A multi-agent research skill for Claude Code. Tell Claude "do quick research on X" and it will:
- Check the current state of the topic (so it doesn't recommend stuff that's been deprecated)
- Spawn 2–4 research helpers in parallel
- Pull from multiple sources
- Write a clean, cited Markdown report and save it to
docs/research/in your project folder
Cuts research time roughly in half vs doing it manually, and the output is consistently cited so you can verify everything.
Not really. Claude Code doesn't have a package manager. "Skills" and "agents" in Claude Code are just files in specific folders that Claude reads automatically.
To "install" this, you copy 2 small bits into 2 folders on your computer. That's it. No download tool, no install command, no account.
quick-research-export/
├── README.md ← you're reading this
├── skills/
│ └── quick-research/
│ ├── SKILL.md ← the main skill instructions
│ └── references/
│ ├── architecture.md
│ └── prompts.md
└── agents/
└── research-assistant.md ← the helper agent the skill calls
Two pieces:
skills/quick-research/— the skill folder (Claude readsSKILL.mdto know what this skill does)agents/research-assistant.md— the helper agent the skill spawns to do parallel research
You're going to copy these two pieces into the corresponding folders in your home directory.
On your machine, Claude Code looks in:
- Skills:
~/.claude/skills/(one sub-folder per skill) - Agents:
~/.claude/agents/(one file per agent)
The ~ symbol means "your home folder." On Mac that's /Users/yourname/. On Linux it's /home/yourname/. The .claude folder is hidden (starts with a dot) — you may need to enable "show hidden files" in Finder to see it.
If the ~/.claude/skills/ or ~/.claude/agents/ folder doesn't exist yet, just create it.
Copy the entire skills/quick-research/ folder from this pack into ~/.claude/skills/.
After copying, you should have:
~/.claude/skills/quick-research/SKILL.md
~/.claude/skills/quick-research/references/architecture.md
~/.claude/skills/quick-research/references/prompts.md
Copy the agents/research-assistant.md file from this pack into ~/.claude/agents/.
After copying, you should have:
~/.claude/agents/research-assistant.md
If you have Claude Code open, close it and reopen it so it picks up the new files. New sessions will see the skill automatically.
Just talk to Claude Code normally. Any phrasing that signals "do real research" will trigger it:
- "Do a quick research on the top 10 AI coding assistants right now."
- "Compare Notion vs Obsidian vs Logseq for personal knowledge management — give me a real report."
- "Research the EU AI Act and how it affects small businesses."
- "Find me the best espresso machines under $500 in 2026."
Claude will:
- Check today's date and run a few baseline searches to know what's current
- Spawn parallel research helpers
- Synthesize their findings
- Save a Markdown report to
docs/research/in your project folder - Give you a summary in the chat
The report file will be named like docs/research/2026-05-26-espresso-machines.md so you can find it later.
The skill primarily uses Claude Code's built-in WebSearch. If WebSearch ever fails, it can optionally fall back to Exa (a search API).
You don't need to set this up. If you don't have an Exa account, the skill works fine on WebSearch alone.
If you ever want the fallback:
- Sign up at exa.ai (they have a free tier)
- Get your API key from the dashboard
- Add this to your shell profile (
~/.bashrcor~/.zshrc):export EXA_API_KEY="your-key-here" - Restart your terminal.
Skip all of this if you're just starting out.
Claude doesn't seem to use the skill when I ask for research.
- Make sure you fully closed and reopened Claude Code after copying the files.
- Check the file paths are exactly:
~/.claude/skills/quick-research/SKILL.mdand~/.claude/agents/research-assistant.md. - Try being more explicit: "Use the quick-research skill to research X."
The report isn't saving to docs/research/.
- The skill saves to the project folder you're currently working in. If you're in
/Users/yourname/Documents/myproject/, the report goes to/Users/yourname/Documents/myproject/docs/research/. - If the
docs/research/folder doesn't exist, the skill should create it. If it doesn't, just create it yourself.
I'm getting search rate limits.
- WebSearch has limits per session. The skill is designed to stay within them (max 10 sub-agents, max 3 iterations), but if you run lots of research back-to-back you may hit limits. Wait a few minutes and try again.
- It does not replace deep multi-day research. For massive projects, use Gemini Deep Research or ChatGPT Deep Research.
- It does not access paywalled academic databases — just the public web.
- It does not remember past research between projects (each report is standalone in that project's
docs/research/).
For everyday "I need a solid 1-page writeup on X" requests, it's the right tool. For "I need to understand the entire history of quantum computing," go deeper elsewhere.
Skill design inspired by Anthropic's published multi-agent research system and LangChain's Open Deep Research. Cleaned and packaged for general use.