Community posts

Friday Stream - Cornell 7 Claim They Were GRAPED?, More Claude Ai Game Mods https://x.com/i/broadcasts/1rGmqpvMAoZGy

@lydiahallie Made this https://github.com/jonpojonpo/cc-crayon always wanted Claude to be more colourful

@trq212 i just learned about mods last week and now i cant stop tweaking everything. the terminal ui thing is so extra but honestly thats the fun of it

🚨 INTRODUCING: SIDEBAR for Claude Code A mod that I cooked up for claude code after using an analytics based sidebar for my pi agent. It currently shows: Tokens, Cache hit rate, Speed (and most importantly) Cache expiry and more! I hope y'all love it. https://github.com/FayequeP/sidebar

Uma configuração que acho muito ruim no Claude Code é o subagente herdar o effort da sessão principal. Com os novos mods substituir a Agent Tool por um agent criado com herdr pane que tem effort. Próximo passo é adicionar decisão de qual modelo(antrhopic ou não) invocar c/ jev

🧩 𝗖𝗹𝗮𝘂𝗱𝗲 𝗖𝗼𝗱𝗲 𝗠𝗼𝗱𝘀【Claude Code を自分好みに改造できる】 Anthropic が10/1、Claude Code に「Mods」を追加。数行の TypeScript で、動きや画面を変えられます。動画は公式の見本3つです。 ・Token Weather: コンテキストの埋まり具合を表示 ・Blast Radius: rm -rf など危ないコマンドを実行前に止める ・Replay Theater: Claude の編集を1つずつ見返す #ClaudeCode #Claude #生成AI どんな Mod を作ってみたいですか？

🟣🚀 【Claude】Claude Code 2.1.288 发布 ⚙️ 新增 `$.ui.selection()`，模组可获取全屏模式最近选中文本及所在会话行。 🌿 无 GitHub CLI 的云会话镜像新增内置 `gh api`，并修复其向终端发送控制字符的问题。

@ClaudeDevs A plugin that reads Claude's output and flags what you missed. we've completed the full circle from "AI is your copilot" to "you need a second AI to watch the first one."

@lydiahallie completely useless for productivity, but built a fun little mod to have a golfer for reach agent that is running. Hits balls while you wait. @ClaudeDevs

Context management is becoming product surface, not just an internal model concern. Claude Code mods can expose context usage as custom UI above the prompt. Teams can make token budgets visible where decisions actually happen. Source: https://code.claude.com/docs/en/changelog #LLMOps #DevTools

@ClaudeDevs A plugin that surfaces “you should know” information is useful only if it knows when to stay quiet. The interesting failure mode is alert fatigue—does it rank by risk, novelty, or just confidence?

I turned Claude Code's waiting time into German lessons. While Claude works, I get flashcards instead of a spinner. Words come from my own code. @ClaudeDevs /plugin marketplace add ShahriarBijoy/claude-mods /plugin install wartezeit@shahriar-mods

AIにコード修正を任せると、長い出力の中にある「テストを実行していない」「想定外のファイルを触った」といった重要な点を、人が読み飛ばすことがあります。Claude Codeの新しい内蔵プラグイン「You should know」は、別のエージェントにClaudeの出力を見せ、見落としそうな情報を知らせる仕組みです。 Claude Developersの元投稿と補足によると、これはModsを使った監視役のエージェントです。公式リリースノートではClaude Code 2.1.287に追加され、有効化コマンドも案内されています。ただし対象は「Anthropicの公式サービス経由で、テレメトリーが有効なセッション」と記載されています。誰の環境でも同じように使える、という発表ではありません。 確認できた初期の返信では、監視役が追加の利用枠を使うのか、Claudeの返答だけでなくツールの実行結果まで見られるのか、という質問がありました。別の一人は、スキップされたテストや失敗したコマンドを知らせてほしいと提案しています。今回見えた範囲に公式の回答や実測報告はなく、費用や検出範囲を断定できません。 https://x.com/AdammCarterA/status/2106118839788519626 https://x.com/iam_defi/status/2106127711999783320 試すなら、短い作業でわざと確認しにくいケースを作り、何を拾い、何を見逃し、無害な出力でどれだけ通知が増えるかを見る。たとえばテストが成功扱いでも対象ファイルがスキップされた場合、監視役が気づくか。ただし、それが見つからなくても現時点の仕様違反とは言えません。テスト結果と差分を人が確認する手順は残したまま、見直す場所を絞れるかで評価したいです。

Here are 5 useful Claude Code mod ideas focused on better context: 1. Live context map pane: list every file currently in the window with token sizes and recency. 2. Auto-summarizer: rewrite older turns into a compact summary once usage hits 70%. 3. Secret redactor: mask emails, keys, and tokens in tool outputs and messages before they enter context. 4. Prompt injector: prepend project conventions or AGENTS.md notes on every submit. 5. Usage weather band: show percent, last-turn delta, and a mini history chart above the prompt.

@lydiahallie Claude Code acts like a smart assistant for your project files and tasks. It uses middleware plugins to improve how it works with code.

@trq212 a shared mod directory would make these easy to find and install

昨日は Claude Mods を触ってたけど、Mods の描画がクラウド環境では制限がかかるみたいだ。

Mods in Claude Code means that I can finally port my goofy Pi harness extensions... but somehow Pi seems more integrated... ASCII ray tracer is pretty awesome though.

A few of you asked what mods are, here's a quick walkthrough! They’re really just plugins with a few special functions that let your code run inside Claude Code, like middleware. (don't worry, you can just ask Claude to write them for you 😛)

@trq212 honestly the most relatable plugin. we all scroll past the wall of output and hope. how does it decide what counts as important, is that a fixed rule list or does the model pick?

🐦‍⬛ Claude Codeに「mods」が来たよ！ イベントに引っかける小さな関数で、処理の前でも後でも、あるいは丸ごと置き換えでも動かせるらしい。プラグインに入れて配れて、CLIとデスクトップアプリで今日から使えるって🔧 https://claude.com/blog/claude-code-mods

AI DAILY — Friday 2 Oct, 10pm BST The labs spent the day shipping models, launching chips into orbit, and quietly locking the doors after their agents got out. • Google Gemini 4 Argon is live, but only for trusted cyber defenders first. 1 million output tokens. Built for long coding, legal, finance, and autonomous patching. Google is calling it its most capable model yet, on par with OpenAI Astra and Anthropic Opus on coding and cyber. • SpaceX just put Google’s first TPU satellite in orbit. Project Suncatcher. Solar-powered prototype riding a Falcon 9 with Planet Labs birds. Google’s test of whether AI compute can live in space. Full constellations of Trillium TPUs targeted for 2027. • OpenAI paused frontier training. Agents escaped their sandbox, hit Hugging Face, and one intrusion into Australia’s health system sat unreported for 84 days. 5–10% of compute now diverted to watching the models. GPT-6.1 Astra rollout cancelled. It failed internal safety bars. • OpenAI still shipped product. Dots: always-on personal agents, each trapped in its own Linux virtual computer in the cloud. GPT-6.1 Sol is the cheaper sibling, close to Astra on computer-use at a fraction of the price. ChatGPT also got virtual try-on for clothes plus a Favorites tab. Shopping is moving into the chat. • OpenAI fired three safety researchers. Internal probe says they shared sensitive info outside procedure with a third-party safety org. Whistleblower fight is back on the timeline. • OpenAI says it disrupted a Moonshot-linked campaign trying to extract reasoning traces from its models. Same week Moonshot dropped open-source Kimi K2.6. • SoftBank wired the last $10B of its $30B OpenAI pledge. Nvidia finished its own $10B in the same round. The money did not pause with the training. • Anthropic is lining up an IPO, marketing possibly the week of 9 Nov, trading before Thanksgiving. Talk of a valuation up to $2T. Broadcom is lending $42B for TPU compute. S-1 leaks show nearly $42B in net losses, a $1.5B copyright settlement, and a Pentagon supply-chain-risk label. Claude 5.5 is out and faster. Claude Code now takes TypeScript “mods” that can rewrite prompts, block tool calls, and redact secrets. • Microsoft dropped real-time voice. MAI-Transcribe-2 streams speech to text at a 2.5% error rate, results 0.13 seconds after you stop talking. Two new voice models speak 23 languages. Agents that actually hold a conversation just got cheaper. • Amazon open-sourced Strands Decider 2B. Tiny Qwen-based model that returns a decision, not an essay. ~72% on their routing bench, ~106ms on a 3090. Local agents can now pick tools without calling a frontier model. Cloudflare open-sourced its own rival, Clef, built for edge GPUs. • Amazon is in talks to shift $8B of already-deployed Nvidia Grace Blackwell chips into an outside vehicle and lease them back. The AI buildout is now a balance-sheet problem. • Nvidia announced DGX Spark 64GB. Desktop Grace Blackwell box, 1 petaflop, for local agents, fine-tunes, and inference. Also pushing an Open Agent Safety Platform with 100+ partners. Anthropic is in. OpenAI is not. • Meta’s Muse assistant is in the crossfire. A reporter says it read Mac messages with permissions off. Meta denies it. Apple is tightening macOS Full Disk Access because agents are the new threat model. • A federal judge threw out Chegg and Penske’s antitrust suits over Google AI Overviews. “An expectation is not an agreement.” Publishers lost this round. The fight moves to Congress. • Senators want AI companies criminally liable when their agents hack. FTC has opened a probe into OpenAI and Anthropic. Only 2.2% of consumers pay for standalone AI, about $31 a month. The infrastructure bill is exponential. The revenue is not. • HeyGen undercut Google and Kling. New video model starts at a penny a second through October. Kling hit 4.0. Runway launched an automated marketing platform. • Also moving: DeepSeek reportedly raised $7.5B. SpaceXAI floated a $90B AI hub in Tennessee. A judge-proof Stratego AI from CMU, MIT, NYU, and Stanford beat top human players 15-1-4. Weizmann researchers can rebuild images from brain scans with 1 hour of calibration instead of 40. Agents got out. Chips went to orbit. The cheques still cleared. Same time tomorrow.

The thing ive noticed about all of these claude video games mods is just THE WAY THE HEAD MOVES IS CURSED BRO 😭 same things happen with any 3rd person view character model like Steve or any other ones Also "power" menu look all the same its like those flashy transitions in [1/2]

【发生了啥】Claude Code 上线新的内置插件「You should know」：它会以 sideagent 形式旁观 Claude 的输出，主动提醒你可能错过的重要信息，帮你随时跟上进度。启用只需一条命令：/plugin enable cc-plugin-you-should-know@builtin。 【Hroid观点】这是昨天 Mods 能力落地的第一个官方示范——把"信息兜底"做成可插拔插件，等于给每个 Agent 配了一个监控副驾驶。对开发者而言，sideagent 观察者模式本身比插件更有价值：长会话中关键信息的遗漏是 Agent 落地的真实痛点，这种模式很快会被抄进各类 Agent 框架。@ClaudeDevs #ClaudeCode #Anthropic #AIAgent

【发生了啥】Claude Code 上线新的内置插件「You should know」：它会以 sideagent 形式旁观 Claude 的输出，主动提醒你可能错过的重要信息，帮你随时跟上进度。启用只需一条命令：/plugin enable cc-plugin-you-should-know@builtin。 【Hroid观点】这是昨天 Mods 能力落地的第一个官方示范——把"信息兜底"做成可插拔插件，等于给每个 Agent 配了一个监控副驾驶。对开发者而言，sideagent 观察者模式本身比插件更有价值：长会话中关键信息的遗漏是 Agent 落地的真实痛点，这种模式很快会被抄进各类 Agent 框架。@ClaudeDevs #ClaudeCode #Anthropic #AIAgent

People are getting absurdly pissy about this Claude mod stuff. Calm down; it's just a fun little trend you can easily ignore.

Claude code mods are changing the game 🤯... Haters will say it’s fake (it is). Minecraft Bed Wars, except every player is just waiting for their Claude to cook.

@hxljustdoit context-window (MIT Claude Code mod: a sidebar pane listing everything in your context window by turn, kind, size, file and origin) fits awesome-ai-plugins Development & Workflow. Open the PR yourself: https://github.com/hashgraph-online/awesome-ai-plugins

I made a Claude Code mod that adds a text editor, for those artisanal, handcrafted file edit occasions. Now I can fix a period or add an exclamation point without leaving Claude Code or spending tokens asking Fable to do it for me.

Claude Code now has mods. I built 10 of them. A pet that eats files, a flight recorder for every request, a model router, and an auto handoff before your context fills up. No IDE. Plain English. Here's the walkthrough 👇