LangChain Agent Skills

Agent-optimized LangGraph, LangSmith and Deep Agents skills that let AI coding assistants cover the full build-to-production lifecycle.

Stars
★ 106
Last updated
10d ago
License
MIT
Primary language
Python

At a glance

How it runs
Agent plugin / skillCLI
Works with
Universal · cross-platformCodex · Claude Code
Cost
Free, no paid service needed
Setup effort
Low · running in minutes
You'll need
npx skills CLIuvPython 3.12Shell / CLINetwork accessLocal filesystem
Typical use
A team using Claude Code to build a LangGraph multi-agent app wants the assistant to follow one set of conventions when writing supervisor or handoff logic.
Not a fit if
  • Developers who want runnable LangGraph sample code, not assistant instructions
  • Teams that do not use coding assistants such as Claude Code, Codex, Cursor or OpenCode
Maintenance
Commits in the last 30 dayslast commit 2026-11-11

What does this agent do, and when should you use it?

This repository is a set of skill packages laid out in the SKILL.md convention, consumed by AI coding assistants such as Claude Code, Codex, Cursor and OpenCode when working in the LangChain ecosystem. It ships nine production-oriented skills — project setup, multi-agent patterns, state management, error handling, testing and evaluation, LangSmith trace analysis, production deployment, and Deep Agents configuration plus write_todos planning — alongside a skill-creator meta-skill for extending the collection. Each skill is a self-contained folder whose entry point is SKILL.md, optionally supported by scripts/, references/ and assets/, loaded through progressive disclosure. Distribution works three ways: the open skills CLI, a Claude Code marketplace with three plugin bundles, and manual copying into .agents/skills, .opencode/skills or .cursor/skills. The skills are harness-agnostic, so the same SKILL.md is read by several assistants without conversion.

Each skill exposes SKILL.md with name and description frontmatter so an assistant can decide when to load it, pushing deeper detail into references/ to save context. Through these files an assistant can: scaffold a project with structure, langgraph.json, env vars and dependencies via langgraph-project-setup; pick supervisor, router, orchestrator-worker or handoff coordination via langgraph-agent-patterns; design state schemas, reducers, persistence, checkpoint inspection and migration via langgraph-state-management; implement retry strategies, LLM-based recovery loops and human-in-the-loop escalation via langgraph-error-handling; write unit/integration tests, trajectory evaluation, LangSmith dataset evals and A/B comparisons via langgraph-testing-evaluation; fetch and organize traces via langsmith-trace-analyzer; deploy and monitor on Cloud, Hybrid or Standalone via langsmith-deployment; and initialize Deep Agents projects and decompose tasks with write_todos. Two Python scripts, init_skill.py and quick_validate.py, create and validate skill packages.

  1. A team using Claude Code to build a LangGraph multi-agent app wants the assistant to follow one set of conventions when writing supervisor or handoff logic.
  2. An engineer taking a LangGraph app to LangSmith Cloud or Hybrid deployment needs the assistant to supply deployment parameters and monitoring steps.
  3. A team adopting Deep Agents needs to initialize Python or JavaScript projects with middleware, backends, subagents and persistence configured.
  4. Someone debugging a production agent asks the assistant to pull and categorize LangSmith traces with langsmith-trace-analyzer.
  5. A platform group building an internal skill library uses skill-creator to draft and validate new skills.
  6. Cursor or OpenCode users who want the same skill set wired in via a remote rule or a copied skills directory.

Key setup facts

Pulled from the install and usage notes below. The project README has the current values.

Install / run
  • npx skills add soba-labs/langchain-agent-skills
  • npx skills add soba-labs/langchain-agent-skills --skill langgraph-agent-patterns
Runtime
Python 3.12
Install notes cover
Windows

How do you install or deploy this agent?

Install every skill for Claude Code, Codex, Cursor, OpenCode and other agents with the open skills CLI:

npx skills add soba-labs/langchain-agent-skills

Or install a single skill:

npx skills add soba-labs/langchain-agent-skills --skill langgraph-agent-patterns

Install as a Claude Code plugin:

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

Clone and copy into an assistant's skills directory (Codex shown):

git clone https://github.com/soba-labs/langchain-agent-skills.git
mkdir -p .agents/skills && cp -r langchain-agent-skills/skills/* .agents/skills/

To work on the repo or run its scripts, you need uv and Python 3.12:

uv venv --python=3.12
uv sync

How do you use this agent?

Pick a folder in skills/, open its SKILL.md, and either paste the relevant sections into your assistant or point it at the file path. Assistants without skill support can be directed explicitly:

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

Create and validate a new skill:

uv run skills/skill-creator/scripts/init_skill.py <skill-name> --path skills/
uv run skills/skill-creator/scripts/quick_validate.py skills/<skill-name>/

Cursor users can add a Remote Rule (GitHub) in Settings → Rules pointing at the repository, then type / in Agent chat to select skills by name. The repo root carries .claude/skills and .agents/skills as symlinks to skills/; Windows contributors may need git config core.symlinks true.

What are this agent's strengths and limitations?

Pros
  • Skills follow the SKILL.md convention and are harness-agnostic, so one copy serves Claude Code, Codex and OpenCode without format conversion.
  • Coverage spans the whole lifecycle — setup, state management, error handling, testing and evaluation, deployment monitoring, plus Deep Agents.
  • Three ready distribution paths: a one-command skills CLI install, three Claude Code marketplace bundles, and plain directory copying.
  • Bundled init_skill.py and quick_validate.py give a concrete toolchain for authoring and validating new skills.
  • MIT licensed and maintained by Soba Labs, described as LangChain Ambassadors; folders can be copied freely into other repos.
Limitations
  • The skills are instruction text for coding assistants; the repo contains no runnable LangGraph application, so output quality depends on the assistant and model.
  • Other than AGENTS.md there is no verification checklist confirming a skill was actually loaded; that must be checked per tool.
  • Some install paths depend on the external npx skills CLI and marketplace mechanics, which need workarounds in offline or air-gapped setups.
  • On Windows the repo-root symlinks may require git config core.symlinks true before skills are discovered.

How does this agent compare with similar options?

Key facts side by side with the most closely related agents.

Agent Source review Form / cost Stars Updated Language Full support on
LangChain Agent Skills This agent 42 · Major gaps Agent plugin / skillFree ★ 106 10d ago Python Codex · Claude Code
Self-Learning Skills 54 · Major gaps Agent plugin / skillFree + model costs ★ 925 2mo ago — Codex · Claude Code
Loop Library: Reusable AI Agent Workflows 75 · Good Agent plugin / skillFree ★ 3.2k 4mo ago JavaScript Codex · Claude Code
Waza 71 · Some gaps CLIFree ★ 7.6k 3d ago Python Codex · Claude Code

How does FollowAgents rate this agent?

FollowAgents source review · FARS-2.1
Major gaps
42/ 100 5-point scale 2.1 / 5
Trust 7/29
Reliability 5/14
Adaptability 12/18
Convention 9/18
Effectiveness 6/13
Verifiability 3/8
Why each dimension lost points
Trust7 / 29 · 1.2/5

least_privilege: skills are docs and scripts with no explicit permission declarations or least-privilege constraints; only a generic 'never commit secrets' rule in AGENTS.md — deducted. user_confirmation: deepagents-planning-todos suggests showing a plan and asking approval, but it is advisory workflow, not an enforced confirmation mechanism — deducted. data_flow_transparency: no description of what data skills read or send; only a note that LangSmith traces may be sensitive — deducted. sensitive_data_handling: AGENTS.md has rules against committing real secrets and requires sanitizing trace exports, but no concrete handling procedure — deducted. dependency_security: no dependency manifest, version pinning, or vulnerability notes — deducted. external_effects: skills scaffold projects and run scripts, but no side-effect boundary is documented — deducted. rollback: no rollback or recovery mechanism described — deducted. source_attribution: README and LICENSE name Soba Labs and MIT, but publisher identity is unverified by FollowAgents — partial, deducted.

Reliability5 / 14 · 1.8/5

self_consistency: README claims '9 production-ready skills' while AGENTS.md lists 10 directories including '1 planned skill' — minor inconsistency, deducted. dependency_availability: README gives uv, npx skills, and plugin install paths, but no version pinning or source verification — deducted. failure_messages: langgraph-error-handling provides error taxonomy and retry strategies, but failure messaging for the skill's own scripts is not shown — deducted.

Adaptability12 / 18 · 3.3/5

audience_and_scenarios: targets AI coding assistants in the LangChain/LangGraph/LangSmith/Deep Agents ecosystem, covering setup, patterns, state, errors, testing, deployment — adequate. capability_boundaries: deepagents-setup-configuration includes a 'choose the right abstraction' comparison table clarifying boundaries between Deep Agents, create_agent, and LangGraph — adequate. trigger_precision: each SKILL.md description enumerates trigger conditions — adequate. environment_fit: README details installation and discovery paths for Claude Code, Codex, OpenCode, and Cursor — adequate.

Convention9 / 18 · 2.5/5

information_architecture: README, AGENTS.md, CLAUDE.md, and skills directory are clearly structured, but AGENTS.md ends with a truncated 'Project Context' section — deducted. install_notes: README provides multiple installation methods — adequate. naming_stability: skill directories use consistent kebab-case naming. examples_and_faq: example directories are referenced, but no complete FAQ appears in the sample — deducted. known_limitations: only scattered notes on Windows symlinks and skipped static checks for dynamic config; no systematic limitations section — deducted. license: MIT LICENSE is explicit. versioning_changelog: marketplace.json carries version 2.0.0, but no CHANGELOG — deducted. maintenance_responsibility: README states Soba Labs maintains the project with contact info, but publisher is unverified — deducted.

Effectiveness6 / 13 · 2.3/5

output_usability: SKILL.md files provide decision tables, code snippets, and script commands usable directly — adequate. marginal_value: content is largely general LangChain best practice overlapping official docs, limiting incremental value — deducted. cost_benefit: many skills with progressive disclosure, but no quantified benefit evidence — deducted.

Verifiability3 / 8 · 1.9/5

claim_traceability: README links to blog and docs, but technical assertions inside skills (e.g., 'Verified' default middleware) lack traceable sources — deducted. cross_source_corroboration: README, AGENTS.md, and marketplace.json corroborate the skill list, but technical details have no external corroboration — deducted. fact_inference_separation: SKILL.md mixes facts and recommendations without clear separation — deducted.

Risks and how to mitigate them
  • Not found in source: dependency securityPin versions and run a dependency audit (npm audit, pip-audit) before installing; prefer running it in a container.
  • Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
  • Publisher identity is unverified by FollowAgents; verify provenance before installing.
  • Skills scaffold projects and run scripts; review in an isolated environment first.
  • No dependency version pinning or security audit information; check dependencies independently.
  • Minor inconsistency between README and AGENTS.md on the number of skills.
Evidence confidence: Low Reviewed Nov 15, 2026 Reviewed revision 0401f727e736
See the full review method →
Maintain this project? Add the FARS badge to your README

The badge shows the current review score and verdict and links back to this page. It refreshes about once an hour after a re-review.

FARS

[![FARS review](https://followagents.com/badge/langchain-agent-skills.svg)](https://followagents.com/en/agents/langchain-agent-skills)
<a href="https://followagents.com/en/agents/langchain-agent-skills"><img src="https://followagents.com/badge/langchain-agent-skills.svg" alt="FARS review"></a>

FAQ

Do these skills execute LangGraph code themselves?
No. A skill is an instruction package with SKILL.md and optional scripts/, references/ and assets/. Claude Code, Codex, Cursor or another assistant reads it and acts accordingly; execution happens in your own project and assistant runtime.
Is there a cost to use them?
The repository is MIT-licensed and free. The skills guide you toward LangChain, LangGraph, LangSmith and Deep Agents; LangSmith deployment and tracing are third-party services, and any charges depend on your account there.
Do I have to use Claude Code?
No. The repository documents Claude Code, Codex, OpenCode and Cursor, each with its own directory and install route, and the SKILL.md format is shared across them.
What is in each of the three plugin bundles?
langgraph-skills covers project setup, agent patterns, state management and error handling; langsmith-skills covers trace analysis and deployment; deepagents-skills covers setup and configuration plus planning and todos. Each bundle also ships skill-creator.
How do I add my own skill?
Run uv run skills/skill-creator/scripts/init_skill.py <skill-name> --path skills/ to generate the skeleton, write the SKILL.md following AGENTS.md, then validate with quick_validate.py.
Is LangChain Agent Skills open source?
Yes. The source code is on GitHub (soba-labs/langchain-agent-skills) under the MIT licence.
Is LangChain Agent Skills safe to use?
FollowAgents' FARS review (Nov 15, 2026) scored LangChain Agent Skills 7/29 on trust and safety. The review did not find these safety controls in the source: dependency security, rollback or recovery path. Before giving it real accounts or data: Pin versions and run a dependency audit (npm audit, pip-audit) before installing; prefer running it in a container.
Is LangChain Agent Skills still maintained?
Yes. The latest commit on GitHub is from 2026-11-11, within the last 30 days.
Who makes LangChain Agent Skills?
LangChain Agent Skills is published on GitHub by the account soba-labs. FollowAgents has not verified which person or company is behind that account.
What is LangChain Agent Skills's official website?
The project's GitHub repository lists https://sobalabs.ai/blog/langchain-agent-skills/ as its website. That link is set by the repository owner; the source code itself is on GitHub (soba-labs/langchain-agent-skills).
Was this page helpful?
View on GitHub ↗ Install ↓

Host ecosystems

Compare agents like this one

The same FARS review applied across the shortlist this agent qualifies for.

Related concepts