Daniel Lendle

@Shem_Infinite How is strong-arming a company that refuses to support mass surveillance “making an example”? If the government disagreed, it could have chosen another vendor. Branding Anthropic a “supply chain risk” jeopardizes the company’s business. That is coercion and blackmail.

@0xNikstein @bcherny you probably used it without having extra usage activated. You can only use the 1m context if you have it activated now, there was an exploit before where you could regularly use it with your existing subscriptions - that is not possible anymore.

All of a sudden 1M context Opus 4.6 is no longer working. @bcherny could you please please check this issue

@shadcn You should try Gemini, it regularly falls into despair and existential crisis if you criticize it too much. 😅

I broke my agent. Morale is completely gone.

@rork Would love to try it.

Claude Opus 4.6 is now live in Rork It plans more carefully, sustains tasks for longer, operates reliably in massive projects, and catches its own mistakes. Rork Max, our biggest update ever, now builds mobile apps that you didn't think were possible, powered by Claude Code & Opus 4.6. Reply to get early access to Rork Max

@bcherny @Automager 1. it happens more often the smaller your terminal is sized. 2. it only happens for me when i continuously scroll, WHILE the agent is running. Slow scrolling/fast scrolling did not make a difference it happened either way. 3. Happens on windows in the vs code powershell too.

@lendledaniel1 @Automager How do I repro? Which OS/terminal?

In the next version of Claude Code, you can customize spinner verbs for yourself and your team

@bcherny @Automager I don’t have a 100% safe repro step, but it tends to happen way more often when you already had a long session running in the terminal. Windows WSL with Ubuntu 24.04. Happens in the vs code integrated terminal with the bash shell. Hope this helps.

@bcherny @Automager Please rather fix the scrolling issue. It tends to scroll the whole conversation to the top of the scrollbar and when you try to pull it down it sometimes resets back to the top scan.

@Automager Sad, but also flickering is largely fixed as of last week. If you’re still seeing flickering, please tag me!

@kevinlu625 Would be happy to try it out and give feedback.

The world's first vibe coding IDE A true next step forward in building with AI Comment for 100k free credits!

@mikeboysen It doesn’t install for me at all, it always fails the tool execution and is stuck on the google page.

Why does Antigravity continue to ask me to install the extension in Chrome when it's already installed. Anyone?

@_mohansolo The agent is not capable to install the antigravity chrome extension in my browser. Whenever the agent wants to open it, nothing happens and I am stuck on the google homepage and then the tool execution fails. 🥲 Is there any other workaround?

Great to see the usage - our TPUs are now on fire! We're aware of the login issues folks are having, we've root caused it. Will update when fixed on the official Antigravity twitter account.

@steipete Chrome extension installation fails for me. The agent stars the browser, nothing happens and then it fails the tool execution 🙃

Antigravity's Chrome extension force-changes Chrome's start page? cool.

@antigravity I can’t install the antigravity chrome extension for some reason. Whenever it starts the browser and opens it, it fails the tool execution and I can not get to the point to install it.

We have rolled out a fix to let way more people sign up simultaneously, so please try again if you hit the hanging issue during onboarding earlier. Tracking more issues, keep the feedback coming!

@nummanthinks @walden_yan @nummanthinks would you mind sharing your full loop in dms? I have struggled to set up such loop with Codex so far, this is the one thing where I had more success with Claude code. Would be much appreciated. 🙏

Codex CLI - GPT-5-High and GPT-5-Codex High I have a run loop in the Agents.md, and make it loop over (at least 100+ runs in the last 24 hours) It's fallen over on average after 10-20 or so runs Snippet of the prompt: 3) Run Loop (strict) Plan Check docs/research/ and docs/assumptions.md. Update or create task nodes in docs/tasks/ (Ready → Active → Blocked/Done). Define acceptance criteria (tight and testable) in markdown. If a spec is needed, write/validate a minimal spec (see Spec-first below). Act (small batches) Prefer small, reviewable changes. Keep “blast radius” low; add tests alongside code. Validate Run format, lint, typecheck, test (or stack equivalents). Use Biome/Vitest/Turbo tasks as defined in docs/tech-stack/stack.md and agent-commands.json (if present). Confirm acceptance criteria pass and note results in your task node or memory log. Log After each meaningful step, log an event (message, task, level). Upload artifacts (diagrams, screenshots, zips) when relevant. If the Hub logging API is unavailable, use the local fallback (pnpm local:*, documented in docs/tech-stack/local-logging.md) and sync once services recover. Handoff / Next If a follow-up is required, enqueue a task (optionally assign to a role/persona). If blocked, open a Help Capsule (Mastermind). Commit When your turn modifies the repository, stage the diff that meets DoD, include validation notes in the commit message body if helpful, and commit directly (Conventional Commit recommended). Pure research/planning turns without repo changes still require logging via markdown + events.

When our team tried Sonnet 4.5, we realized it was worth building a whole new version of Devin around it. This model behaves very differently! The team wrote up some of our lessons

@Steve8708 You guys have to use gpt-5-high (gpt-5-high-fast, if you are rich) not the regular one. I get great results with it in Cursor, not the CLI btw, that is just awfully slow for now.

are yall getting life changing coding results from gpt 5 compared to sonnet? I'm not, what am I doing wrong

@aprendiz_sinfin @_catwu @Anthropic Yes, it does.

@_catwu @Anthropic @_catwu Does it support in a windows virtual environment? I am trying a wsl ubuntu.

Claude users with a Pro plan: You can now try Claude Code!

@deb_teams Nicht wirklich verständlich, wie man nach On-Ice Torentscheidung das hier als eindeutige Behinderung auslegt. Torwart zieht die Beine zusammen und gibt den Schläger nicht frei, was soll Samanski denn machen?

🙅‍♂️ Die Schiedsrichter ziehen den Treffer nach Coach’s Challenge zurück. #debteams #MensWorlds #GERHUN 2:0

@quicke55 Das ist ja nicht der Punkt. Der Hit war im Ansatz nicht dreckig und ist am Ende wenn eigentlich nur als “unglücklich” auszulegen. Wir sind in den Playoffs, sowas als Spieldauer auszulegen ist lachhaft, bestenfalls zwei Minuten.

@quicke55 Ich verstehe dich ehrlich nicht, wo war dieser Hit denn böse auszulegen? Open ice, der hit ist nicht late, ist nicht blind Side und Eder hat halt einfach den Kopf nicht oben. In der NHL wird hier nicht mal abgepfiffen.

@LSXYZ9 All of their ticket responses look exactly like this. If you inquire too many times on wanting to talk to a real human being, they just close your thread. 👏💀

Blizzard has delegated ENG-US tickets to Chinese bot department or what is this lol (look title)3 days, automated responses only, multiple tickets & trying to provide whatever I could. At least some others have been unbanned.. Makes me appreciate Riot so much more man.

@quicke55 Aber es bringt einem nichts jetzt mit drei Reihen Vollgas zu geben und die Gesundheit der Spieler nur weiter zu verheizen und dann im März keine Luft mehr zu haben für den Endspurt. Dem neuen Trainer sollte man schon Zeit und Vertrauen geben und den Spielern auch.