tavily-research skill (Tavily): in-depth research reports from many sources

Makes your agent research a topic in depth with Tavily search and write a report built from many sources.

Installs: 22.8K; on skills.sh
Stars: 491; on GitHub
Files: 2; besides SKILL.md
Updated: Oct 2026; last repo commit

tavily-research makes your agent split a question into non-overlapping subtopics, search and read sources for each, and write one report with the sources cited. It triggers on requests to research, investigate or analyze something in depth. A single fact is a tavily-search job instead.

It works through the tvly command line tool from Tavily (the SKILL.md allows Bash(tvly *)), and it depends on the tavily-search skill for how to search. Search and extract work without an account, up to a rate limit. With subagents it writes notes to research_notes/ and the report to reports/ in your working directory.

Good for

Analysts researching a market, policy, technology or trend
Anyone comparing options across several criteria
Writers and students who need a sourced literature review

Install tavily-research

Claude Code

Copy this prompt.

Install the tavily-research skill in Claude Code. Run this command:

curl -fsSL https://cli.tavily.com/install.sh | bash
tvly init

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

tavily-research tells the agent to read tavily-search for every search and page read, so install both. The agent also needs the tvly command line tool or Tavily MCP tools. Search and extract work without an account up to a rate limit.

From the project README. Source.

What's in the skill

When it loads

Research a topic in depth with Tavily. Use this skill when a question needs many sources gathered and weighed together: analysing a market, policy, technology or trend, comparing options across several criteria, reviewing the literature, or any request to research, investigate or analyze something in depth. A single fact or answer to find, even one with many clues, is a tavily-search task. For one CRM record use tavily-entity, for a table of matches use tavily-list, and for a recurring check use tavily-monitor.

Sections in SKILL.md

Steps
1. Ask only when the answer would change the research
2. Split the question
3. Research the subtopics together
4. Fill thin parts, then write
5. Write and deliver the answer

Key facts

Author: Tavily (tavily-ai)
Repository: tavily-ai/skills (skills/tasks/tavily-research)
Installs on skills.sh: 22,781 as of 2026-11-02
GitHub stars: 491 for the whole tavily-ai/skills repo, 2026-11-02
License: MIT (as detected by GitHub)
Last commit to the repo: 2026-10-30
What is inside: 3 files including SKILL.md (two reference files: researcher.md and report-writer.md)

What it includes

Questions only when a term is ambiguous
The agent asks when a term has two meanings, you assumed a place, or a name matches two entities. Otherwise it starts.
Read · 1. Ask only when needed
3 subtopics by default, 4 for comparisons, 5 for entities
Subtopics must not overlap and must carry your limits on dates, places and scope.
Read · 2. Split the question
One subagent per subtopic, or parallel searches on its own
With subagents it creates research_notes/{title}/ and reports/ and starts one general-purpose subagent per subtopic. On its own it runs searches in parallel and keeps findings in context.
Try: “Research the state of EU battery recycling rules and compare how three countries enforce them.”
Runs code · 3. Research in parallel
Another search round before writing
Runs more searches for any section that would otherwise be thin.
Runs code · 4. Fill thin parts
One report, saved to reports/ only if you ask
Follows report-writer.md. With subagents, one agent writes reports/{title}.md from the notes. The full report is also given in the reply.
Write · 5. Write and deliver
Subagent rules: about 8 searches per subtopic, stop by 12
Each subagent uses tavily-search, extracts pages it cites, prefers authoritative sources, and writes notes with URLs next to each fact.
Runs code · references/researcher.md
Report shape: finding title, answer paragraph, cited sections
The writer cannot search. A claim that is not in the notes with a URL does not go in the report.
Write · references/report-writer.md

From the official docs

Permissions

Medium risk

The skill sends your search queries to Tavily and fetches web pages. It writes notes and reports in your working folder, so keep private details out of prompts.

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

Your research questions go to Tavily
Queries and URLs are sent to Tavily through the tvly CLI. Leave private or confidential details out of research prompts.
The README installs the CLI with a piped script
The README setup runs curl -fsSL https://cli.tavily.com/install.sh | bash. You can instead use uv tool install tavily-cli or pip install tavily-cli, both listed in the README.
It installs another skill if one is missing
researcher.md tells subagents to run npx skills add tavily-ai/skills --skill tavily-search when it is missing. Install tavily-search yourself first to avoid a surprise.

Technical details

Talks to	Tavily (search and extract queries and URLs, via the tvly CLI). Web pages the agent extracts
Sign-in and keys	tvly init signs in and stores a Tavily credential for the CLI; TAVILY_API_KEY is named for unattended runs.
Scopes	Runs tvly commands (allowed-tools: Bash(tvly *)) Writes research_notes/ and reports/ in the working directory Can start subagents that each run searches

Questions

What does the tavily-research skill do?

It makes your agent run a multi-step research job. The agent splits your question into subtopics, searches and reads sources with Tavily, and writes a report that cites them.

Does it need a Tavily account or API key?

Not for basic use. The README says tvly search and tvly extract work without authentication, with a keyless rate limit, and that map, crawl and research commands need authentication. For unattended runs it names the TAVILY_API_KEY environment variable.

How do I install it in Claude Code or Codex?

Use the install box at the top of this page. It installs tavily-search alongside, because tavily-research relies on it. You also need the tvly CLI, which the README installs with its guided setup. A Claude app upload tab is not offered because nothing in the SKILL.md says the Claude app can run tvly.

Do I need tavily-search too?

Yes. The SKILL.md says to search and read pages as tavily-search describes, and the researcher file tells subagents to install it with npx skills add tavily-ai/skills --skill tavily-search if it is missing.

When should I use a different Tavily skill?

The description says: a single fact is a tavily-search task, one CRM record is tavily-entity, a table of matches is tavily-list, and a recurring check is tavily-monitor.

Does it save files on my machine?

With subagents, yes. It creates research_notes/{title}/ and reports/ in the working directory and writes reports/{title}.md. Working on its own, it saves a file only when you ask for one.

Is it safe to install?

It runs tvly commands that send your search queries to Tavily and fetch web pages, and the Tavily CLI installer is a script piped from curl in its README. Read what the agent proposes, and do not put private details in research prompts. Review any report before you rely on its figures.

Information

updated Nov 2, 2026

Repository: tavily-ai/skills
Folder: skills/tasks/tavily-research

tavily-ai/skills skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 5

tavily-ai/skills README · github.com
tavily-search SKILL.md · github.com
tavily-research SKILL.md · github.com
researcher.md · github.com
report-writer.md · github.com