How to Install the Deep Research Skill in Claude Code

The Deep Research skill turns Claude Code from a thing that answers questions into a thing that goes and finds out. Instead of one search and a summary, it fans out across many sources, fetches the actual pages, checks claims against each other, and hands you a cited report. If you've ever spent two hours with 30 browser tabs open trying to get to the bottom of something, this is the tool that does that part for you.

Installing it takes about two minutes. Using it well takes a little more thought, because the quality of what comes back depends almost entirely on how you scope the question. Here's the whole thing.

What the Deep Research skill does

A normal Claude query might run one or two web searches and synthesize an answer. Deep Research runs a structured process: it breaks your question into sub-questions, searches each one, fetches the real sources rather than trusting snippets, looks for claims that contradict each other, and then writes a report with citations you can check. The output is closer to what a research assistant would hand you than to a chatbot reply.

That makes it worth installing for a specific kind of task: anything where you need to be confident the answer is right and you need to see where it came from. Competitive research, due diligence, "what's the current state of X," literature scans. For "what's the capital of France," it's overkill.

Before you start

You need two things: Claude Code installed and working, and a terminal you're comfortable running a few commands in. That's it. If you're on Claude Code already, you're ready.

Step 1: Find your skills directory

Claude Code loads skills from a skills folder in your config directory. On Mac and Linux it's at ~/.claude/skills/. On Windows it's under your user folder at %USERPROFILE%\.claude\skills\.

# Mac / Linux
cd ~/.claude/skills/

# If the folder doesn't exist yet, create it
mkdir -p ~/.claude/skills/
cd ~/.claude/skills/

Step 2: Install the skill

A skill is a folder containing a SKILL.md file. Installing means getting that folder into your skills directory. If the Deep Research skill is in a Git repo, you clone it:

# From inside ~/.claude/skills/
git clone https://github.com/anthropics/skills.git

# This pulls a collection; the deep-research skill is one folder inside it.
# If you only want that one skill, copy its folder out:
cp -r skills/deep-research ~/.claude/skills/deep-research

If you got the skill as a download or from a marketplace, the job is the same: end up with a deep-research folder containing SKILL.md inside ~/.claude/skills/. The folder name and the SKILL.md are all Claude needs.

Step 3: Verify it loaded

Start a fresh Claude Code session, or restart your current one, so it re-scans the skills directory. Then check that the skill is visible. You can ask Claude directly:

Which skills do you have available? Is deep-research one of them?

If it lists deep-research, you're done installing. Claude reads each skill's description on startup and loads the full instructions only when a task matches, so seeing it in the list means it's ready to fire when you ask a research question.

Step 4: Run your first research task

This is where most people get mediocre results, and it's avoidable. The skill is only as good as the question. A vague prompt gets a vague report.

The weak way to ask:

Research electric vehicles.

That's a topic, not a question, and you'll get a broad, shallow overview. The strong way gives scope, angle, and what you'll do with the answer:

Research the state of solid-state EV batteries as of 2026: which companies
have shipped or announced production timelines, what the main technical
hurdles still are, and how close they are to beating lithium-ion on cost.
Cite sources and flag anything that's a press claim versus an independent test.

That last instruction matters most. Asking it to distinguish a company's marketing claim from an independent result is exactly the kind of adversarial check the skill is built to do, and naming it up front gets you a sharper report.

Step 5: Read the report like a skeptic

When the report comes back, don't just take the conclusion. The reason this skill is worth using over a plain answer is the citations, so use them. Click through to a couple of the sources for any claim that surprises you or that you're going to act on. The skill checks claims against each other, but it's not infallible, and a source it trusted might be wrong. Treat the report as a very well-organized starting point, not a verdict.

For anything high-stakes, due diligence on a vendor, a medical question, a financial decision, verify the load-bearing facts yourself. The skill saves you the hours of gathering and cross-referencing. It doesn't remove your judgment from the loop.

The security note you shouldn't skip

A skill is instructions and sometimes code that run with your permissions. Before installing any skill, open its SKILL.md and read it. It's plain text. You can see exactly what it tells Claude to do and what external services it calls. The Deep Research skill needs web access to do its job, which is expected. What you're checking for is anything beyond that: does it send your local files anywhere, does it call a service it doesn't need to. Install from the official Anthropic repo when you can, and audit anything from a source you don't recognize. We go deeper on this in our guide to auditing skill security.

When to use it, and when not to

Use Deep Research when the answer needs to be defensible and you need to see the sources: market research, technical state-of-the-art questions, comparing claims across vendors, any "is this actually true" investigation. It's also the right tool when you'd otherwise spend an afternoon reading.

Skip it for quick factual lookups, for questions Claude already knows cold from training, and for anything where you don't actually need citations. Each research run does real work and costs more tokens than a normal reply, so save it for the questions that earn it.

Installed once, it sits quietly until you ask a question that needs it. That's the right way to think about the whole skills system: a set of capabilities Claude reaches for when the task calls, not tools you have to manage by hand.

A real example, start to finish

Say you're deciding whether to switch your team's project tool. Here's the session. You ask Deep Research to compare the three tools you're considering on pricing, integrations, and what recent users actually complain about, with sources. It comes back with a cited rundown, and the complaints section, pulled from real reviews and forums, is the part that saves you, because that's the stuff vendor sites never tell you.

Then you keep going in the same session. You ask Claude to put the comparison into a spreadsheet so you can score each option against your own criteria. If you have a spreadsheet skill installed, it builds the table. You ask for a one-page summary to send your team, and a document skill turns it into something shareable. One question became a research report, a scoring sheet, and a memo, without you opening a single browser tab.

That chaining is the point. Deep Research does the finding; the other skills do something with what it found. They load independently and hand off inside one conversation.

FAQ

Is the Deep Research skill free? The skill itself is free to install. You pay for the Claude usage it consumes, and because it runs many searches and fetches full pages, a deep research task costs more tokens than a normal chat reply. Use it for questions that justify the spend.

How is it different from just asking Claude to search the web? A normal search runs once or twice and summarizes. The Deep Research skill breaks your question into parts, searches each, fetches real sources, cross-checks claims for contradictions, and returns a cited report. It's a process, not a single lookup.

Do I need coding skills to install it? No. The install is a couple of terminal commands you can copy from this guide. Using the skill is plain-language prompting. If you can open a terminal and run git clone, you can install it.

Why doesn't the skill show up after I install it? Almost always because the session was already running when you added the folder. Restart Claude Code so it re-scans the skills directory. Also check the folder is directly inside ~/.claude/skills/ and contains a SKILL.md file.

Can I trust the report it produces? Treat it as a strong, well-sourced starting point, not a final verdict. The skill cross-checks claims, but sources can be wrong. For anything you'll act on, click through the citations and verify the key facts yourself.

What else pairs well with it? A spreadsheet skill for crunching any data the research surfaces, and a document skill if you want to turn the report into a polished deliverable. The skills load independently, so you can chain them in one session.