Guide · Agent skill
How to use the Deep Research Methodology agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The Deep Research Methodology skill (skill name: deep-research) is a set of written instructions that tells an AI agent how to research a topic before answering or creating content. Instead of running one web search and writing from memory, the agent surveys the topic, breaks it into subtopics, searches each one, reads key sources in full, and checks for data, examples, expert views and criticisms. This guide explains what the skill does, how to install it, how to trigger it with natural requests, and what to watch for.
Deep Research Methodologybytedance/deer-flow · MIT Download the skill Quick start
- 1Download the ZIP. Download the deep-research skill ZIP from this page.
- 2Unzip into a skills folder. Unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder for one project.
- 3Or upload in the Claude apps. In the Claude apps, go to Settings → Capabilities → Skills and upload the skill.
- 4Check the tools. Make sure the agent has a web search tool and a web_fetch tool available.
- 5Make a research request. Ask a question such as 'research X' or 'compare X and Y', adding a time frame if you need current information.
- 6Review the sources. Read the agent's answer and check the claims that matter, especially for medical, legal or financial topics.
What is an agent skill and what does this one do?
An agent skill is a folder containing a SKILL.md file with instructions that an AI agent loads when a task matches the skill's description. The deep-research skill's description tells the agent to use it instead of a basic web search for questions that need online information, such as 'what is X', 'explain X', 'compare X and Y', or 'research X'.
When loaded, the skill tells the agent not to rely on a single web search or its general knowledge. It follows four phases: a broad survey of the topic, targeted deep-dive searches for each subtopic, a diversity check for facts, examples, expert opinions, trends, comparisons and challenges, and a synthesis check before writing.
- Phase 1: broad exploration and identifying subtopics
- Phase 2: targeted searches and reading full pages with web_fetch
- Phase 3: checking for data, cases, expert views, trends, comparisons and criticisms
- Phase 4: a checklist that asks whether at least 3-5 angles were searched before writing
How do you install the deep-research skill?
Download the ZIP file from this page. Then choose one of two routes. For a personal setup available across your projects, unzip the download into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder.
If you use the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.
- Personal: unzip into ~/.claude/skills/
- Single project: unzip into the project's .claude/skills/ folder
- Claude apps: Settings → Capabilities → Skills, then upload
How do you trigger the skill with natural requests?
You do not need a special command. The skill is matched by its description, so ordinary requests work. Asking the agent to research, explain, compare or investigate a topic is enough. The skill is also meant to load before content generation, so you can mention a presentation, article, report, UI design, or video when you ask for it.
If you want to be sure the research happens, say so directly, for example by asking for research before drafting. Include a time frame when you need current information, because the skill tells the agent to check the current date and use it in searches.
- 'Research the current state of solid-state batteries, including limitations.'
- 'Explain how vector databases work and compare them with traditional databases.'
- 'Before you write the slide deck, research the latest trends in remote work tools.'
- 'Find recent developments in offshore wind this month and summarise them.'
What does the skill need to run?
The skill needs two tools available to the agent: a web search tool, and a web_fetch tool to read full pages. Without web search the agent cannot follow the process. Without web_fetch it can only work from search snippets, which the skill specifically warns against.
The skill runs no scripts and does use the network, because searches and page fetches send queries to external services.
- A web search tool available to the agent
- A web_fetch tool to read full pages
- Runs scripts: no
- Uses network: yes
What should you know before using it for sensitive work?
The skill only guides how the agent searches. It does not check sources, so it is not a substitute for verified accuracy in medical, legal or financial decisions. Confirm those facts with qualified professionals and primary documents.
Because queries and page fetches go to external services, the skill is not suited to offline work or to privacy-sensitive material. The process is also slow and token-heavy by design, so it is more than you need for a quick factual question that one lookup can answer.
- Not a source checker: verify medical, legal and financial claims yourself
- Not private: queries are sent to external search and fetch services
- Not fast: expect many searches and page reads
What tips get better results?
Give the agent specific context. The skill's own examples show that 'enterprise AI adoption trends 2024' beats 'AI trends'. Adding phrasings like 'case study', 'statistics', 'expert interview', 'limitations' or 'criticism' helps the agent reach the diversity types the skill asks for.
For time-sensitive questions, make the time frame explicit. The skill's guidance is to match precision to intent: month and day for 'today', month for 'recently', and year for 'trends'.
- Name the topic and context precisely
- Ask for benefits and limitations together to get a balanced view
- Say what the output is for (report, slides, article, UI) so research covers the needed facts
- State a time frame such as this week or this month for current information
Topic example from the skill: "AI in healthcare" → dimensions: diagnostic AI, treatment recommendation, administrative automation, patient monitoring, regulatory landscape, ethical considerationsWhat people use it for
- Gathering facts, data and case studies before writing an article or report
- Collecting current trends and examples before building a presentation
- Researching a topic from both benefits and limitations for a balanced write-up
- Collecting real-world context before designing a UI or frontend mockup
Frequently asked questions
Is the deep-research skill free?
The skill is published under the MIT licence, as stated on this page. Check the licence in the source repository for the full terms before redistributing it.
What do I need to use the deep-research skill?
You need an agent that has a web search tool and a web_fetch tool for reading full pages. The skill runs no scripts but uses the network, since searches and page fetches go to external services.
How do I install the deep-research skill in Claude?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. In the Claude apps you can upload it under Settings → Capabilities → Skills.
Does the skill check whether sources are accurate?
No. The skill only guides the search and reading process and does not verify sources. For medical, legal or financial decisions, confirm facts with qualified professionals and primary documents.
Why is the research slow, and when should I skip the skill?
The process uses many searches and full page reads by design, so it is slow and token-heavy. For a quick factual question that one lookup can answer, a simple search is the better choice.
How is this different from a normal web search?
A normal search often returns one set of snippets. This skill asks the agent to survey the topic, break it into subtopics, read key sources in full and check for data, examples, expert views and criticisms before writing.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use deep-research if you want well-sourced answers, a report, article, presentation or UI design that needs current facts, or a view of both benefits and limitations. Pick something else if you need a quick lookup, offline or privacy-sensitive work, or verified accuracy for medical, legal or financial decisions.
Download the skill