langchain-ai/langchain-skills
14 skills137k installs15.2k starsGitHub
Install
npx skills add https://github.com/langchain-ai/langchain-skillsSkills in this repo
1Deep Agents MemoryA skill for configuring memory and persistence in Deep Agents. Covers ephemeral (thread-scoped) and persistent (cross-thread) file storage, filesystem access, and hybrid routing. Developers use it when building agents that must retain state across sessions or coordinate file operations within a thread.13.8kinstalls2Langgraph PersistenceLangGraph Persistence manages durable graph state through checkpointers that save/load state at every super-step, thread IDs that isolate separate conversation sequences, and Stores for cross-thread long-term memory. Developers use it when building multi-turn agents that must remember conversation history, support time travel/replay from past checkpoints, or share user preferences across sessions. Key workflows include selecting appropriate checkpointers (InMemorySaver for testing, PostgresSaver for production), configuring subgraph persistence modes for interrupt and parallel execution, and using Stores to access user facts outside thread scope via the Runtime object.12.5kinstalls3Langgraph FundamentalsLangGraph is a framework for modeling agent workflows as directed graphs. Developers use it to build complex, stateful orchestrations by defining nodes (work functions), edges (routing logic), and state schemas with reducers for accumulation. Key workflows include linear pipelines, conditional branching, parallel fan-out with Send, and dynamic routing via Command. Graphs must be compiled before execution and support streaming (values, updates, messages, custom) and error handling (retry policies, tool error recovery, human interrupts). Essential for fine-grained control over multi-step agent systems that require persistence, loops, or human-in-the-loop checkpoints.12.5kinstalls4Langchain RagLangChain RAG skill covers the complete pipeline for indexing and retrieving documents to augment LLM generation. Developers use it to load documents from files, web, or databases; split content into manageable chunks with configurable overlap; convert text to embeddings; and store/search vectors in persistent stores (Chroma, FAISS, Pinecone). The workflow chains document loaders, RecursiveCharacterTextSplitter, embedding models (OpenAI), and vector stores to enable semantic search and context injection into LLM prompts, with support for metadata filtering and retrieval strategies like MMR.12.4kinstalls5Langchain FundamentalsThis skill covers creating LangChain agents using create_agent(), the recommended method for building production agents. Developers learn to define tools via @tool decorator or tool() function, configure agents with models and system prompts, and implement middleware for human-in-the-loop workflows and error handling. Key workflows include setting up checkpointers for conversation persistence across invocations, using middleware to intercept and control agent loops, configuring structured output for typed responses, and avoiding common pitfalls like missing tool descriptions, infinite loops, and incorrect result access patterns.12.3kinstalls6Langgraph Human In The LoopLangGraph's human-in-the-loop patterns enable pausing graph execution, surfacing data to users, and resuming based on their feedback. Developers use interrupt() to pause at a node and Command(resume=value) to resume with the human's input. Three requirements must be met: a checkpointer (InMemorySaver for dev, PostgresSaver for prod), a thread_id in the config, and JSON-serializable payloads. Common workflows include approval flows (pause, show draft, route to send or reject), validation loops (re-prompt until valid input), and multiple parallel interrupts (resume all at once via a map). Critical: all code before interrupt() re-executes on resume, so side effects must be idempotent (upsert, not insert).12kinstalls7Langchain MiddlewareLangchain-middleware provides production-grade middleware patterns for LangChain agents, enabling human-in-the-loop (HITL) approval workflows for sensitive tool calls, custom request/response hooks, and structured execution control via the Command resume pattern. Developers use it to pause agent execution before email sends, deletions, or other high-risk actions, then approve, edit arguments, or reject with feedback. Key workflows include configuring per-tool HITL policies with MemorySaver checkpointer and thread_id tracking, implementing wrap hooks for retry logic and guards, and before/after hooks for logging and state inspection. Requires explicit checkpointer and thread_id configuration for state persistence across interrupts.11.4kinstalls8Deep Agents OrchestrationDeep Agents Orchestration provides three middleware components for building agent workflows: SubAgentMiddleware for delegating tasks to specialized subagents, TodoListMiddleware for multi-step task planning and tracking, and HumanInTheLoopMiddleware for approval-based control over sensitive operations. Developers use this when building complex agent systems requiring task isolation, long-running operation planning, or compliance-mandated human oversight. Key workflows include configuring custom subagents with domain-specific tools, automatically creating todo lists for multi-step tasks, and setting up interrupts that pause execution until human approval via approve/reject/edit decisions. Requires checkpointers and thread IDs for state persistence across invocations.11.4kinstalls9Deep Agents CoreDeep Agents is an opinionated LangChain/LangGraph framework providing built-in middleware for task planning, context management, memory persistence, and human approval workflows. Developers use it to scaffold complex agent applications with TodoListMiddleware for task decomposition, FilesystemMiddleware for file-based context, SubAgentMiddleware for delegating work to specialized agents, and SkillsMiddleware for on-demand capability loading. Key workflows include configuring create_deep_agent() with tools, system prompts, subagents, storage backends, and interrupts; designing SKILL.md files for progressive disclosure; and maintaining conversation state via thread IDs and checkpointers.11.2kinstalls10Langchain DependenciesThis skill provides authoritative guidance on installing and versioning LangChain 1.0+ (current LTS), LangGraph, Deep Agents, LangSmith, and provider-specific packages. It covers the independently-versioned package structure, minimum runtime requirements (Python 3.10+, Node.js 20+), framework choice between LangGraph and Deep Agents, and pinning strategies for each package group. Developers invoke it when setting up new projects, troubleshooting import errors, or deciding which integrations to install. Key workflows include selecting the right orchestration layer, pinning langchain-community conservatively (NOT semver), migrating from legacy community imports to dedicated packages, and configuring environment variables for API keys.11.2kinstalls11Langgraph Clilanggraph-cli is an agent skill for the LangGraph CLI lifecycle from project scaffolding through local development, Docker validation, and LangGraph Platform deployment. Commands include langgraph new for template scaffolding, langgraph dev for hot-reload local servers without Docker on port 2024, langgraph build for Docker images, langgraph up for production-like Docker Compose stacks with Postgres on port 8123, and langgraph deploy for LangSmith Deployments requiring LANGSMITH_API_KEY. The langgraph.json reference documents required dependencies and graphs mappings plus optional env, python_version, dockerfile_lines, and related keys for Python and JavaScript projects. Templates include deep-agent-python, agent-python, and JavaScript project starters. Typical workflow scaffolds with new, configures graphs and env, develops with dev, validates with up --recreate, deploys to platform, and monitors via deploy logs. Gotchas note Apple Silicon Buildx requirements, dependency resolution from pyproject.toml or package.json, and that dev runs without Docker while system packages must exist locally. Developers reach for it when bootstrapping or shipping LangGraph agent graphs.4.3kinstalls12Swarmswarm is a LangChain agent skill for processing many independent items in parallel through a table handle workflow. create builds one row per file, glob match, or pre-parsed task record; run dispatches an instruction template with required responseSchema across rows and returns completed, failed, skipped, and failures counts. Omit subagentType for cheap direct model classification; set subagentType when rows need tools, file access, or multi-step reasoning. Sources include glob or filePaths for one-file-one-row work, or tasks arrays parsed inside eval from JSONL, CSV, or chunked readFile loops for large files. Aggregation uses rows with plain JavaScript filters and counts without spawning extra subagents. Chaining passes updates tables in place; filter supports equals, notEquals, in, exists, and and/or combinations for retries on failed rows. batchSize controls auto-batching capped at ten dispatches by default with optional per-row functions clamped between one and fifty. Technical notes require importing @/skills/swarm only in eval blocks that call it, cap console output around five kilobytes, and never write directly to .swarm directories.4.3kinstalls13Managed Deep Agentsmanaged-deep-agents is an agent skill for LangSmith hosted Deep Agents runtime covering deploy, thread, run, MCP, and file-tree operations. Prerequisites include Managed Deep Agents preview access, a LangSmith API key, and clients such as deepagents-cli 0.2.2 or managed-deepagents and @langchain/managed-deepagents packages defaulting to the /v1/deepagents API. The CLI scaffolds agent.json, AGENTS.md, tools.json, skills, and subagents directories, supports deploy dry-run, MCP server registration, OAuth connect flows, and paste-ready tools.json snippets from mcp-servers tools. Backends use state for lightweight agents or sandbox with thread or agent scope, policy_ids, idle_ttl_seconds, and delete_after_stop_seconds for code execution workloads. SDK workflows create agents and durable threads, stream events with values, updates, and messages-tuple modes, and expose a LangGraph client adapter for React useStream without exposing API keys in browser code. REST fallbacks document agents, threads, runs, MCP servers, MCP tools, and auth session resource groups. Developers reach for it when deploying hosted deep agents, wiring MCP interrupt_config approvals, or building React chat against.4.2kinstalls14Ecosystem PrimerThe ecosystem-primer skill "INVOKE FIRST for any LangChain / LangGraph / Deep Agents agent building project before consulting other skills or writing any agent code. Required starting point for up to date info on framework selection LangChain vs LangGraph vs Deep Agents vs hybrid composition , agent patterns, install, environment setup, and which skill to load next." It covers deep Agents top layer, harness - batteries-included toolkit built on LangChain + LangGraph. Ships with planning, file management, subagent spawning, and memory out of the box.. Key workflows include langGraph middle layer, runtime - low-level orchestration for durable execution, custom control flow, and stateful workflows. LangChain agents run on top of LangGraph.. Developers invoke ecosystem-primer when the task matches the triggers and reference files in SKILL.md for grounded, stepwise execution. Reference files and progressive disclosure keep context focused while preserving concrete commands, configuration fields, and validation checks copied from the upstream documentation.3.3kinstalls