Same Hyperliquid trader. Same 90 days.
Copied at 10x: liquidated 27 times.
Copied at 5x: 4 times, and more than double the profit.
I built a free tool that runs this test on any Hyperliquid trader before you put money in. https://x.com/i/article/2114489356862070784
Strike Copilot: the full guide to setup and every skill
TL;DR: Strike Copilot is a free, open-source toolkit that runs inside Claude Code on your own computer. You ask in plain English, and it checks traders, replays them on your own fee and leverage, and sets up copies only after you say yes to each change. Install it, type "Set me up", and run a review once a week.
Most copy-trading mistakes happen before the first copy
Most copy traders pick a lead from the leaderboard and find out a week later what they signed up for. By then the big choices are locked in: trader, copy mode, leverage and caps.
I built Strike Copilot with Claude Code to make those choices with numbers instead of hunches. Its checks are generalised from the scripts I use on my own two Strike accounts. On my 12 Oct review, one replay changed a setting: a lower total cap would have done better, so I lowered it.
What is Strike Copilot, and what isn't it?
It's a folder of scripts and instructions that Claude Code reads. You chat, Claude runs the scripts, and you get plain-English answers. It isn't a bot or a service: nothing runs on anyone's server, and nothing is sent to me.
The ground rules are written into the tool itself:
- It never places, changes or cancels your own orders, withdraws or transfers. Order endpoints aren't wired up, and Strike's API keys can't withdraw, by design.
- Your private key stays on your computer, in a hidden file (.secrets/strike_api.key). It's never shown or sent anywhere.
- Every account change needs a yes. Claude shows the exact request as a dry run ("DRY RUN: nothing sent"), explains it, then gives you a menu: Make this change / Skip this one / Explain more / Stop setting up. No option is marked as recommended.
- It sends data only to Strike and Hyperliquid, through their public APIs. There's no tracking.
After that menu, it can set leverage and margin mode, add margin to an open position, subscribe to a copy, and change, stop or set a stop loss on a subscription.
What do you need before you start?
You need a Strike account, Claude Code, Python 3.10 or newer, and about 30 minutes for the first run, most of it the sweep running by itself. It works on Mac and Windows.
- A Strike account (wallet or email login), funded with the money you plan to copy with.
- Claude Code, Anthropic's assistant, in a terminal or the Claude desktop app. It needs a Claude subscription (Pro or Max) or an Anthropic API account. Install guide: claude.com/code.
- Python 3.10+. On a Mac, check with python3 --version. On Windows, install it from python.org and tick "Add Python to PATH".
How do you set it up? A step-by-step checklist
Setup is one conversation, one step at a time.
1. Download the toolkit. With git: git clone https://github.com/benryandev/strike-copilot.git. Without git: press the green Code button on the GitHub page, then Download ZIP, and unzip it.
1. Install its one package. In the folder, run python3 -m pip install -r requirements.txt (use python on Windows if python3 isn't found).
1. Start Claude in the folder. Type claude, or open the folder in the desktop app. Then type Set me up.
1. Make the API key on your computer. Claude creates a key pair and shows you only the public half.
1. Add it on Strike. Open app.strikefinance.org/api-keys while logged in. Enter a name, paste the public key, and fill in Days to expire (the tool suggests 90; leave it empty and the key never expires). Press Authorize. Don't press Generate: that makes a key in your browser, and the tool never gets the private half.
1. Say "done". Claude checks the key works and reads your account and any sub-accounts.
1. Pick where to copy from. Claude recommends a sub-account just for copying.
1. Answer the referral question, if it appears (see below).
1. Answer eight multiple-choice questions, in two rounds:
- Balance: $250, $500, $1,000, $5,000 or your own amount
- Leverage to start from: 5x suggested (or 3x, 10x, 20x). The comparison later tests 10x, 5x, 3x and 2x against each trader and picks one
- Most of the balance in open copies at once: 50% suggested (your total cap)
- The dip you could sit through: 20% suggested (deeper-dipping traders get flagged)
- Markets never to copy (e.g. PUMP-USD, NIGHT-USD, stocks and commodities)
- Copy mode: either, fixed margin or fixed ratio
- Margin mode: isolated (suggested) or cross
- Review day: Monday suggested
10. Run your first sweep when Claude offers.
Your answers live in profile.json, on your computer only. Change them any time: "Change my balance to $2,000".
My rule: make a sub-account just for copying, and set an expiry on every API key. Both cap the damage if something goes wrong.
What does each skill do?
There are six skills. The mode comparison isn't one of them: it's a script that sweep and check-trader run, and you can ask for it directly.
Sweep: who's worth copying for my account?
A sweep checks Strike's scored set of about 5,000 Hyperliquid wallets plus Strike's own traders in 20–30 minutes. Ask for a wide sweep to cover the whole Hyperliquid leaderboard, about 45,000 wallets, in about an hour.
Each filter comes from copying for real. At least 70% of the last 90 days' realised profit must be on markets Strike lists and you copy, across at least 20 round trips. The trader's edge must be at least 20 bp per dollar, because a copy round trip costs about 15 bp in fees and price gap.
You get 2–3 options, each with its catch and its free copier slots. If nothing passes, it says so: no lead is better than a bad lead.
Check-trader: should I copy this one?
Paste a Hyperliquid address or a Strike trader's link and you get the same checks in about 30 seconds, each fail explained in one plain sentence. A profile shows the whole account. A copier only gets the perp trades on Strike's markets, after fees and the price gap. I don't rely on any one score. I run my own checks.
Compare: fixed margin or fixed ratio?
Fixed margin puts the same dollar margin on every copied entry. Fixed ratio scales each copy to the trader's trade size relative to their account.
The tool replays both over the last 90 and 30 days, on your balance, fee and exclusions, at 10x, 5x, 3x and 2x, with liquidations included. It scores each by profit per dollar of drawdown: net profit divided by the biggest drop. The clearly higher score wins. Within 10%, the smaller drop wins, because it compounds better and it's easier to sit through. If both modes lost money, it recommends neither.
Leverage is tested, not assumed. The replay checks every hourly high and low against Strike's maintenance margin, so a copy that would have been liquidated is counted as liquidated. It then picks the leverage with the most profit per dollar of drawdown.
You get a table per period (net, lowest point, biggest drop, fees, volume, skips), a verdict with the reason, and the settings to use.
Caps: how much can one copy hold?
Strike has two caps: max margin per symbol and max total margin. An entry that would go past a cap is skipped, not shrunk, and closes are never blocked.
The caps skill replays no caps, then 3–4 options side by side. Caps that never bind change nothing in the past: they guard against the trader changing behaviour. Caps that bind often cut both losses and winners.
Apply-copy: set it up without missing a setting
You choose: through the API, or numbered steps to do it yourself in the app. Through the API, it sets margin mode and leverage on the trader's markets, then subscribes with both caps and your exclusions. Each step gets its own dry run and menu.
Leverage comes first because, by default, copies use your leverage and margin mode on each market, not the trader's, and copying won't change them for you. (A copy with its own sub-account can now be set to follow the trader's leverage instead, up to a cap you choose.) The first trade on a market locks its settings. When I tested on a new account, some markets started on Cross at 20x.
Your first copy lands on the trader's next trade. Positions they already hold aren't copied.
Review: five minutes a week
On your review day, say "Review my account". Claude checks for a newer version and asks before installing it, then shows your balance, drop from peak and each copy's profit and loss. It flags only what needs a decision:
- you're past the dip you said you could sit through
- a trader has gone idle, or their profit is moving to markets Strike doesn't list
- how far each open position is from liquidation
- a copy the trader no longer holds
- your copies that were force-closed this week
- copies your caps skipped this week, and roughly what they'd be worth now
- if other people copy you: how many, and the profit share you've earned
- a market whose leverage or margin mode drifted from your profile
If nothing needs attention, it says so in one line.
Between reviews, I use Copilot Watch, a free Telegram and Discord bot I also built. It tells you when a copy is close to liquidation, when the trader has closed a position your copy still holds, or when one of their trades wasn't copied. It reads Strike's public data only and never asks for a key. If you'd rather not install anything, Copilot Check checks and paper-copies a wallet in your browser.
When a copy gets close to liquidation
Ask "Add $50 margin to my BTC position" and it shows the new liquidation price before you say yes. It only ever adds margin; it never removes it. I did this three times on 22 Oct, after Copilot Watch flagged two copies.
You can also set a stop loss on a whole copy ("Set a $300 stop loss on my copy of 0x…"): once the copy has lost that much since it started, Strike stops it and closes its positions. Ask for a risk check to see what price shocks and weekend gaps on stock and gold positions would do, and ask "Show my recent changes" to see everything the tool has changed, with before and after values.
A worked example: testing leverage on my own trader
I copy one trader on my sub-account at a fixed $100 per trade and 10x. The tool used to suggest 10x too. Then I added liquidations to the replay, and 10x stopped looking sensible.
On 26 Oct I ran the comparison on that trader: a $7,000 account, up to half of it in copies, isolated margin, the last 90 days.
At 10x, fixed margin had the biggest drop and 27 liquidations for one of the smallest profits. The trader runs much of the book on cross, so it rides out dips that wipe a 10x isolated copy.
The verdict was neither: no leverage passed, and at 2x, the safest tested, both modes lost money over the last 30 days. So the tool won't recommend this trader right now. That's the check doing its job. It's why the setup now starts at 5x and tests every trader before suggesting a leverage.
The tool sized fixed margin at $25 per trade here, because this trader has held up to 128 trades at once, so these aren't my sub-account's own results. The direction matches what my sub-account has lived through, though: about $1,900 lost to liquidated copies of trades the trader held on to.
What about the referral code?
I share a referral code, BenRyan, and earn a share of the fees of people who use it. The tool handles that like this:
- Setup checks whether you already get a referral discount. If you do, referrals never come up.
- If you have no discount, it asks once, says plainly that I get a share of the fees, and says the tool works the same either way. Say no and it never asks again.
- It can't set a code itself. Strike only allows that in the app, so if you say yes, you apply it yourself. It never touches a code you already have.
One more disclosure: more trading volume helps my referral tier. So whenever the recommended copy mode trades more volume than the other, the result says so, with both volumes. The verdict still follows the drawdown rule above.
What are the limits worth knowing?
Replays are simulations, and some copy-trading facts affect results whatever tool you use.
- Replays aren't fills. Open copies are valued at Hyperliquid's hourly prices between trades. Every copied close is assumed to land 10 bp worse than the trader's, as a copy-cost allowance. Liquidations are modelled from hourly highs and lows; funding payments and Strike's liquidation fee aren't.
- By default, copies use your leverage, not the trader's. A trader at 3x on cross can ride out a drop that liquidates a 10x isolated copy. It happened to one of my copies in the 8 Oct sell-off: a $50 NEAR long.
- A trader's liquidation closes your copy. If the trader is liquidated, Strike closes your copy by the same fraction. It doesn't work the other way: your copy can be liquidated while the trader holds on, as above.
- Stopping a copy closes what it copied. In the app or through the API, stopping closes your copied positions at market. The app doesn't cancel resting copied orders, so check your open orders after. To stop new copies but keep what's open, the tool excludes every market instead: open copies still close when the trader closes. It always asks first: stop and close, pause new copies, or don't stop.
- Adding margin moves the liquidation price away, but puts more money into that one position. If the trader holds on and the price keeps going, you lose more than the original margin.
- Fixed ratio sizes off the trader's whole account. If the trader withdraws, every later copy gets bigger.
- Each trader takes 40 copiers at most.
- What it can't do: place or close trades, withdraw, set a referral code, or create sub-accounts.
If a script fails, Claude explains it and suggests reporting it on GitHub Issues, without keys or account IDs.
FAQ
Can Strike Copilot withdraw my money or trade for me? No. Strike API keys can't withdraw, and the tool never places, changes or cancels your own orders. The only account changes it can make are leverage, margin mode, adding margin, and copy subscriptions with their caps and stop loss, each after a dry run and your yes.
Does it cost anything? The tool is free and MIT-licensed. Claude Code needs a Claude subscription or API credit. Strike charges its normal trading fees on copies.
How do I remove it? Delete the API key on Strike's API keys page, then delete the folder. Nothing else is installed except one Python package (cryptography).
Checks first, then the copy
Strike Copilot puts the checks before the money: the right trader and mode, leverage set first, caps tested, and a weekly review. It changes nothing without your yes, and it can't withdraw.
For the background, my Strike deep dive covers the venue and its copy trading, and how I pick a trader to copy goes deeper on the checks behind the sweep. The pinned launch post is here, and the how-to series started with why most copy-trading mistakes happen before the first copy, then API keys.
Get it free on GitHub: github.com/benryandev/strike-copilot. New to Strike? Open Strike here (my referral link, 5% off fees).
Not financial advice. I'm a Strike user, not affiliated with Strike Finance.