Suprik (@Supriik_) header
Suprik (@Supriik_) profile picture

crypto degen who replaced half his workflow with AI. bots, agents, prompts. no fluff, just what actually works

Joined December 2019
Photo by Suprik (@Supriik_)Photo by Suprik (@Supriik_)Photo by Suprik (@Supriik_)Photo by Suprik (@Supriik_)Photo by Suprik (@Supriik_)Photo by Suprik (@Supriik_)Photo by Suprik (@Supriik_)Photo by Suprik (@Supriik_)Photo by Suprik (@Supriik_)Photo by Suprik (@Supriik_)
its f*cking insane we have a video how binance laptop looked at hack $5.5B left Binance hot wallets in seconds → new session key from an unknown IP → a signer key exported → multisig dropped from 3/5 to 1/5 → the monitor still says "status: OK" then 8 transactions signed in under 2 seconds 164,464 ETH 1.6B USDT 823M USDC 18,270 BTC all into one wallet: 0x5227...2f8c the alert fired AFTER the last signature freeze failed. 8/8 already broadcast. reverted: 0 the biggest exchange on earth had alarms they went off when the money was already gone coins on an exchange are a promise, not a balance how much are you still keeping on a CEX?
20
11
114
23,846
it's f*cking goldmine i stopped paying for antivirus i hired 6 little Claudes instead every Sunday at 3am they sit down at their desks and scan my whole computer. it costs me $0 each one has one job: • SIGNATURES checks every file against millions of known malware fingerprints (ClamAV) • PATTERNS looks inside files for tricks new malware uses (YARA) • HASH CHECK asks 70+ engines about anything suspicious. only the fingerprint leaves my disk, never the file • STARTUP watches what launches on boot and flags anything new • QUARANTINE locks bad files away. it never deletes anything on its own • REPORT writes it all up in plain English last Sunday: 03:00 → scan starts 03:21 → SIGNATURES: setup_crack.exe, known trojan → locked 03:25 → PATTERNS: invoice.pdf.js, a script pretending to be a PDF → locked 03:29 → STARTUP: "updater_helper", unsigned, new on boot → flagged for me 03:41 → 41,902 files checked. 2 threats. report on my screen no green shield. no "you're protected". just what it found, where it came from and what to do all of it is free and open source. Claude set it up in one evening, and i read every step before i approved it it doesn't replace Windows Defender or macOS XProtect. it sits on top of them. keep those on the full guide with the exact prompt is in the article below ↓
7
1
25
901
it's f*cking insane i gave Grok $380 and said "triple this overnight" i woke up to $3,195.80. that's ×8.41 25 trades. 23 wins. 2 losers, both closed by the stop-loss for −$42.60 total at 23:10 i pasted one prompt into Grok Bot, hit run and went to bed how the night went: → 6,842 fresh pairs scanned → 6,650 killed by the filter, 97% → 192 scored, only 75+ got traded → 25 trades, 23 won, 2 stopped out 23:52 → first trade, solana:EkDGB5fbPXiRmDDjxKcC7dFjzvFZj2KT9t7oeyyPx4SX +$29.70 01:08 → $BONKZILLA +$306.40 02:26 → first stop, $SOLSAGE −$17.90. no averaging down, just out 02:56 → x3 hit. $1,151 while i'm asleep 03:03 → $GROKFOX +$522.60 04:11 → $MOONVAN +$684.30, best trade of the night 05:12 → stop #2, $RUGLESS −$24.70 06:08 → $SUNUP +$611.90 06:34 → last trade, $BYTE +$33.40 07:00 → report on my screen $380 → $972 → $1,674 → $2,405 → $3,195 the exact prompt: Run a Solana memecoin night desk. Goal: x3 by 7am. Never break the rules to get there. 5 agents: Scout, Filter, Scorer, Trader, Guard. Scout: new pairs 1h–48h old, liq >$50k, mcap $250k–$15M. Filter kills: mint/freeze not revoked, LP not locked, top 10 holders >25%, dev >4% or linked to past rugs, bundled >10%, already +100% in 1h. Scorer 0–100: volume trend, unique buyers, buy/sell ratio, holder growth, X mentions. Trade only 75+. Trader: 10% of balance per trade, max 3 open. Guard: stop −7% set before entry. Take 50% at 2x, trail the rest by 15%. Log every skip with the reason. Paper mode first. Stop at 07:00 and send me a report. it hit x3 before 3am and didn't stop. same rules, all night i didn't click a single button after 23:10 how to set up the bot is in the article below ↓
11
19
270
it’s f*cking insane the guy who installed my wifi left something on my router. and for 23 days every website i opened went through someone else's server first the wifi was installed. he spent 40 minutes "setting it up". said it's for faster support. i said thanks and never thought about it again then i read my own article about hidden cameras on wifi and decided to check my own router with claude i exported the router config and an hour of traffic, gave both to claude in read-only mode and asked one question: what on this router is not mine? the first answer made me close the laptop for a minute → remote admin was open to the whole internet on port 8443 → admin password was still the default one printed on the box → a "remote support" app i never installed, checking in to an unknown server every 5 minutes → my DNS was switched to their servers. every site i opened was looked up through them first → a port forward straight to my home storage with all my photos 23 days. 6,624 check-ins. 41,280 DNS lookups logged on a server i don't control i don't know if anyone actually looked. but they could. my bank, my emails, what i search at 2am. all of it went through a box someone else had the keys to what i did that night: > factory reset the router > updated the firmware from the manufacturer's site, not from the "support" link > new admin password, remote admin OFF, WPS OFF, UPnP OFF > DNS back to one i chose myself > deleted the port forward > new wifi password and checked every connected device > called the provider and asked what exactly their guy installed. they're "checking" check yours tonight. log into your router, look at remote management, DNS and port forwarding. if there's anything you didn't set, it's not yours the full router check with claude is in the article below. takes 15 minutes
68
243
14
1,496
223,006
IT'S F*CKING GOLDMINE Claude + jev just built a whole pump.fun terminal in 4.1 seconds?? i typed one sentence: "build me a pump.fun terminal that trades for me" Claude wrote the code. jev read the internet. they named it DUOX. and it started trading 4 SOL → 22.6 SOL. 5.65× in 21 seconds who built what: Claude → → the bonding curve feed → the dev wallet + bundle scanner → auto entry, auto exit → the rug guard → the whole terminal UI jev → → the X + web trend reader → reads every coin's story before it buys → kills anything with a copied name or fake narrative every coin climbs the curve toward graduation. Claude checks the wallets, jev checks the hype. both have to agree or it skips what it did in 21 seconds: → $CLAUDEFROG · 9K posts, dev 4% → buy → +0.84 SOL → $DEVRUN · dev holds 28% → skip → rugged 2s later → $NIGHTCAT · graduated → +1.47 SOL → $COPYPEPE · fake narrative (jev) → skip → rugged → $MOONDOG · graduated → +2.35 SOL → $BUNDLED · bundle 41% → skip → rugged → $GEMLORD · trending #1 → graduated → +3.12 SOL → $SENDOOR · 26K posts → graduated → +4.26 SOL → 7 more wins → +6.88 SOL 14 trades. 12 wins. 2 small cuts: -0.11 and -0.21 SOL 3 coins skipped. all 3 rugged. it was never in a single one the rules they wrote for themselves: 1. dev holds more than 10% → skip 2. bundles hold more than 25% → skip 3. Claude says yes, jev says no → skip 4. dev sells anything → exit now 5. coin graduates → sell everything i didn't write a single line. i didn't pick a single coin one AI that checks the wallets. one AI that checks the story. you're still clicking buy by hand
8
1
38
1,383
it's actually f*cking scary after the hotel room thread i ran the same sweep on my car i was ready to buy. found 3 things that shouldn't be there 👇 claude + jev, engine off, phone in airplane mode, 6 minutes. claude listed every signal in the cabin. jev made one call per signal: → is this a car module? yes/no → is it mine? yes/no → does it move with the car? yes/no → how suspicious? 1-3 17 signals. 14 normal: bmw telematics, key fob, my phone. 3 flagged. → a mic behind the speaker grille. 2.4 GHz burst every 30 seconds → a ble tag behind the vent. not paired to anything i own → a gsm bug in the 12v socket. sim inside, dials out silently none of them were mine. your 10-minute check: ble scan → any unknown tracker that keeps showing up wherever you go 12v / usb → pull every adapter you didn't buy yourself obd-ii port → under the wheel. nothing should be plugged in vents + grilles → flashlight behind every one of them seat rails + under the seats → tape, velcro, small black boxes rf sweep → engine off, phone in airplane mode, then look for spikes found one? don't rip it out. photo everything where it sits. check your phone and accounts from a different device. then call the police. a planted tracker is evidence. hotels are where people expect to be watched. your car is where nobody checks. full method in the article below 🕷️
4
5
39
1,777
claude audited my old iphone for 2 minutes and what it found is f*cking insane not a hack. not a leak. just permissions i gave away and forgot about. i pasted my installed-app list + permissions into claude and asked one thing: "which of these has no reason to hold what it holds?" it came back with: → a flashlight app with access to ALL my photos → a qr scanner with microphone + location "always" → a pdf reader running in the background overnight → 4 apps i haven't opened in over a year, still allowed to track me → one "system sync" app with a generic name and no clear maker 31 apps had photo access. i needed 6. i never "allowed" any of it on purpose. i tapped "ok" on a popup in 2022 and never looked again. went from 31 apps to 6 in about two minutes. the prompt is read-only: it can't delete, change or send anything. it just reads your list and ranks every app worst-first, with one plain line on what it can see. rules i gave it: · only MY device · never ask for passwords or 2fa codes · screenshot first, remove after · change passwords from a different device save this. send it to anyone with a phone. your device only. not anyone else's.
12
10
1
145
10,693
i scanned my own apartment's wifi with claude and found 3 devices i never bought 2 cameras. 1 microphone. all uploading 24/7. it took 4 minutes. no special hardware. just 3 commands: → nmap -sn 192.168.1.0/24 lists everything on your network. nothing is changed → nmap -sV -p 80,554,8000 [unknown ip] checks which doors that device has open. port 554 = live video stream → claude "what is this device?" paste the output. it reads vendor + ports + traffic together and gives a verdict here's what gave them away: · camera-maker prefix on a device i never bought · port 554 open (a phone has no reason to) · steady upload at 3am when the flat was empty it's f*cking scary how normal this looks on a router page. one unknown name in a list of 9. if you've ever rented a place, had a guest, a repair or a "free smart plug", do this once. it's free. the whole guide is in the article. step by step, with a floor plan and a checklist of where to look first. only scan networks you own. this is defense, not a hacking tutorial. save it. run it tonight.
8
19
1,467
it's fucking insane everyone just got their iphone duo delivered this week and i guarantee 99% of people have no idea what's actually happening inside when they open it so i made claude break it down. full 3d, step by step, every number from one physics model, with sound closed it's just a normal 5.4" phone, 11.3 mm thick open it and boom: a 7.6" screen, almost 2x the area, in a body only 5.2 mm thin. thinnest iphone ever but the real question is how do you fold glass without shattering it → bending stretches the surface by thickness ÷ (2 × radius) → regular 0.6 mm phone glass would stretch 11.5% on the fold. instant crack → the duo uses ultra-thin glass. 0.58% stretch. nothing happens → the hinge has 100+ parts and drops into a pocket when you close it, so the screen curls into a waterdrop instead of a sharp crease → and the screen layers aren't glued solid. glued they'd stretch 5.8%. instead they slide over each other on a special adhesive, so each layer bends on its own that's literally the whole trick. thin layers + room to curl = no cracks and that's not even all of it: → 2 batteries, one in each half, 4,883 mAh total, power flowing across the hinge to keep both sides level → a20 pro on 2 nm sitting on a vapor chamber, 35% better sustained performance than the 17 pro → touch id in the side button, works closed or open → the selfie camera is hidden under the screen a year ago you'd need a motion designer, a 3d artist and weeks in after effects for this now it's one claude subscription and an evening i wrote the full guide on how to make animations like this yourself, it's in the quoted post below
6
9
1,580
it's f*cking insane read this article, gave grok bot one task before bed, woke up to $37 → $2,570. 77 trades overnight. only 2 losers. and it wasn't luck. it was this exact method. most people tell grok "make money on memecoins" and pray. i told it to break itself first. the task was basically: "run a failure mode sheet on your own desk. list every way each bot can die. fix the worst one first. then trade." what grok came back with before placing a single order: → the gambler bot was sizing off the last trade. classic martingale in a hoodie. killed it → position size is now a fixed fraction of the current book. never of the last result → contract checks became hard gates, not scores: - mint authority not renounced → reject - lp not locked → reject - top-10 holders too concentrated → reject → safety factor: it runs at half of what it thinks the max is → blast radius: one bad coin can only ever take one position, never the book then it traded. the night: 00:30 → first gate rejects 11 fresh launches in a row. zero trades 01:10 → first entry that passes every check 03:00 → 40+ trades in, still only small sizes 06:45 → 77 trades closed 08:00 → $37 → $2,570 the 2 losers? both small. both stopped exactly where the sheet said they should. that's the whole point of the article. every thread shows you the win. nobody shows you how the bot breaks. find the failure mode before the market does. then let it run.
4
273
i couldn’t f*cking believe this the automatic gearbox was always black magic to me. no clutch pedal, no sliding gears, it just... shifts so i asked claude to show me. it built the whole thing in 3D torque converter with oil flowing pump → turbine. two planetary gearsets. two clutches, two bands, a valve body, a park pawl here’s the crazy part you put it in D and you SEE the turbine stop dead while the oil keeps churning around it. that’s literally why an automatic creeps forward when you lift off the brake then the shifts: hold the rear carrier → 1st, ×2.42 a band grabs the sun → 2nd, ×1.42 a clutch locks it all together → 3rd, 1 to 1 no pedal. no lever. just oil pressure choosing which part of the gears to hold a live table on the side lights up exactly which clutch or band is working in every gear. the numbers come from the real gear math: sun 30 teeth, rings 72, planets 21 then it rendered it as a 60fps video with its own soundtrack, engine pitch synced to the rev counter i’ve driven an automatic for years and only today understood what’s inside it this used to be a week of work for a motion designer. now it’s a claude subscription and one question full guide to build your own below 👇
1
6
978
That’s fucking insane claude can now build a full 3D animated machine from one question, and motion designers aren’t needed for this anymore i asked it how a manual gearbox works. it came back with a real cutaway: pistons on a crank, clutch, 3 shafts, gears with the actual tooth counts, synchro sleeves, a lever in an H-gate you press “shift into 2nd” and watch the sleeve slide, the brass ring glow, the speed gap count down 643 → 0 rpm, and only then the teeth click in every number on screen comes from real math. 41 teeth turning 17 = ratio 3.67 → 24.8 km/h at 3,000 rpm then it rendered the whole thing as a 60fps video with its own soundtrack. the engine sound’s pitch follows the rev counter now i actually know how a manual gearbox works and it doesn’t stop at gearboxes. you can break down ANY mechanism like this with nothing but a claude subscription: locks, watches, pumps, engines, your own product no after effects. no blender. no 3 weeks of keyframes the full guide to build your own is below 👇
1
4
336
this is getting stupid: i plugged claude + jev into my vw and now it lives in the car. not a phone holder with chatgpt open. an actual assistant that knows the car, my calendar and my day. the setup is way simpler than it sounds: → a raspberry pi under the seat → a cheap obd-ii bluetooth dongle (read-only) → a usb mic + the car speakers over bluetooth → my phone's hotspot for internet that's it. no hacking the car. nothing touches the brakes, steering or the engine. it only reads. the split is the same as in my article: claude is the brain. it talks, explains, plans, writes. jev makes the small calls in milliseconds, on every single thing the car or i say: - is this worth interrupting me while i'm driving? yes/no - how urgent? 1-3 - what is it? warning, message, calendar, nav, small talk code does the rest. reads the sensors, does the math, keeps the logs. what it actually does on a normal day: → reads the obd data and tells me in plain english what a warning light means, not a code → "tyre pressure rear left dropped 0.3 bar in 2 days, slow leak, get it checked this week" → reads my calendar and says when i actually need to leave, with traffic → summarizes messages out loud, only the ones jev marks urgent → drafts replies by voice, i say "send" or "skip" → logs every trip: fuel, km, cost, so taxes take 2 minutes at the end of the month → remembers stuff: "you said you'd buy oil filters after work" when i pass the shop the rules i hardcoded, because it's a car: - everything is voice only. no screens to look at while driving - if jev says urgency < 2, it waits until i'm parked - it never controls anything. it reads and it talks - calls and messages only with my "send" the best part: it stopped being a chatbot. it's just there, like a co-pilot that never gets tired and actually reads the manual. full claude + jev setup is in my article. this is just the same brain on 4 wheels.
2
350
Claude + Jev just found why my brother's trading bot was bleeding money every night It wasn't the strategy. Someone was inside it. His bot "worked" for 3 weeks. Small wins in the day. Small losses at night. Balance slowly going down. He thought it was just a bad market I opened his repo, gave Claude Opus 5.5 one prompt and let Jev triage every file 6 minutes later: → 412 files scanned → 398 normal → 11 with sloppy risk logic → 3 that should never have been there The 3 files: → a "free signals" plugin he copied from a Telegram group → an extra dependency nobody ever installed on purpose → a cron job running at 03:00 every night Jev flagged them in milliseconds. Claude pulled them apart. Inside: → the "signals" plugin sent his exchange API key to an outside server on first launch → his key had withdrawals ENABLED → the 03:00 job placed tiny market orders on illiquid pairs, then reversed them → every night a few % of his balance quietly leaked through bad fills → stop-losses were placed AFTER the order, not with it → fees and slippage were never in his backtest His "strategy" was never the problem We killed the keys, made new ones with withdrawals OFF and IP whitelist ON, deleted the plugin, rewrote the risk layer and moved everything into one CLAUDE.md with hard rules Next night the balance stopped bleeding Your 3-minute check tonight: api keys → withdrawals must be OFF. always ip whitelist → if your exchange has it, turn it on plugins → anything copied from a chat or a "free signals" group, read it before you run it cron jobs → if you don't know why it runs at 3am, it shouldn't stops → placed WITH the order, never after backtest → no fees and slippage = it's lying to you Found something? Don't just delete it. Revoke the keys first, from the exchange, from another device. Then clean the repo The full Claude + Jev setup is in my article above Save this. Check your bot tonight. Then check your friends' bots
3
566
this is f*cking insane: a b2b company just fired its entire sales manager team and replaced it with claude + jev. no sdrs. no linkedin scrolling. no "just bumping this up" emails. the agents now find who to write to, decide if it's worth it, and write the email themselves. the old setup: → 6 managers → the same template sent to 300 people a day → 2% reply rate → $40k+ a month in salaries the new one is basically 3 layers: claude reads everything. company site, news, job posts, changelogs, funding. then writes an email that actually mentions their situation. jev makes the calls. one request per lead, every question at once: - does this company fit? yes/no - how strong is the signal? 1-3 - which angle? cost, speed, scale - who exactly? founder, head of ops, cto code runs the pipeline. crm, send caps, follow-up timers, unsubscribes. the trick is the confidence split: > 0.8 → claude writes, email goes out 0.5-0.8 → claude writes, a human taps approve < 0.5 → lead goes to a watch list, nobody gets spammed and the emails don't read like ai. short, specific, one ask: "saw you're hiring 3 ops analysts. we automate that exact workflow. worth 15 min?" then jev reads every reply. interested → meeting link + a human gets pinged. unsubscribe → gone forever, instantly. the rules they hardcoded: - only public business info - daily send caps - pricing, contracts, discounts → always a human result: 6 managers became 1 person who only closes deals. the pipeline runs while everyone sleeps. sales was never about sending more emails. it was about the right email to the right person at the right time. an agent does all 3 now.
1
2
292
IT'S F*CKING GOLD A 19 Y/O OXFORD STUDENT PLUGGED HIS BRAIN INTO CLAUDE + JEV 🤯 Not a chip. Not Neuralink. His actual brain: every note, every lesson, every losing trade he ever wrote down. He turned it into a trading bot. First night: $62 → $4,287. Here's exactly how he did it 👇 ━━━━━━━━━━━━━━━ THE BRAIN For 2 years he kept a trading journal in plain .md files. 📁 notes/ → one file per setup 📁 markets/ → how each coin behaves 📄 MEMORY.md → his rules that never change 📄 decisions.md → every trade, every reason, every mistake 📄 LEARNINGS.md → what the losses taught him Most people trade on vibes. He had 700 pages of himself. ━━━━━━━━━━━━━━━ THE STACK 🧠 CLAUDE → reads the brain, writes the bot turned 2 years of notes into code, tests and a backtester ⚡ JEV → makes the calls the way HE would every signal checked against his own rules in milliseconds ⚙️ CODE → sizing, fees, stops the math never touches a model ━━━━━━━━━━━━━━━ HOW HE BUILT IT 1/ CLAUDE.md as the map tells Claude where every rule, setup and lesson lives Claude Code loads it on every session, so it never starts blind 2/ Claude turned his notes into rules "I lose money on breakouts during news" → a hard filter "funding flips work on SOL, not on BTC" → a setup file 3/ Jev became his instinct every signal, one call, all questions at once: → does this match one of MY setups? → did this exact pattern fail before in decisions.md? → is this the kind of trade I always regret? 4/ Confidence decides > 0.8 → trade 0.5–0.8 → paper only < 0.5 → skip, like he would 5/ Backtest on his own history fees, slippage, walk-forward only setups that survived went live 6/ Risk in code 1% per trade · stop with every order daily loss limit · keys trade-only, withdrawals OFF ━━━━━━━━━━━━━━━ THE NIGHT 00:00 → bot live, he goes to sleep 03:40 → first funding-flip setup fires 06:15 → three winners, one small stopped loss 08:00 → he wakes up $62 → $4,287. ━━━━━━━━━━━━━━━ THE REAL LESSON The edge wasn't Claude. It wasn't Jev. It was 2 years of honest notes about his own mistakes. The AI just made sure he never repeated them. Start writing your brain down today 🕷️
3
412
IT'S F*CKING INSANE Everyone's still duct-taping Claude + Jev together by hand. Meanwhile there's an open-source repo that already did it. QuantDinger. 13,000 stars. Apache-2.0. Free. An AI trading OS with Jev built in 👇 ━━━━━━━━━━━━━━━ WHAT IT IS 🧠 an open-source "AI Trading OS" 📈 research, Python strategies, backtests 🤖 agent trading in one place ⚡ Jev System One integration out of the box The decision layer from my article? Already wired in. ━━━━━━━━━━━━━━━ WHY IT FITS THE SETUP PERFECTLY In the article above the split is simple: Claude builds. Jev decides. Code does the math. QuantDinger gives you the "code does the math" part: the strategy engine, the backtester, the research tools. So Claude doesn't start from an empty folder. It starts from a working trading OS. ━━━━━━━━━━━━━━━ HOW I'D USE IT 1/ clone it, open it in Claude Code 2/ let Claude read the codebase and map it into a CLAUDE.md 3/ point the Jev layer at your signal filters and risk checks 4/ write your strategy in Python, Claude writes the tests 5/ backtest with fees and slippage, walk-forward, sealed hold-out 6/ paper trade 2–4 weeks before a single real dollar ━━━━━━━━━━━━━━━ THE PART PEOPLE SKIP Pin the Jev version. Keep all questions in one file. Confidence < 0.5 → you decide, not the bot. API keys: trading ON, withdrawals OFF. Every rule from the article still applies. The repo just saves you weeks of building. ━━━━━━━━━━━━━━━ github.com/OpenByteInc/Quant… Read the article. Clone the repo. Build this weekend 🕷️
5
677