1. Install all 5
In Claude Code
Open Claude Code and paste this. Claude does the whole install for you.
Clone https://github.com/TakeOff-Team/claude-research-codes and run its install.sh. If anything fails, fix it and run it again.
Then restart Claude Code and type /. Your codes will be in the list, in every project.
In the Claude app
- Turn on code execution. Go to Settings, then Capabilities, and turn on "Code execution and file creation".
- Download the 4 zips. fact-check, papers, deep-research and competitors.
- Upload them. Go to Customize, then Skills, click +, and upload each zip.
- Add the fifth one. Download last30days.skill and upload it the same way.
- Turn on web search, then type
/in a chat to see your codes.
Everything is on GitHub if you want to read exactly what each code tells Claude to do.
2. The 5 codes
/fact-check: Find out what is actually true
Give it a document, a link or pasted text. It pulls out every claim that could be proven right or wrong, checks each one against real sources on the live web, and gives each one a verdict: true, partly true, out of date, false, or "can't back it up". You get the score, what to fix first, a source link for every claim, and corrected lines ready to paste.
When to use it. Run it on anything with facts in it before you send or post it, especially anything an AI wrote. It also catches the classic AI mistakes, like a real number credited to the wrong source.
/fact-check Check this document before I send it to a client: [paste your document or drop in the file]
/last30days: See what real people are saying right now
Give it any topic. It searches places like Reddit, X, YouTube, TikTok and Hacker News for what people have said in the last 30 days, ranks it by what people actually engaged with, and writes one summary with citations.
When to use it. Use it before you write about a topic, launch something, or buy something. You get the conversation happening now instead of whatever a search engine ranked two years ago.
/last30days What are small business owners saying about using AI for customer support?
/papers: Get the real research, in plain English
Give it a question. It searches free research databases like OpenAlex and PubMed, finds the papers, and tells you what they found, how strong the evidence is, and where the papers disagree. Every paper comes with a link, and each one is labeled so you know if it was read in full or just the abstract.
When to use it. Use it any time you catch yourself saying "studies show". It only cites papers it actually found during the run, so you are not trusting a citation from memory.
/papers Does creatine actually improve memory?
/deep-research: Get a full report on a big question
A normal answer does one search and summarizes the first results. This one works in rounds. It searches, reads, looks at what surprised it and what is still missing, then searches again. It also goes looking for people who disagree. You get a full report with a clear answer, the case against that answer, what is still unknown, and every source cited.
When to use it. Use it for real decisions: which tool to buy, whether to enter a market, whether an idea holds up.
/deep-research Is a 4-day work week a good idea for a small business with 10 employees?
/competitors: Get a full profile on everyone you are up against
Give it competitor websites. It reads their homepage and pricing, then goes outside their site to see what customers say in reviews. You get a profile on each one with pricing, strengths and weak spots, plus a side-by-side table and the openings nobody is filling. If a price is not public, it says so instead of guessing.
When to use it. Use it before you price something, write a sales page, or pitch a client. Tell it your size, like your subscriber count or team size, so it can look for the price that applies to you.
/competitors beehiiv.com kit.com mailerlite.com. My business is a free AI newsletter with 2,700 subscribers.
3. Use them together
The best move is to run one code on another code's work. Have Claude write the report, then have it check the report as a separate job. A second pass that's only looking for mistakes catches things the first pass missed.
/deep-research [your question] When the report is done, run /fact-check on it. Fix anything that comes back false, out of date or unsupported, then show me the final version and a list of what changed.
5 things to know
- Web search has to be on All five read the live internet. In the Claude app, turn on web search. Claude Code has it built in.
- /last30days has a quick setup Reddit, Hacker News, Polymarket and GitHub work right away. The first time you run it, its setup wizard adds X, YouTube, TikTok and more. Some of those may ask you to connect an account or add a key.
- The big ones use more of your limit /deep-research and /competitors read a lot of pages, so they use more of your Claude limit than a normal chat. A focused question gets you a better report and costs less than a broad one.
- Some pages can't be read Paywalls, login walls and pricing calculators can block Claude. The codes are told to say what they couldn't read instead of guessing, so read that part of the report.
- Click the links on anything big These codes show their sources so you can check them, and they can still get things wrong. Before you bet money or your name on an answer, click through.
Credit
/last30days is built by Matt Van Horn. The pack installs it straight from his repo. Go give it a star.
The other four were written for this pack, and these public skills were studied for ideas: fact-check by jwynia, research-papers by Firecrawl, deep-research by 199 Biotechnologies and competitor-profiling by Corey Haines.