langchain-ai/deepagents
7 skills11.3k installs169k starsGitHub
Install
npx skills add https://github.com/langchain-ai/deepagentsSkills in this repo
1Langgraph Docslanggraph-docs is a documentation-fetch agent skill for solo builders shipping AI agents with LangGraph. When you ask about graph agents, state machines, orchestration, or concrete implementation steps, the skill fetches LangChain’s llms.txt index, selects the two to four most relevant official pages, and pulls them with fetch_url before answering. That keeps responses aligned with current Python LangGraph docs instead of stale training cutoffs. Use it during backend and agent-tooling work when you need tutorials, conceptual grounding, or API reference detail. It is read-oriented—no local codegen scaffold—and assumes your environment can call fetch_url over the network. Best for builders already committed to LangGraph who want fast, cited guidance; less ideal when you need a full project generator or non-LangGraph stacks only.3.4kinstalls2Web Researchweb-research is a Deep Agents skill that turns vague “look this up online” requests into a disciplined research operation for solo builders and agent operators. It requires creating a dedicated folder, writing a research plan that splits the question into non-overlapping subtopics, then spawning subagents per subtopic with clear questions and instructions to persist findings to files before synthesis into a cited report. Use it when you need current information, option comparisons, or evidence-backed writeups—not for executing code changes inside your repo. The workflow mirrors how a one-person team should work: plan first, parallelize investigation, then merge sources with attribution so Idea and Validate decisions are defensible. It pairs naturally with agent stacks where the parent coordinates tools and child agents handle bounded web tasks, reducing hallucinated “facts” from a single overloaded pass.2.8kinstalls3Social MediaSocial-media is a Deep Agents workflow skill for solo builders and founders who need distribution-ready posts without skipping research or visuals. Every run starts by delegating research through the task tool to a researcher subagent, with findings persisted to research/[slug].md before any copy is written. The skill then enforces a concrete deliverable structure: LinkedIn content lives under linkedin/<slug>/post.md and X threads under tweets/<slug>/thread.md, each paired with a mandatory image.png so posts are not shipped as text-only drafts. It covers hooks, hashtags, thread structure, and repurposing longer content for social platforms. Most often used in Grow for content cadence, it also fits Launch when you are announcing a ship or validating interest in public channels. Intermediate complexity because it assumes a multi-step agent stack rather than a single prompt.1.7kinstalls4Arxiv Searcharxiv-search is a compact agent skill and script that searches the arXiv preprint repository and returns human-readable paper titles and summaries for a free-text query. Solo builders and indie researchers use it during early discovery when they need fast pointers to ML, systems, or domain papers without leaving the agent session. The implementation wraps the official arxiv Python client, sorts by relevance, and limits results with a max_papers parameter defaulting to ten. It is intentionally narrow: network search and formatting, not citation management, PDF download orchestration, or automated literature reviews. Install the arxiv package in the environment where the agent runs; otherwise the skill returns a clear install command. Pair it with note-taking or synthesis skills when you move from raw hits to a product or architecture decision.1.3kinstalls5Skill CreatorSkill Creator is LangChain DeepAgents’ meta skill for turning specialized knowledge into durable agent capabilities. It answers when to author a new SKILL.md, how to name and describe triggers so agents invoke the right procedure, and where files must live so the CLI and compatible tools resolve them predictably. The readme documents four load locations—from user-level deepagents aliases through project `.agents/skills/`—and explains override order when duplicate skill names collide. Solo builders use it whenever they outgrow one-off prompts: packaging deploy checklists, API conventions, or review rubrics into skills their coding agent can load on demand. It pairs naturally with validation and iteration loops after the first draft. This is procedural scaffolding, not a hosted marketplace or MCP server; you still commit files to git and test invocation in your agent of choice.720installs6Query Writingquery-writing is a LangChain deep agent skill that proceduralizes how coding agents turn natural-language questions into executed SQL. For simple requests it enforces a five-step loop: pick the table, pull schema, draft SELECT with WHERE ORDER BY LIMIT, run through `sql_db_query`, and format results for the user. Harder questions start with `write_todos` to map foreign keys, aggregation needs, and JOIN graphs before touching the database. It is aimed at solo builders wiring analytics into SaaS backends or debugging production data without opening a GUI client. The skill does not replace migrations or ORM modeling; it optimizes read-path correctness and keeps exploratory queries bounded. Operate-phase incident triage can reuse the same flow when logs point to data inconsistencies, but the primary home remains Build backend work alongside Deep Agents toolchains.673installs7Schema ExplorationSchema Exploration is an agent skill for solo builders and indie teams who must understand a SQL database before committing to queries, dashboards, or automated agents. It prescribes a repeatable ritual: enumerate every table, pull detailed schema including three sample rows per table, infer foreign-key style links from naming patterns, then summarize purpose and connectivity for the user’s question. It is aimed at Claude Code and similar agents already wired to LangChain Deep Agents sql_db_list_tables and sql_db_schema tools—not a standalone schema migration or ERD drawing product. Use it when someone asks what tables exist, what columns mean, or how entities relate, especially on legacy or sample databases you did not design. The outcome is a trustworthy mental model that reduces wrong joins and hallucinated column names in downstream SQL generation.673installs