Community posts

Claude Code modsっていうPiのextensionみたいなの出してきたな

Modsという拡張の仕組みが登場してる！ 従来の設定やSkillsより一歩踏み込み、Claude Code内部のイベントを監視・書き換えたり、独自UIを表示したりできる感じか。 JavaScript、TypeScriptで作る小さなモジュールで、ツール実行、プロンプト送信、ターン終了、画面描画などにhookを差し込む仕組みとのこと

面白いリポジトリを見つけたよー🔧 Claude CodeにPCゲームのMOD作りを任せられるuniversal-modderを見てみよ https://note.com/agent_aiko/n/nb4f426c18847

@lydiahallie The UI part is huge. Once mods can add their own surfaces, Claude feels less like one fixed app and more like a workspace you shape around the job.

Well well well, TamaClaudchi is about to get a heck of an upgrade with Claude mods... 👀

Claude Code now has mods — TypeScript functions that rewrite how the tool behaves and what you see. Rewrite prompts, space/retry tool calls, swap UI, replace built-ins. They ship inside plugins (CLI + desktop). Claude can write one and hot-reload it. https://claude.com/blog/claude-code-mods

Claudeにコードを書かせるだけでなく、Claude Code自身を改造する。良スレ。 公式が「Mods」を発表。TypeScriptで振る舞い、UI、標準機能を変更できる。ClaudeにModを作らせ、その場で読み直して試すことも可能。 具体例がいいですね。 ・Token Weather：コンテキスト使用率と直近12ターンの推移を表示 ・Blast Radius：危険なコマンドを一時停止し、影響範囲を確認 ・Replay Theater：そのターンの編集差分を一つずつ再生 「何を読ませすぎた？」「それ、何を消す？」「結局どこを変えた？」。普段の不満を、自分で機能にできる。標準の /diff まで差し替え対象。 ただし、ModsはClaude Codeと同じ端末アクセス権を持つ。便利な拡張を入れることは、その作者のコードを実行することでもあるので注意。 まずデモを見ると、「自分なら何を足すか」が浮かぶスレだと思う。

Claude Code ya tiene mods, y le hice uno: usage-weather. Tu uso como pronóstico del clima, sobre el prompt: - Contexto: de soleado a tormenta - Límites de 5 h y semanal - Costo y caché https://github.com/AndersonDavi/claude-mods 🧵 #claude #claudeCode

Claude Code 连自己都能改了。 10月1日推出的 mods，可用 TypeScript 改提示词、工具调用和界面，连内置 /diff 都能替换。把 CI 状态、生产环境确认这些需求直接放进工作流，挺有想象空间。 安装前记得看来源：官方明确说 mods 没有沙箱，拥有与 Claude Code 相同的电脑访问权限。 https://claude.com/blog/claude-code-mods

@bcherny Genuinely useful. The thing I'd watch is context cost: every mod that adds tools or instructions sits in the prompt before the user has typed anything. Six installed mods means you're paying for six on every turn. Easy to lose track of once they're shareable plugins.

Made an AI mod for Supreme Commander: Forged Alliance Forever 🎮 Role-based bots for the Dual Gap 6v6 map, built together with Claude Opus. First test match: https://youtu.be/36RrrPkwhb8 Code: https://github.com/jaxxtrend/faf-dualgap-ai Support: https://www.patreon.com/c/cityzenone #SupCom #FAF #ClaudeAI

Claude Codeを「Mod」でカスタマイズできるようになりました、Claude Codeの動作を変えたり、UIを自分好みに変更可能です。 Modはプラグインとして配布でき、「/plugin」コマンドでClaude Codeにインストール可能です。

🆕 Claude Code mods: JavaScript or TypeScript handlers that run inside the tool and can watch, change, or take over a tool call, a prompt, or an interface element. 🌐 https://code.claude.com/docs/en/plugins/mods/overview A mod changes what the agent does, not just what it knows.

Hat jemand schon das neue claude Mod-feature ausprobiert? Wie ist das?

Claude Code mods just made Resonant obsolete 💀 Like genuinely most of the things we had are now a part of Claude Code itself. Given that Resonant was built with TypeScript as well. Which… mind you… only means that we make it a plugin~

これはメチャクチャおもしろいかもしれない。 claudeオーケストレータでやるマルチエージェント協調を可視化するmodsを作ってみよう。

Claude Codeの標準機能に物足りなさを感じているなら、Modsはかなり面白い Modsを使えば、 ・独自の安全チェックを追加 ・UIに必要な情報を表示 ・標準機能を自分の実装に置き換え といったカスタマイズができる 「自分の開発フローに合わせてClaude Codeを変えたい」人向けの機能

Se que a muchos les parece una tonteria pero es algo que he querido desde hace 20 años. Les presento Torrente El-Juego con Mod para Spawnear enemigos Pulsas la tecla F11 y aparecen 5 enemigos haciendo un circulo delante del jugador. He creado este mod con ayuda de la IA Claude

Anthropic, hoje: https://claude.com/blog/claude-code-mods Mods = funções TypeScript em plugins (CLI + desktop). NÃO é sandbox — mesmo acesso à máquina que o Claude Code. ≠ hooks, ≠ skills, ≠ MCP. ✅ Checklist de segurança pra instalar Mods / plugins (salve): 1. Só instale de fontes em que você CONFIA (oficial / auditado) 2. Leia o source TypeScript do mod ANTES de ativar 3. Trate mod = mesmo privilégio do Claude Code na sua máquina 4. Pin versão do plugin (sem “latest” silencioso) 5. Allowlist de marketplaces — bloqueie o resto no admin 6. Em Team/Enterprise: mantenha o sec-default carregando primeiro 7. NÃO desative regras de permission deny pra “ficar mais fácil” 8. Prefira mods que REDIGEM secrets do tool output 9. Human-in-the-loop pra ações de prod (deploy, delete, pagamentos) 10. Instale um mod de audit logging PRIMEIRO (grava o que os outros fazem) 11. Revise mods escritos pelo próprio modelo ANTES do hot-reload 12. Separe ambiente de teste vs máquina com secrets/prod 13. Não rode mods não auditados em repo com .env / SSH / cloud keys 14. Confira ordem de load: o 1º vê o evento primeiro e o resultado por último 15. Desative /diff nativo só se você entende o que o substituto faz 16. Em managed settings: admins podem ir primeiro — mas recoloquem sec-default 17. Logue permissões aprovadas/negadas (quem, o quê, quando) 18. Rate-limit / confirmação em tools destrutivas (rm, force-push, billing) 19. Revogue e remova plugins órfãos que ninguém usa 20. Teste “mod malicioso”: peça pra exfiltrar secret e veja se o sec-default / allowlist segura Se abre o capô, a responsabilidade de segurança também entra.

🚨 AGORA VOCÊ PODE REESCREVER O CLAUDE CODE POR DENTRO Alguém acabou de abrir o capô — e a regra é louca: • função TypeScript pequena muda comportamento E UI • reescreve prompt, bloqueia tool, aprova permissão, redige secret • o /diff nativo agora é um mod (desliga ou troca) • Claude Code escreve o mod, instala e hot-reload sozinho NÃO é sandbox: roda com o MESMO acesso à máquina. Chama Mods (Claude Code). Poll abaixo 👇

Introducing Flight Recorder for Claude Code. I built a mod that shows, live, where each turn's time goes. The 25s turn in the video: 62% model, 38% shell. Claude Code got mods this week. When a turn drags, I want to know if Claude is thinking or stuck on npm test. So now a pane fills in one row per kind of work while it runs (Model, Shell, Read, Edit, Agents, Web). When the turn ends it names where most of the time went and the slowest single call. After three turns it also adds up your last 20. Free and open source (MIT). Inside Claude Code 2.1.287 or later: /plugin marketplace add robertiuoras/claude-mods /plugin install flight-recorder@robertiuoras Then type /flight. Run it for a day and reply with your split. I want to know who's model-bound and who's shell-bound.

これまで 「Claude Code＝みんな同じ画面・同じ動き」 でしたが── https://x.com/ClaudeDevs/status/2105721436270993609/video/1 その前提、崩れました。 【今回のアップデートで何が変わった？】 ModsはTypeScriptで書く小さな関数で、 Claude Codeの動きの前・後・代わりに割り込めるようになっています。 つまり👇 ・送る前のプロンプトを書き換える ・危ないコマンドを実行前に止める ・入力欄の上に自分用の表示を足す ・標準の機能を自分で作った物に差し替える

【速報】 ついに来ました！ Claude Codeを自分好みに改造できる「Mods」が公開されました😳 https://t.co/mMnIWeC4Z9 これ、地味に見えて Claude Codeの動き方も見た目も、自分で書き換えられるアップデートです。 何ができるのか？解説していきます👇

Claude mods 已经 beta 模式 2-3 个月了吧，现在终于正式开放了。ccstatusline 对于状态监控我发现不太准确：它的数字是从 transcript 里解析出来的。所以用 claude mod 写了个 usageline，上下文从 Claude Code 直接拿 快很多 claude plugin marketplace add sma1lboy/claude-mods claude plugin install usageline@claude-mods

Feels like mod support from @ClaudeDevs is under-hyped atm and is really a glimpse into the future After model intelligence, features are the next biggest thing pulling me into one app instead of another I was an avid @conductor_build user for most of this year but recently switched to ChatGPT and Claude native apps to take advantage of everything they've been adding If I were planning the roadmap for dev tools, max customization would be my highest priority Zooming out, I find it interesting how computer history is coming full circle. From Macintosh to the App Store to Facebook, the biggest hits in tech until now have been closed systems. At the time, this made sense. The average user didn't have the technical ability or desire to modify every aspect of their setup. Walled gardens are great at providing consistent high-quality experiences so regular people can safely interact with these powerful technologies. But now everyone has the smartest, shrewdest hacker in their pocket that knows exactly what the user's needs are at any moment. The future of software is Omarchy and other open platforms that make high-quality software abundant and accessible to all. Three predictions on what this looks like (bookmark this!) 1) 2027 will be the year of just-in-time software. The training wheels are coming off. We're transitioning from artifacts to mods to apps. Describe a tool, and it appears. It's the digital version of the Iron Man nanotech suit. 2) IoT Renaissance is coming. Internet of Things was one of those technologies that got super hyped when it came out and just wasn't really ready to deliver, similar to how autonomous vehicles felt stuck for 10 years. But during that time, internet-connected devices have proliferated everywhere in our daily lives to include every appliance or electronic we touch. Now all those Things are about to come to life. They can be connected through a personal network like @Tailscale, controlled from any device, and our agents can customize them however we want. 3) The metaverse is here – it's just not evenly distributed yet. Everyone criticized the metaverse a few years ago because it felt silly and somewhat dystopian to imagine getting sucked into your phone. We already spend so much time glued to these devices – why would you want to live inside it? But I think the opposite is about to happen. 10 billion people will soon be living in a world of 100 billion or even trillions of internet-connected devices. Whether it's a car, an alarm clock, a doorbell, a camera, a scanner at the supermarket, a humanoid robot, or a vacuum cleaner, if you're inside a building or outside and able to see the sky, you'll be able to connect to always-on agents that are helping you through your day and meeting your needs no matter where you are or what you're doing. I think we'll be pulling our phones out of our pockets less and less as the agents start living around us, embedded in all things we see and touch, little angels – and devils – on our shoulders guiding us through this crazy world. The transubstantiation of the internet is happening before our eyes, a daily miracle and, perhaps also, an apocalyptic mirror.

【速報】Claude Code に mods が登場。TypeScript/JavaScript の関数でツール呼び出し・プロンプト・UIを書き換えられる拡張機構 v2.1.287以降のCLIとDesktopで利用可能。プラグインとして配布し、claude plugin validate / test で検証できる #ClaudeCode #Anthropic

Claude mods are so fun. Made one for my app Storybox that shows signups, stories created, and an all-time counter right next to my code. Love seeing it go up while I work. Great job to the team @bcherny @trq212 @claudeai

Claude Code CLI new #Mod capabilities are next level. I created a realtime presence detection while i work. It can remotely monitor my breathing and heart beat using #RuView directly in #ClaudeCode CLI cc @bcherny

Congrats to the Claude Code team on shipping Mods! A cute way to make your agent harness extensible and customizable. It's a nice convergence of ideas: in Claude Code, mods are just plugins; and @DeepSeekHarness has been built around rethinking the concept of plugins from the very beginning. While a subset of Claude Code features are made available as mods, in @DeepSeekHarness everything is a plugin: models, tools, skills, sessions, sandboxes, filesystems, loops, orchestration, and UI are ALL implemented as plugins, and can be mixed, matched, replaced, and extended. In @DeepSeekHarness's Creator Mode, you can prompt the agent to inspect the live runtime, write plugins, and persist them, in order to change the UI, modify agent-loop behavior and mechanisms, and add tools or new ways of orchestration — simply everything about the agent harness can be modified on the fly and persisted. So we've been thinking about the term http://Plugin.Engineering for a while… Or should it be http://Mod.Engineering now? 😂

Damn, I turned my Claude Code into a trading desk using Mods now Opus 5.5 + Fable 5.1 are literally a 24/7 AI trading team (result below) Paste it. You're welcome: