Guide · Agent skill

How to use the Deep Research agent skill (cited reports)

Updated 2026-10-29 · written with AI from the project's own documentation and checked against our catalogue data

The deep-research skill is an agent skill that makes an AI agent run a multi-source web research process and return a written report where each claim links to its source. It is aimed at professionals doing competitive analysis, market sizing or due diligence, and at students or analysts who want a cited overview of a topic. By the end of this guide you will know what the skill does, how to install it, how to trigger it with a normal request, and what it needs before it works.

Deep Research with Citationsaffaan-m/ECC · MIT Download the skill

Quick start

1Download and unzip. Download the ZIP from this page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder.
2Or upload in the app. In the Claude apps, go to Settings → Capabilities → Skills and upload the skill.
3Configure search tools. Set up the firecrawl and/or exa MCP servers in ~/.claude.json or ~/.codex/config.toml, with your API accounts.
4Make a research request. Ask in plain language, for example "Research the current state of nuclear fusion energy".
5Answer clarifying questions. Reply to one or two questions about your goal and depth, or say "just research it" to use defaults.
6Review the report. Read the executive summary, check the cited sources and look for unverified or gap flags before acting on it.

What is an agent skill and what does deep-research do?

An agent skill is a folder with a SKILL.md file that tells an AI agent how to handle a certain kind of task. When the skill is installed and matches your request, the agent follows its workflow instead of answering from memory alone.

Once installed, the deep-research skill has the agent break a topic into 3-5 research sub-questions, search the web through firecrawl or exa tools, read the most promising pages in full, and write a structured report. Each claim in the report is meant to link to its source. The skill also asks the agent to flag claims that only one source supports as unverified, and to say plainly where it could not find good information.

Splits the topic into sub-questions and runs 2-3 keyword variations for each
Reads 3-5 key sources in full rather than relying only on search snippets
Produces a report with an executive summary, themed sections, key takeaways, a source list and a methodology note
Labels estimates, projections and opinions separately from facts

Install the deep-research skill

Download the ZIP from this page. Unzip it into your personal skills folder, ~/.claude/skills/, or into a project's .claude/skills/ folder if you want the skill available only inside that project.

Alternatively, upload the skill in the Claude apps under Settings → Capabilities → Skills.

The download contains agents/openai.yaml and SKILL.md. The SKILL.md file is the part the agent reads for its instructions.

Personal install: ~/.claude/skills/
Project install: .claude/skills/ inside your project folder
App install: Settings → Capabilities → Skills

Set up the search tools the skill needs

The skill does not search the web by itself. It relies on external search tools, either the firecrawl MCP server, the exa MCP server, or both. The source says both together give the best coverage.

MCP servers are configured in ~/.claude.json or ~/.codex/config.toml. You will also need API accounts for firecrawl and/or exa; the source does not state key details, so check each provider's setup for how to obtain and supply them.

Without at least one of these tools configured, the skill cannot deliver the cited search process it describes.

firecrawl tools used: firecrawl_search, firecrawl_scrape, firecrawl_crawl
exa tools used: web_search_exa, web_search_advanced_exa, crawling_exa

How do you trigger the skill with a natural request?

You do not need a special command. The skill is meant to activate when you ask for in-depth research on a topic, ask for a competitive analysis, technology evaluation, market sizing or due diligence, or use words such as "research", "deep dive", "investigate" or "what's the current state of".

The agent may ask one or two quick clarifying questions about your goal and the depth you want. If you say "just research it", it proceeds with reasonable defaults.

"Research the current state of nuclear fusion energy"
"Deep dive into Rust vs Go for backend services in 2026"
"Research the best strategies for bootstrapping a SaaS business"
"What's happening with the US housing market right now?"
"Investigate the competitive landscape for AI code editors"

What does a finished report look like?

The skill asks for a fixed structure: a title line with the generation date, source count and confidence level, an executive summary, numbered theme sections with inline citations, key takeaways, a numbered source list and a methodology note listing the queries run and sources analysed.

For short topics the full report is posted in chat. For long reports the agent posts the executive summary and key takeaways and saves the full report to a file.

# [Topic]: Research Report
*Generated: [date] | Sources: [N] | Confidence: [High/Medium/Low]*

## Executive Summary
## 1. [First Major Theme]
## Key Takeaways
## Sources
## Methodology

Safety, limits and tips

The skill does not run scripts, but it does use the network, because it sends searches and reads web pages through the MCP tools. Be aware that it can only read what those tools can access, so private or paywalled material is out of reach.

Treat the report as a starting point. The skill cross-references sources but does not verify them beyond that, and web sources may be outdated or wrong. Do not make medical, legal or financial decisions on the report alone.

For broad topics the skill can use Claude Code's Task tool to run several research agents in parallel, with the main session combining their findings.

Ask for recent sources if recency matters; the skill prefers sources from the last 12 months
Check the flagged unverified claims before relying on them
Use it for research, not quick factual questions, because the multi-step search is slow and token-heavy

What people use it for

Competitive analysis of AI code editors with sourced comparisons
Market sizing and growth trajectory for a product category
Due diligence background on a company, investor or technology
A cited overview of a technology for a student or analyst
A repeatable research process for any topic when firecrawl or exa is already set up

Frequently asked questions

How do I install the deep-research skill?

Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. You can also upload it in the Claude apps under Settings → Capabilities → Skills.

What does the deep-research skill require?

It requires at least one of the firecrawl MCP server or the exa MCP server, configured in ~/.claude.json or ~/.codex/config.toml. You also need API accounts for firecrawl and/or exa, as the source does not give key details.

Is the deep-research skill free?

The skill is licensed under MIT. The firecrawl and exa services it depends on are separate products, and the source does not state their pricing, so check their terms directly.

Does the skill verify its sources?

Not fully. It cross-references claims and flags anything supported by only one source as unverified. It does not verify sources beyond that, so web material may still be outdated or wrong.

Can I use it without firecrawl or exa?

The source says the skill relies on these MCP tools for searching and reading pages, and it lists no other search method. Without at least one of them configured, the cited search process will not work.

How does it compare with a normal chat answer?

A normal answer draws on the model's memory, while this skill searches the web through external tools and writes a report with links to its sources. It is slower and uses more tokens, so it suits thorough research rather than quick factual questions.

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 already have firecrawl or exa set up and need a repeatable, sourced research process for competitive analysis, market sizing or due diligence. Pick something else for quick factual questions, for medical, legal or financial decisions made on the report alone, or for private or paywalled material the search tools cannot reach.

Download the skill