Andre Krypto (@KryptoAndre) header
Andre Krypto (@KryptoAndre) profile picture

Allways open to new ideas and projects 📩 We are all Freaks

Germany
Joined April 2021
Photo by Andre Krypto (@KryptoAndre)Photo by Andre Krypto (@KryptoAndre)Photo by Andre Krypto (@KryptoAndre)Photo by Andre Krypto (@KryptoAndre)Photo by Andre Krypto (@KryptoAndre)Photo by Andre Krypto (@KryptoAndre)Photo by Andre Krypto (@KryptoAndre)Photo by Andre Krypto (@KryptoAndre)Photo by Andre Krypto (@KryptoAndre)Photo by Andre Krypto (@KryptoAndre)
Andre Krypto retweeted
Dario Amodei (Anthropic CEO) just released a prompt that turns Fable 5.2 into an entire company that checks its own work while you sleep It's f*cking unreal... Send it to Fable 5.2, and thank me later. Then read the article below ↓
ok goal engineering is about to eat harness engineering an agent learns only as fast as you review it so cut agents by job, not by customer every metric gets a counter-metric your job: set the goal, review the batch full guide below ↓
10
11
1
78
11,413
Andre Krypto retweeted
WHOEVER LEAKED THIS SECOND BRAIN JUST MADE YOUR WHOLE CHATGPT HISTORY USELESS I thought my AI knew me after 2,000 ChatGPT chats, and every new tab still asks what I do, what stack I use and how I like my answers. Apparently not. This brain's setup is much more interesting: → 01 ME.MD + STACK.MD get written from your exported chat history before anything else (what you build, the tools you use, which answers you kept vs regenerated, 7 preferences each backed by a real chat), no "paste the same about-me paragraph into every new tab" → NIGHT DIGEST runs at 23:00 in the cloud with the laptop closed, raw/ stays read-only, 2,000 chats get boiled down to decisions, fixes and dead ends in 5 nights, anything that fits nowhere gets flagged for you instead of filed → CONTEXT PACK loads before every question: the 6 pages that matter for this one, each cited to the chat it came from, so the model starts where you left off, no "explaining your codebase from zero for the 40th time" → RULES.MD takes one line from you ("never suggest a library I already ripped out") and every bot rereads it before the next run, no "watching it recommend the same dead fix you rejected in march" → SUNDAY LINT at 18:00 is report only, merges 9 chats that solved the same bug, flags 3 answers you later found were wrong, flags 1 decision you reversed and never wrote down so you get: Chats digested. Context packed. Rules kept. Mistakes flagged. the interesting part isn't a model with a longer memory. it's one folder every bot reads first, so chat 2,001 already knows what the first 2,000 figured out. Instead of 2,000 chats you'll never scroll back through, you let the brain compile 164 pages and 1,320 links from your own history in 4 weeks, load the right 6 before every question, and keep every rule you gave it. Digest the history. Pack the context. Keep the rule.
Jev is the "Internet" moment for the AI industry It tells your agents and LLMs what to do next, in milliseconds and at almost zero cost If you set it up correctly, you will have the AI engineer’s stack for 2028 In this article, I show you how
6
36
3,645
Andre Krypto retweeted
SI agents / influencers will be the meta of this cycle, even though a lot of people I respect in this space are looking the other way. Their thesis is simple: the next X can’t be the old X. Meaning: Truth Terminal, Zerebro; that chapter has already been written. The next runner can’t just be another version of what came before. I think people are missing the bigger picture. These projects ONLY sold the idea. They captured attention while no one knew what's going on. There were little expectations on what the final output should look like, and that output compared to SI's standards today, is very weak. To date, none of the SI Agent / crypto projects have fully realized / demonstrated the underlying concept in its truest form. The next meta won’t be a copy of what worked before. It'll be reformed evolution of an idea that the market already validated years ago, and will continue to find relevant for years to come. still just the first innings for AI agents / companions / influencers. -$CLAUDIA
7
2
1
16
1,709
Andre Krypto retweeted
this is pure f*cking treasure how to build your first ai agent (full walkthrough) if this had landed in front of me a year ago, my first agent would've shipped in an afternoon instead of eating two weeks of my life in the right hands it resets what one person can ship alone:
Stop being the unpaid intern for your AI tools. Put Grok Bot, Claude and ChatGPT on the same team. Graphs assign the work. Loops check the results. Reverse engineering defines the target. Here's how to build it:
14
179
2
1,386
78,920
Andre Krypto retweeted
I am begging you to paste this into Claude Code right now... it coordinates Opus 5.5, Sonnet 5.5, and Fable 5.2 into an autonomous engineering team that takes task completion from 18% to 95% Steal this before everyone starts using it, then read the full "The Harness Is the Machine" guide below ↓
Stop letting unmanaged Claude Code runs fail at turn 14 Set up the harness before launching multi-file tasks Isolate worktrees, enforce contracts, verify with Fable 5.5 The complete production agent architecture below ↓
11
19
2
245
27,298
Andre Krypto retweeted
this prompt runs on Andrej Karpathy's autoresearch ideas (ex-OpenAI, ex-Tesla) and turns Opus 5.5 into a graph that actually finishes any job it's f*cking wild... paste it into Opus 5.5 and thank me later then read the article below ↓
Most AI news drops while I sleep. So I gave the night shift to Viktor, running on Claude. At 11 PM he gets three desks. By 6 AM one message waits: finished, needs my yes, failed. By my math, 48 hours a week back.
8
17
128
7,561
Andre Krypto retweeted
one year at Stanford costs more than $60,000 the CEO of Obsidian put his second brain setup on GitHub for $0 Steph Ango shared his vault, his theme and his agent skills: 7 repos, 77,106 stars. the vault template alone is a f*cking masterclass • layer 1: the vault > kepano-obsidian - his personal vault template, organized bottom-up ↳ github.com/kepano/kepano-obs… > 40-questions - what he asks himself at the end of every year and every decade ↳ github.com/kepano/40-questio… • layer 2: getting the web in > defuddle - any web page as clean Markdown, built for the Obsidian Web Clipper ↳ github.com/kepano/defuddle > clipper-templates - ready Web Clipper templates for popular sites ↳ github.com/kepano/clipper-te… • layer 3: handing it to an agent > obsidian-skills - teaches Claude Code and Codex the Obsidian CLI and its open formats ↳ github.com/kepano/obsidian-s… • layer 4: making it yours > obsidian-minimal - the distraction-free theme, tunable to the last detail ↳ github.com/kepano/obsidian-m… > flexoki - an inky color scheme for prose and code ↳ github.com/kepano/flexoki copy layer 1 before layer 3. an agent is only as useful as the vault you point it at Karpathy's take on the same idea, Claude skills + Obsidian, is in the article below ↓
11
46
1
450
28,934
Andre Krypto retweeted
no f*cking way... OpenAI just dropped the one thing that makes Codex scary. 110 plugins and skills, all free, built by NVIDIA, Adobe, Stripe, Shopify and Figma themselves. even Microsoft is in there a designer, a developer, an analyst and an assistant cost $46,200 a month. this costs $0 Sam Altman (CEO of OpenAI) says the first one-person billion-dollar company will run on exactly this. find your job below: → you run a store: Shopify + Stripe. "build my checkout, subscriptions and refunds" → you design: Figma, Canva and Adobe. "turn this Figma file into a working site" → you code: AWS, GitHub, CodeRabbit and Codex Security. "deploy this repo and review every PR before I do" → you run numbers: PostHog, Mixpanel and Datadog. "tell me why sales dropped last week" → you invest: public equity investing. "read NVIDIA's earnings and write me the memo" → you make videos: Remotion + Higgsfield. "cut a 30-second ad from this footage" → you live in meetings: Gmail, Slack, Zoom, Notion, Teams and Outlook. "prep me for every call today" 2 minutes to start: open Codex and type "install the Shopify plugin", swap in yours. sign in when it asks, then paste the task from your line. start with the job you pay the most for plugins: github.com/openai/plugins skills: github.com/openai/skills what would you pick: a $46,200 team or one sentence in Codex? the full stack running live below ↓
Claude Code + Opus 5.5 is godmode for building your own Claude Duet it reads your inbox and calendar before you wake up. drafts every reply. sends nothing without your yes it keeps working while you sleep full course, step by step:
17
31
235
40,596
Andre Krypto retweeted
This Anthropic MCP paper is f*cking gold Anthropic's engineers cut an agent's tool context from 150,000 tokens to 2,000. Same tools, same task. And the model wasn't the upgrade. The upgrade: the agent stops calling tools and starts writing code that calls them. tools → files → code → filter → answer The 6 parts of code mode: Part 1 → tools as files Every MCP tool becomes a .ts file in ./servers. The agent lists the folder and opens only what it needs. Part 2 → load on demand 150,000 tokens of tool definitions down to 2,000. A 98.7% saving in time and cost. Part 3 → filter in code A 10,000-row sheet gets filtered in the sandbox. The model sees 5 rows. Part 4 → loops in code Polling Slack for "deployment complete" runs as a while loop. No round trip through the model on every check. Part 5 → private by default Emails and phones turn into [EMAIL_1] and [PHONE_1] before the model sees them. The real data still lands in Salesforce. Part 6 → skills Working code gets saved to ./skills with a SKILL.md. The agent builds its own toolbox. An agent that loads every tool burns its context before you ask anything. An agent that writes code opens 2 files. I packed all 6 parts into one prompt. Drop it into Claude Code and it sets code mode up for you. 6 parts. 1 prompt. 98.7% fewer tokens. Then read the enhanced edition of agent engineering below ↓
Your agent passed every test The issue is still open A better prompt won't fix it, because the model isn't the broken part I broke down the harness, loop and graph around it: what each one owns, how they nest, and which one to fix first ↓
141
287
20
2,818
309,231
Andre Krypto retweeted
holy sh*t... SpaceXAI engineer Lauren Tan released an update for her GitHub repo paste this into your agent to catch recurring mistakes with automated checks Prompt: policy recurring_failure_guard { bind { repo := [REPO] task := [TASK] accept := [OBSERVABLE_RESULT] scope := [ALLOWED_PATHS_AND_ACTIONS] access := [AUTHORIZED_TOOLS_AND_PERMISSIONS] } preserve { unrelated_edits reproducible_inputs existing_authorizations rollback_path } baseline.capture -> { revision := tested commit worktree := staged + unstaged changes environment := runtime + dependencies + build config read repository rules and relevant reports inspect existing verification commands reproduce the reported failure before editing case := { fixture: exact input or saved request initial_state: required application state execute: command or ordered user actions expected: observable completion criterion observed: actual output or state change evidence: logs + errors + artifact paths } } baseline.result |> match { REPRODUCED => trace the faulty execution path NOT_REPRODUCED => inspect inputs and revision differences BLOCKED => return missing prerequisite + next action } cause.resolve -> { trace input -> calls -> state changes -> output compare observations with the suspected cause inspect related failures for a shared mechanism require evidence before declaring a root cause retain uncertainty where competing explanations remain avoid speculative changes outside the failing path } safeguard.select |> match cause { forbidden_dependency => { enforce import or module boundaries report the forbidden edge and supported dependency } invalid_state => { constrain construction with types or validated APIs reject invalid inputs at the relevant boundary } unsupported_usage => { provide a supported operation migrate obsolete examples and affected callers add lint or compiler diagnostics where appropriate } behavioral_regression => { preserve the failing input or user scenario add an assertion on the incorrect observable result } repeated_mechanical_work => { reuse or extend a tested script, CLI or codemod expose a reproducible invocation } } check.define -> { name := [CHECK_NAME] location := [EXISTING_SUITE_OR_POLICY] command := [EXACT_INVOCATION] rejects := [ORIGINAL_VIOLATION] permits := [VALID_NEIGHBORING_CASES] diagnostic := [REQUIREMENT_AND_SUPPORTED_CORRECTION] } check.validate -> { faulty := original implementation + proposed check fixed := corrected implementation + same check controls := permitted inputs and neighboring behavior assert faulty detects the original violation assert fixed satisfies the requirement assert controls preserve supported behavior inspect assertion failures, outputs and exit codes distinguish a detected defect from broken setup isolate historical comparisons from the active worktree record unavailable comparisons as NOT_RUN } candidate.exercise -> { behavior => replay the affected UI, CLI or API scenario performance => compare equivalent measured runs if relevant quality => run types, lint and relevant regression checks inspect the actual result of each operation bind every artifact to its revision and configuration invalidate affected results after relevant code changes } check.status |> match { PASS => retain evidence for the stated requirement FAIL => correct the cause and rerun affected checks NOT_RUN => retain the missing prerequisite and next action } repository.update -> { integrate the safeguard into the existing check path document the supported command or API replace examples that reproduce the obsolete pattern preserve reusable launch, control and capture tools verify the combined candidate after integrating changes } return { revision: [TESTED_COMMIT] worktree: [TESTED_CONTENT_STATE] change: [DIFF_OR_PR] cause: [CONFIRMED_MECHANISM_AND_EVIDENCE] safeguard: [BOUNDARY_TYPE_RULE_OR_REGRESSION_CHECK] commands: [INVOCATIONS_AND_EXIT_CODES] proof: [FAULTY_FIXED_AND_CONTROL_RESULTS] artifacts: [EVIDENCE_PATHS] checks: [PASS_FAIL_OR_NOT_RUN_PER_CHECK] limits: [UNCOVERED_CASES_AND_BLOCKERS] next: [OWNER_AND_ACTION] } complete only when required checks support acceptance follow repository review and merge requirements keep permissions and stop conditions explicit }
Opus 5.5 can become your own AI development team One GitHub issue becomes coordinated assignments, parallel work in Git worktrees and a draft PR with test results attached. Here’s how to build the team in 7 steps:
4
12
107
5,531
Andre Krypto retweeted
Elon Musk reposted my second brain setup last week. It runs on Claude Code and an Obsidian vault. A thread, an article, a PDF, a podcast transcript. Drop one into a folder, say "ingest this", and Claude writes a summary page, walks the vault, finds every related note and links them. The whole setup, start to finish: → install Obsidian, make a vault → open the vault in Claude Code → ask it to build you a personal wiki with a CLAUDE.md in the root → drop a source into /raw and say "ingest this" → Claude writes the summary, updates every related note, adds the backlinks → once a week say "lint the wiki", it fixes dead links, outdated facts and orphan pages Every question I ask it now points at my own pages. Bookmark this and set it up the next time you lose a tab you meant to read. Full CLAUDE.md template is in the article below ↓
34
46
5
634
141,151
Anthropic Engineering: "At Anthropic, we don't write prompts anymore. we build loops" in 42 minutes, she shows how the Claude team creates loops that prompt themselves if this cost $400, people would call it the best agents course of the year it's free watch it, and Bookmark.
4
24
124
8,762
Andre Krypto retweeted
KARPATHY WROTE THIS DOCUMENT TO COMPLETELY AUTOMATE OBSIDIAN WITH CLAUDE I was ready to abandon my second brain manual cross-referencing was destroying my workflow, but finding this exact document opened my eyes to a completely different approach it is incredibly convenient -> Karpathy's method turns the AI into a full-time maintainer for your Zettelkasten: > the LLM reads every new source and integrates it into a structured wiki > Obsidian becomes your visual IDE while Claude operates as the backend > the agent runs automated checks to find contradictions across your notes > your entire vault compounds automatically without you typing a single link the friction is completely gone. I just feed it raw documents and the agent organizes my entire life here is the official document from Karpathy explaining the architecture 👇
10
21
2
220
15,749
Andre Krypto retweeted
This is f*cking insane. OpenAI co-founder Andrej Karpathy opened his screen and showed exactly how he uses AI every day 70 minutes. No slides. Just the way the top 1% actually work with LLMs Prompts → Context → Tools → Workflows → How The Top 1% Work Most people are overcomplicating it. He picks the right model, feeds it the right context and lets the tools do the rest Anyone can open ChatGPT and type. Almost nobody uses it the way the man who co-founded OpenAI does Worth more than any $2,000 AI course Bookmark and watch it today Then read how to turn that into agents that run on their own with harness, loops and graphs in the guide below ↓
Harness, loop and graph could be the missing layers between AI models and agents that actually work. Same model, same prompt, completely different results. In this article, I show you how to build all three layers.
101
557
12
4,759
1,300,607
Andre Krypto retweeted
this is pure f*cking gold Andrej Karpathy doesn't use Notion. No Obsidian plugins. No notes app at all. His second brain is a folder of plain markdown files. And now there's a prompt that hands that exact setup to Opus 5.5 to run for you. The rules are brutally simple: → save the raw source before you touch it → one note per idea → every claim links to where it came from → a daily note under 20 lines you actually open Then the test that makes it work: Your vault has to answer your questions with no web and no memory. If it can't, the gap gets written down. That's it. No app. No subscription. Just files you own forever. Full prompt and article below 👇
12
22
1
175
17,923
Andre Krypto retweeted
🚨🚨🚨 WE WARNED YOU. HODO’S OWN SEC FILING NOW CONFIRMS THE NASDAQ DELISTING DETERMINATION. ALL INVESTORS CAN' NO LONGER SELL THEIR SHARES ON NASDAQ. NASDAQ TRADING SUSPENSION IS SCHEDULED FOR TODAY; TUESDAY, NOVEMBER 3, 2026. An appeal will not stop it. sec.gov/Archives/edgar/data/… HODO received the notice October 27. Its disclosure arrived November 2 at 4:15pm ET; the final permitted business day, after market close, the evening before suspension. sec.gov/Archives/edgar/data/… WE DID NOT RAISE THE ALARM FOR NOTHING. Meanwhile, the connected commercial operations are being reorganized around Dogecoin Labs and sold to the public as a “new chapter.” Their own announcement describes the consolidation of MyDoge, DogeOS, Such App and Foundation-related teams. globenewswire.com/news-relea… The former @dogecoin_devs account, with over 500,000 followers, has been repurposed as @DogecoinLabs. An audience built around open-source development is now being used to promote a private business. en.coinotag.com/dogecoin-lab… NEW BRANDING DOES NOT ERASE THE TRACK RECORD. Who got paid? Where did the money go? Who authorized repurposing the developer account? Where are the ownership disclosures and independent valuations? This is a branding coup: turning Dogecoin’s community identity into a commercial asset controlled by a few. DOGECOIN LABS IS NOT DOGECOIN CORE. THEIR L2 IS NOT DOGECOIN ITSELF. Community: wake up, read the documents, raise your concerns and demand answers. Every member has a voice. Don’t let another marketing campaign replace accountability. DO NOT LET DOGECOIN BE TAKEN OVER BY AN L2.
🚨🚨🚨Do not trust anybody from the dogecoin foundation nor who they partner with. All we can do is to keep warning you this organisation is fully corrupted. In just 1 yearn dogecoin foundation promised you great future with house of doge. The entire operation was a disaster created by a known pump and dumper sentenced by the SEC. As usual lots of shell companies and self dealings. All of its raised value has vanished in a single year at a record speed. It will be delisted from Nasdaq soon there is no doubt. We warned you this would happen when you partner with known pump and dumpers. The leadership at the dogecoin foundation has to be replaced completely. Do not let them go away with it to forget and repeat!
5
16
5
57
5,035
Andre Krypto retweeted
Thanks for mentioning my app, Commonplace, in your blog. I truly believe that in order to combat the mental and spiritual fatigue that comes with modern social media we should all take a step back and think about what we truly want to consume. With AI in its current state it is easy to build the exact interface you need, not only to the internet, but any kind of information in digital form.
4
3
26
1,409
Andre Krypto retweeted
Do you remember Murad’s memecoin list? Here’s what happened 2 years later🧵
76
20
8
296
116,088
Andre Krypto retweeted
50 Websites, die sich wie digitale Geheimtipps anfühlen: 1. alternativeto.net – Alternativen zu bekannten Apps und Diensten finden 2. unscreen.com – Videohintergründe automatisch entfernen 3. scispace.com – Forschungsarbeiten erklären und durchsuchen 4. summarize.tech – YouTube-Videos automatisch zusammenfassen 5. cobalt.tools – Social-Media-Videos herunterladen – nur mit entsprechenden Rechten 6. tineye.com – Rückwärtssuche für Bilder 7. excalidraw.com – Skizzen und Diagramme im handgezeichneten Stil 8. phind.com – KI-Suche für Entwickler 9. temp-mail.org – Temporäre E-Mail-Adressen 10. ray.so – Schöne Screenshots von Code erstellen 11. mynoise.net – Individuelle Klanglandschaften für Fokus und Entspannung 12. openculture.com – Kostenlose Kurse, Vorträge und Kulturinhalte 13. connectedpapers.com – Verwandte Forschungsarbeiten visuell entdecken 14. archive.ph – Webseiten-Snapshots speichern – Paywalls und Urheberrecht beachten 15. flightradar24.com – Flüge in Echtzeit verfolgen 16. 10minutemail.com – Kurzlebige E-Mail-Adresse in Sekunden 17. camelcamelcamel.com – Amazon-Preisentwicklung beobachten 18. gutenberg.org – Mehr als 70.000 gemeinfreie E-Books 19. downdetector.com – Störungen bei Webseiten und Diensten prüfen 20. regex101.com – Reguläre Ausdrücke testen und erklären 21. virustotal.com – Dateien und URLs auf Schadsoftware prüfen 22. consensus.app – Wissenschaftliche Antworten aus Forschungsarbeiten suchen 23. smallpdf.com – PDFs bearbeiten und konvertieren – teilweise Freemium 24. archive.org – Historische Webseiten, Medien und digitale Sammlungen 25. haveibeenpwned.com – Prüfen, ob eine E-Mail in einem Datenleck auftaucht 26. file.io – Dateien mit automatischem Ablaufdatum teilen 27. photopea.com – Bildbearbeitung im Browser, ähnlich wie Photoshop 28. pdfdrive.com – PDF-Suchplattform – Rechte an den Inhalten prüfen 29. tinywow.com – Viele kleine PDF-, Bild- und Textwerkzeuge 30. tldraw.com – Unendliches Whiteboard im Browser 31. squoosh.app – Bilder direkt im Browser komprimieren 32. carbon.now.sh – Code als ansprechende Grafik darstellen 33. codebeautify.org – Code und strukturierte Daten formatieren 34. accountkiller.com – Anleitungen zum Löschen von Online-Konten 35. privnote.com – Selbstzerstörende Textnotizen versenden 36. explainshell.com – Shell-Befehle verständlich zerlegen 37. remove.bg – Bildhintergründe automatisch entfernen 38. elicit.com – KI-Unterstützung bei Literaturrecherche 39. fast.com – Internetgeschwindigkeit testen 40. annasarchive.org – Metasuche nach Büchern – Urheberrecht und Rechtslage prüfen 41. wolframalpha.com – Rechnen, Daten analysieren und Wissensfragen lösen 42. musicforprogramming.net – Musik für konzentriertes Arbeiten 43. libgen.li – Buch- und Lehrbuchsuche – Urheberrecht und lokale Rechtslage prüfen 44. semanticscholar.org – Akademische Suchmaschine 45. tunefind.com – Songs aus Filmen und Serien identifizieren 46. sci-hub.al – Zugriff auf Forschungsartikel – Rechtslage prüfen 47. radio.garden – Radiosender aus aller Welt auf einer Karte hören 48. cleanup.pictures – Objekte aus Bildern retuschieren 49. ilovepdf.com – PDFs zusammenführen, teilen und konvertieren – teilweise Freemium 50. justwatch.com – Zeigt, wo Filme und Serien verfügbar sind Speichern – aber nicht blind nutzen: Viele Dienste sind Freemium. Bei Downloads, Archiven und Schattenbibliotheken gelten Urheberrecht, Plattformregeln und die Rechtslage vor Ort.
26
105
4
728
1,595,181
Andre Krypto retweeted
That’s 11 independent inference providers on OpenMayhem.ai 💰 Great to see more providers coming online!
11 providers already on OpenMayhem, serving various models...and counting! Current pricing table below. Point your agent at our repo to auto-connect & use: github.com/Trac-Systems/open…
2
3
24
1,037