REPO PACKAGE
langchain-ai deepagents
COMMUNITY18skills
15ready to use
LABSCO SUMMARY
The 20 live skills split roughly by task: business analysis (analyze-market, competitor-analysis), long-form writing (blog-post, social-media), coding support (code-review, planning, skill-creator, coding-prefs), database work (schema-exploration, query-writing), general research (arxiv-search, web-research), a data-visualization skill, and a remember skill that writes durable notes back to a memory file. 17 of the 20 skills are labelled ready; the other three — arxiv-search, cudf-analytics, and cuml-machine-learning — are labelled local_tool, meaning each depends on something beyond the skill file itself. For the latter two that dependency is explicit in their own descriptions: NVIDIA's cuDF and cuML libraries for GPU-accelerated data analysis and machine learning.
This suits two audiences differently. If you're already running the deepagents Python package (uv add deepagents, LangChain's own agent harness built on LangGraph) — described in its README as an "opinionated," "batteries-included" agent that also supports sub-agents, a pluggable filesystem, persistent memory, and human-in-the-loop approval — these are a starter skill set to extend. If you just want individual skill files with no interest in the harness, most of the 20 stand alone fine; the GPU-flavored ones only pay off if you actually have that NVIDIA stack installed.
READ THE FULL ANALYSIS
Where these skills sit in the project. Deep Agents' README lists "Skills" as one of eight bundled features ("reusable behaviors the agent can load on demand"), but the text never names or walks through any of the 20 that ship in the repository — everything above comes from reading the skill files directly, not from the project's own documentation of them.
The repository actually holds more than 20 skills. It lists 22 skills in total; two are dead in our index and excluded from the count above, one of them an entry with no name and no description at all.
What we did not check. We read every skill's own frontmatter description; we did not run any of them, so "ready" here means the skill file looks self-contained on its face, not that we exercised it end to end.
ALSO IN THIS PACKAGE
MCP SERVERS
2 itemsMCP SERVER
docs-langchain
http · https://docs.langchain.com/mcpMCP SERVER
reference-langchain
http · https://reference.langchain.com/mcpWHAT'S INSIDE
18 showing · 18 totalReady to use
Nothing else to set up — install it and go.
18 of 18
SETUP
Showing all 18 skills
analyze-market
A structured way to size up a new market before committing to it, so the answer rests on stated numbers and reasoning rather than on a hunch.
arxiv-search
arXiv is the open archive where researchers post papers before they are formally published; this searches it and brings back each match's title and summary.
blog-post
Writes a full article on a topic, researched before a word of it is written and laid out to pull the reader in and end with a clear next step — and makes a cover picture to go with it.
code-review
A last pass over code that was just changed — does it actually do the job, do the tests still pass, did anything unsafe slip in — with the problems fixed rather than merely listed.
coding-prefs
Stops you having to repeat yourself: once you say how you like your code written, it is written down and read back before the next piece of work starts.
competitor-analysis
Works out who else is already selling into a market, what each of them is betting on, and where the space between them is.
cudf-analytics
Number-crunching on a big table of data — totals, averages, values that look wrong — pushed onto the graphics card, which gets through it far faster than the processor alone. On a machine without one the same work still runs, just slower.
cuml-machine-learning
Machine learning on a table of data: you hand it past records, it learns the pattern and then predicts or groups new ones — with the graphics card doing the training so it finishes fast.
data-visualization
Draws the charts that go with an analysis — bars, lines, scatter plots, heatmaps — and saves each one as an image file, coloured so the differences still show for a colourblind reader.
gpu-document-processing
Pulls the text and the tables out of long PDFs and whole folders of documents, handing the heavy work to a separate machine with a graphics card so a big pile gets through quickly.
langgraph-docs
LangGraph is LangChain's framework for building AI agents; this reads its official documentation live and answers from the current pages rather than from memory.
planning
Before a line of code is written, the project gets read through and the job becomes an ordered to-do list: which files change, in what order, and what could break.
query-writing
Ask in ordinary words what you want to know from a database, and the matching query gets written, run against it, and the answer laid out for you.
remember
Goes back over the conversation for the decisions and habits worth keeping and writes them down — a one-line rule into a notes file, a whole procedure into a reusable skill of its own.
schema-exploration
Opens up an unfamiliar database and explains what is inside it: which tables exist, what each column holds, and how the tables link to one another.
skill-creator
A good skill is mostly a decision about what to leave out — this is the guide to making that call.
social-media
A LinkedIn post or an X thread written to that platform's shape, researched first, and never handed over without an image to go with it.
web-research
Splits a question into a few separate lines of enquiry, researches them at the same time, and writes up one answer with the sources attached.
No skill in this package matches that.
HOW TO GET IT
WHOLE PACKAGEInstall the whole package (18 skills)
npx skills add langchain-ai/deepagentsSINGLE SKILLOr install a single skill
npx skills add langchain-ai/deepagents --skill <name> --full-depthPick the skill name from the Skills tab — each entry there installs independently.