator/scripts/quick_validate.py skills//.
4. Open a PR and include:
   • Clear summary of what changed and why.
   • Reference to related issue or PLAN.md item (if applicable).
   • Validation notes (script checks, quick_validate.py output, manual verification).
5. Request review and respond to feedback quickly so we can merge safely.

Keep changes focused and actionable. AGENTS.md is the single source of truth for workflow, tooling, and architecture rules.