Melanie 💖 @melafpgt

Running several coding agents gets messy fast. This terminal makes their work visible. GhosttyEXTREME is a macOS terminal fork for developers running AI coding agents across multiple projects. It helps you follow what each agent is doing by showing live status, the files it reads and edits, its project context, and the changes it leaves for review. Key features: • Agent sidebar – see each agent’s task, status, timer, and last action across tabs • Live code following – an editor opens files agents read and animates their edits as they happen • Change review + undo – inspect an agent’s diff, comment on lines, or restore a snapshot from before its turn • Cross-agent hand-offs – pass a task, recent conversation, diff, failed commands, and project memory to another session • Agent races – run the same task in separate git worktrees and compare the resulting diffs It’s open-source (AGPL-3.0 license). Link in the reply 👇

EXO-K 엑소케이 'History' MV (Korean Ver.) hits 100M views! Thank you EXO-L for your love and support🤍 youtu.be/vdejiaoEhFc #EXO #엑소 #weareoneEXO #History #EXO_History #MAMA

Good night everyone As the evening shadows deepen and the world slows down, let a warm, comforting stillness wrap around your heart. Exhale all the lingering rush of the day, allowing yourself to settle into a space of pure rest and quiet comfort. As you lay your head down tonight, may God’s gentle mercy hush every anxious thought and His perfect peace watch over your mind. You can drift off to sleep with absolute confidence, knowing you are held securely by His infinite love and timeless faithfulness. May His protective grace blanket your home, shielding you and your loved ones through the dark hours. Sleep deeply and beautifully, letting the divine quiet restore your body and refresh your soul so you can awaken to a peaceful, blessed new day. Good night, and may the Lord bless you always.

𝐓𝐑𝐔𝐌𝐏 𝐀𝐃𝐌𝐈𝐍 𝐇𝐀𝐋𝐓𝐒 𝐌𝐄𝐃𝐈𝐂𝐀𝐈𝐃 𝐅𝐔𝐍𝐃𝐈𝐍𝐆 𝐓𝐎 𝐌𝐈𝐍𝐍𝐄𝐒𝐎𝐓𝐀 𝐎𝐕𝐄𝐑 𝐅𝐑𝐀𝐔𝐃 — 𝐎𝐙 𝐑𝐄𝐕𝐄𝐀𝐋𝐒 𝟔𝟓𝟎,𝟎𝟎𝟎 𝐃𝐔𝐏𝐋𝐈𝐂𝐀𝐓𝐄 𝐄𝐍𝐑𝐎𝐋𝐋𝐄𝐄𝐒 The Trump administration announced it is 𝐟𝐫𝐞𝐞𝐳𝐢𝐧𝐠 𝐟𝐞𝐝𝐞𝐫𝐚𝐥 𝐌𝐞𝐝𝐢𝐜𝐚𝐢𝐝 𝐩𝐚𝐲𝐦𝐞𝐧𝐭𝐬 𝐭𝐨 𝐭𝐡𝐞 𝐬𝐭𝐚𝐭𝐞 𝐨𝐟 𝐌𝐢𝐧𝐧𝐞𝐬𝐨𝐭𝐚 until state officials prove they're actually policing fraud, a top health official said at a briefing this week. “𝘞𝘦'𝘳𝘦 𝘢𝘯𝘯𝘰𝘶𝘯𝘤𝘪𝘯𝘨 𝘵𝘰𝘥𝘢𝘺 𝘵𝘩𝘢𝘵 𝘸𝘦 𝘩𝘢𝘷𝘦 𝘥𝘦𝘤𝘪𝘥𝘦𝘥 𝘵𝘰 𝘵𝘦𝘮𝘱𝘰𝘳𝘢𝘳𝘪𝘭𝘺 𝘩𝘢𝘭𝘵 𝘤𝘦𝘳𝘵𝘢𝘪𝘯 𝘢𝘮𝘰𝘶𝘯𝘵𝘴 𝘰𝘧 𝘔𝘦𝘥𝘪𝘤𝘢𝘪𝘥 𝘧𝘶𝘯𝘥𝘪𝘯𝘨 𝘵𝘩𝘢𝘵 𝘢𝘳𝘦 𝘨𝘰𝘪𝘯𝘨 𝘵𝘰 𝘵𝘩𝘦 𝘴𝘵𝘢𝘵𝘦 𝘰𝘧 𝘔𝘪𝘯𝘯𝘦𝘴𝘰𝘵𝘢 𝘪𝘯 𝘰𝘳𝘥𝘦𝘳 𝘵𝘰 𝘦𝘯𝘴𝘶𝘳𝘦 𝘵𝘩𝘢𝘵 𝘵𝘩𝘦 𝘴𝘵𝘢𝘵𝘦 𝘰𝘧 𝘔𝘪𝘯𝘯𝘦𝘴𝘰𝘵𝘢 𝘵𝘢𝘬𝘦𝘴 𝘪𝘵𝘴 𝘰𝘣𝘭𝘪𝘨𝘢𝘵𝘪𝘰𝘯𝘴 𝘴𝘦𝘳𝘪𝘰𝘶𝘴𝘭𝘺 𝘵𝘰 𝘣𝘦 𝘨𝘰𝘰𝘥 𝘴𝘵𝘦𝘸𝘢𝘳𝘥𝘴 𝘰𝘧 𝘵𝘩𝘦 𝘈𝘮𝘦𝘳𝘪𝘤𝘢𝘯 𝘱𝘦𝘰𝘱𝘭𝘦'𝘴 𝘵𝘢𝘹 𝘮𝘰𝘯𝘦𝘺”, the official said. Providers on the ground already got paid, the halt targets the state government itself for refusing to clean up its own books. CMS Administrator Dr. Mehmet Oz followed with the numbers that triggered the crackdown. His team cleared a Biden-era backlog and found the duplicate-enrollment problem was 𝐭𝐞𝐧 𝐭𝐢𝐦𝐞𝐬 𝐰𝐨𝐫𝐬𝐞 than expected, jumping from an estimated 100,000 cases to 𝟔𝟓𝟎,𝟎𝟎𝟎 𝐩𝐞𝐨𝐩𝐥𝐞 double-dipping on taxpayer-funded health coverage, about 550,000 of them on Medicaid itself. “𝘕𝘰 𝘰𝘯𝘦 𝘭𝘰𝘴𝘵 𝘵𝘩𝘦𝘪𝘳 𝘪𝘯𝘴𝘶𝘳𝘢𝘯𝘤𝘦. 𝘞𝘦 𝘴𝘢𝘷𝘦𝘥 𝘵𝘢𝘹𝘱𝘢𝘺𝘦𝘳𝘴 $4.4 𝘣𝘪𝘭𝘭𝘪𝘰𝘯, 𝘮𝘢𝘬𝘪𝘯𝘨 𝘺𝘰𝘶𝘳 𝘭𝘪𝘧𝘦 𝘮𝘰𝘳𝘦 𝘢𝘧𝘧𝘰𝘳𝘥𝘢𝘣𝘭𝘦, 𝘢𝘯𝘥 𝘸𝘦 𝘤𝘭𝘦𝘢𝘯𝘦𝘥 𝘶𝘱 𝘵𝘩𝘦 𝘳𝘰𝘭𝘭𝘴”, Oz said, crediting the cleanup with $𝟒.𝟒 𝐛𝐢𝐥𝐥𝐢𝐨𝐧 𝐢𝐧 𝐬𝐚𝐯𝐢𝐧𝐠𝐬 in 2025 alone, with duplicate cases already down to roughly 190,000 this year. The crackdown builds on a CMS finding from last year that flagged 2.8 million Americans potentially enrolled in two or more Medicaid or ACA exchange plans at once, an overlap CMS estimated could cost taxpayers up to $14 billion a year (CMS Newsroom). Minnesota isn't an abstraction here, it's one of the states that has repeatedly made national headlines for Medicaid fraud involving fake autism clinics, phantom housing stabilization services, and feeding-program scams that have already produced federal indictments. Washington just told Minnesota's government: fix it, or the money stops. 𝐂𝐥𝐞𝐚𝐧 𝐮𝐩 𝐭𝐡𝐞 𝐛𝐨𝐨𝐤𝐬, 𝐨𝐫 𝐥𝐨𝐬𝐞 𝐭𝐡𝐞 𝐜𝐡𝐞𝐜𝐤. Source: @EricLDaugh x.com/EricLDaugh/sta…

Eric Daugherty @EricLDaugh

a week ago

🚨 JUST IN: Minnesota may NEVER get back the $259 MILLION in Medicaid money the Trump admin froze over suspected fraud, because they CAN'T PROVE THE CLAIMS ARE LEGIT — POLITICO Imagine that 🤯 Dr. Oz and JD Vance gave Minnesota a simple deal in February: prove the claims are

74 774 3K 92K 92

Oil has had one wild ride since 2000, and it's not exactly calming down now. In 2008, booming demand from China and a flood of speculators sent Brent to a record $144 a barrel. Months later, the global financial crisis hit and prices crashed below $40. By 2016, Brent had sunk to $26 after the U.S. shale boom flooded the world with oil and Saudi Arabia refused to cut production to prop prices up. Then came 2020. COVID lockdowns killed demand overnight, storage tanks filled to the brim, and U.S. crude briefly went NEGATIVE, meaning sellers were literally paying people to take it off their hands. In 2022, Russia invaded Ukraine, the West slapped sanctions on Moscow, and Brent rocketed back up to $133. Now it's climbing again, with Brent settling at $106.65 yesterday and WTI at $93.74, as the Iran war drags on, the Strait of Hormuz stays under blockade and Houthi missiles rain down on Saudi Arabia. Every few years, the world finds a new way to break the oil market, and this year it's going for the full combo. Writer: Julie x.com/OilandEnergy/s…

I used Claude Fable 5.1 and GPT-6 Astra to build a self-learning trading algorithm The result: +$1,840.82 in two days It makes about 17 trades a day or 521 a month, with an average trade size of $53.2 Six AI agents track new coins on Solana and learn to make trading decisions as one system I called it QuantumOrb GitHub repo:github.com/VALIXxDaniro/Q… Here’s how it works: SCOUT tracks new launches, bonding curve activity and PumpSwap swaps, then records the trading data CARTOGRAPHER detects shared wallet funding sources, same-bundle activity and connections to the token creator JUDGE analyzes traders from their transaction history and estimates their performance and edge after fees, entry delay and copybot activity QUANT calculates execution on the curve or inside the pool, accounting for fees, market impact, priority fees and Jito tips GUARD checks holder concentration, connected wallet groups and launch risks EXECUTOR models opening and closing positions through take-profit, stop-loss or time expiry At the center of the system is the Orb Every launch is classified into three probability states: +60% is reached first −30% is reached first Neither threshold is reached within 60 minutes Most profitable trades: $45 → $158 (+$113) $37 → $146 (+$109) $31 → $125 (+$94) Organic buyer flow shifts probability toward one scenario Connected-wallet activity, sells and holder concentration adjust the estimate in the opposite direction With every new signal, the state updates. After that, QUANT determines whether the setup still makes sense after all execution costs, while GUARD decides whether the risk is acceptable Every launch that passes GUARD is used for learning, including the ones the system decides to skip One hour later, the actual outcome is recorded and fed back into the model That feedback becomes part of the next decision.

Daniro @Dan1ro0

7 months ago

x.com/i/article/2036…

21 30 255 666K 474

Your coding agent shouldn’t send raw secrets to the model. Baize is an open-source coding-agent CLI for builders who want tool use, multiple model backends, and privacy controls in one local workflow. It helps you work through coding tasks without exposing sensitive values directly to the model by replacing detected PII with placeholders before requests and restoring it in the response. Key features: • Reversible privacy masking – detects items such as emails, phone numbers, API keys, and bank cards before model requests • Local-first operation – the README says it has no remote database and runs completely locally • Flexible model backends – supports DeepSeek, OpenAI-compatible APIs, GLM, Qwen, Kimi, and local Ollama • Agent toolchain – includes shell execution, file operations, search, web fetch, background tasks, and todo management • Extensible workflows – supports skills, subagents, hooks, and MCP servers for adding domain knowledge and tools It’s open-source (MIT license). Link in the reply 👇

🇨🇭 To prove an earlier point… this Swiss man is out walking these CUTE-ASS rams every day because he couldn’t bring himself to kill them! He probably gave them all cute names… hard to kill something named ‘Grace’ or ‘Ethel’! Writer: Jamie

Si je te propose un rendez-vous ce soir, tu acceptes ou tu as peur ? ︎ ︎ ︎ ︎ ︎ ︎ ︎ ︎ ︎ ︎ ︎◯ J'accepte ︎ ︎ ︎ ︎ ︎ ︎ ︎ ︎ ◯ J'ai peur ︎ ︎ ︎ ︎ ︎ ︎ ︎ ︎ ︎ ︎ ︎ ︎ ︎ ︎ ︎ ︎ ︎ ︎ ︎ ︎ ︎ ︎ ︎ ︎ ︎◯ Ça dépend

it's f*cking insane.. Anthropic just dropped a Claude Code setup that makes tokens up to 97% cheaper Opus brains, Sonnet bill: > Sonnet runs your session and writes the code > Haiku subagents grab the small tasks > Opus steps in only when Sonnet gets stuck > it checks the work, digs into the problem and puts Sonnet back on track $19.50 saved on every 1M tokens a Haiku 5.5 subagent writes, compared with Opus 5.5 one command sets it up: npx claude-code-templates@latest --setting model/sonnet-opus-advisor-haiku-subagents the easiest money you'll save on Claude Code this year

Mnimiy @Mnilax

4 months ago

x.com/i/article/2068…

7 10 136 71K 252

Starlink offers fast, reliable internet in the most rural and remote locations. Portable and easy to set up, Starlink Mini works wherever you are. Order online in minutes.

holy sh*t. Official Anthropic tip for Claude Code: Once Opus 5.5 is your main model, stop leaving Haiku 5.5 sitting idle and stop burning Opus tokens on sweeps Sonnet 5.5 and Haiku 5.5 can tear through turn mass-agent sweeps into deterministic code with /workflows claude --effort ultracode Opus 5.5 authors the workflow script and synthesizes the final PR Sonnet 5.5 executes precision refactors in isolated worktrees at medium effort Haiku 5.5 runs at 340 tok/s as the dedicated pipeline filter, cutting cognitive load at three critical bottlenecks: → before dispatch: sweeps the full codebase in parallel to extract AST candidates without context bloat → on schema validation: drops malformed tool calls and false positives before Sonnet touches them → before synthesis: verifies that every candidate strictly conforms to the target JSON schema Haiku 5.5 filters. Sonnet 5.5 refactors. Opus 5.5 ships Jev engineering is the same move one layer down: the forks that need no thinker (which file, which tool, retry or stop) resolve in V8 in 16ms, and frontier models only wake up when execution paths actually split Author on high. Refactor on medium. Keep Haiku at 340 tok/s. - the full workflow architecture > Opus 5.5 on high drafts the workflow script in .claude/workflows/ > pipeline() fans out Haiku 5.5 at 340 tok/s across target files > parallel() schema fuzzers drop false positives before patch generation > Sonnet 5.5 refactors verified files in isolated git worktrees > intermediate state lives in V8 variables, keeping context drag at 0 tokens > 1h prompt cache prefix hold eliminates redundant system prompt processing paste the setup and this prompt into Claude Code ↓ "Configure my Claude Code workspace for dynamic workflows: 1. Audit ~/.claude/workflows and .claude/workflows for existing scripts: > Draft workflow definitions for codebase audits, parallel migrations, and schema validation > Pin model: haiku, effort: low for AST sweepers, and model: sonnet, effort: medium for refactor workers > Configure workers to operate in isolated git worktrees to eliminate edit conflicts 2. In ~/.claude/settings.json, enable dynamic workflows: > Set ultracode = true > Lock main session effortLevel to high > Set subagentPromptCacheTtl = '1h' to share prompt prefixes across parallel fan-outs 3. Set workflow runtime controls: > Set autoContinueAtUsageLimit = true so long batch runs automatically resume after limit resets > Set CLAUDE_CODE_WORKFLOW_PREFIX_STAGGER_MS = 5000 for prompt prefix caching > Set large workflow warning threshold to 25 agents 4. Add a rule to ~/.claude/CLAUDE.md: all multi-file refactors must pass an adversarial Haiku schema check before patches merge to main Show me every change as a diff first. No edits until I say go." ↳ code.claude.com/docs/en/workfl…

mirku @mirku21

a month ago

x.com/i/article/2098…

29 78 680 2.1M 2K

Anthropic's research PM (ex-DeepMind) just showed the exact loop Anthropic runs to pick a model for every task. i turned it into a prompt you paste into Claude Code: 1 → build an eval from ~20 real tasks, log cost next to every score 2 → set the quality bar from your margin 3 → fix the prompt before you blame the model 4 → free levers first: caching, batch at 50% off, compaction 5 → sweep model × effort. "low and medium are the new high" full prompt in the image, their own words in the webinar below ↓

Shoopy @0xShoopy

2 weeks ago

Anthropic engineers: "low and medium are the new high" in a 48-min webinar, Lucas Gonzalez and Ben Lehrburger show how Anthropic picks the model for every task, and take a support agent from 79% to 94% resolved while cost per ticket drops 5x: 14:02 → the default ladder: Opus

10 7 68 5K 40

OPUS 5.5 + DOTS + JEV CLOSED 312 TICKETS IN ONE WEEK. 25 engineers closed 96 That org chart costs $343,750/month in payroll. The agent costs $2,055/month in tokens. That's 167x less. The split is simple: → Opus 5.5 writes the code → Dots writes the tests → Jev reviews everything → Nothing reaches "done" without Jev's approval And here's the part most people miss: The team has 1,000 hours available each week. Only 410 hours are actually spent writing code. That's 41%. The other 590 hours go to meetings, reviews, on-call, customers, and decisions. The agent is taking the 41%, not replacing the entire engineering organization. Now look at the other side. A business bills $120,198/month from 402 customers paying $299. After fees, hosting, the agent, and 3 humans, $66,887/month remains before tax. At that revenue level, you can afford roughly 8.7 engineers at most. So that business never had 25 engineers to replace in the first place. That's the part worth understanding. Al isn't simply replacing "25 engineers.!" It is changing how much output a small organization can produce with the same human decision-making layer. SAVE THIS SO YOU DON'T MISS IT.

motyka_9 @motyka_9

2 months ago

x.com/i/article/2091…

4 9 111 819K 148

Yet another Polymarket bot made $63.7K by buying the most obvious outcomes on 5-min Crypto Up/Down markets BTC, ETH, XRP, or SOL - it doesn’t matter Once UP or DOWN hits 0.99 odds, it places limit orders Fills automatically collect maker rebates This setup saved thousands in fees while making an extra $5.5K Across 6 months and 14.3K trades, it lost just 45 times Can someone build me a bot like this? Profile: polymarket.com/@0x9ad6c5316ae…

keyboard_arrow_left Previous keyboard_arrow_right Next

Leshuk @leshuuuk

3 weeks ago

Can you really bond 5-min crypto Up/Down markets on Polymarket? Yes, you can and someone has already automated it This trading bot buys 0.99 odds across different short-term crypto markets: > 2-3 trades per hour > Uses limit orders only > Position size capped at <0.8% of

keyboard_arrow_left Previous keyboard_arrow_right Next

23 2 90 12K 73

this was never supposed to leave the studio group chats Jev doesn't generate images or write code. it picks. layout A or B, serif or sans, cut here or there. code does the drawing. that's why it never ships something half-broken. want a whole design system from one sentence? -> jev-design: type "an 18th century printed book" and the dashboard becomes one a second or two later github.com/bilune/jev-des… want full UI screens from a one-line prompt? -> jev-design-test: real shadcn screens in ~1-2 seconds, Jev never writes a line of JSX github.com/bhaiG-de/jev-d… want Figma to follow plain sentences? -> jev-figma-design-plugin: Jev decides, code draws github.com/voidning/jev-f… want a long video cut into shorts? -> jevcut: 62 clips out of a 97-minute podcast, about 3 cents of Jev per hour of video github.com/VBS2004/jevcut want a video editor with Jev inside? -> HyperEdit: FFMPEG + Remotion, powered by Jev github.com/kevinbadi/hype… the catch: Jev only chooses from options a human already built. no vocabulary, no design. and it's still behind a waitlist. the taste is yours. the thousand tiny choices don't have to be. bookmark this. pick one. ship something this week.

kiosa @thegreatest_sv

a month ago

x.com/i/article/2102…

23 79 462 1.2M 1K

The next generation Starlink V5 has a smaller form factor and lightweight design with greater power efficiency. With speeds up to 375+ Mbps, Starlink V5 delivers reliable home internet for streaming, video calling, gaming and more. Available in select areas.

i spending 4+ years building institutional level 24/7 trading systems today i wrote a 6-page research paper on exactly how to find profitable strategies 24/7 with Opus 5.5 from scratch along with the COMPLETE CODEBASE opus 5.5 builds the MOST POWERFUL 24/7 trading agents right now here is how to build your first PROFITABLE trading bot with opus 5.5: 1. start with the market, not the strategy, every idea must name who loses on the other side or it has no edge 2. build every signal from closed bars only, one index shift separates a real backtest from one that sees the future 3. rank a whole universe instead of timing one asset, long the top, short the bottom, keep beta near zero 4. test enough variants and the best Sharpe rises with the square root of 2 ln N on pure luck, so most strategies should die 5. fees decide more than signals, 250 trades a year at 0.09% per round trip costs 22.5% before slippage 6. size so a 2 ATR stop loses 1% of equity, halt everything at max drawdown and change one rule per night the exact system I have been running for the past 5 days is in the article below (INSANE REULTS):

Roan @RohOnChain

2 weeks ago

opus 5.5 is the most dangerous AI model for trading right now it gives you 1M token context and always-on reasoning that can discover new profitable trading strategies for you 24/7 if you set it up correctly, you gain superpowers x.com/i/article/2104…

76 83 657 1.2M 2K

How many tabs are open right now that you're "going to read later"? Be honest. You won't. Nobody does. Later never comes because nothing brings it BACK. A second brain with Opus 5.5 reads it, files it, and puts it in front of you the week it matters. Full course below. Bookmark this.

CyrilXBT @cyrilXBT

3 weeks ago

x.com/i/article/2104…

33 73 670 932K 1K