Community posts

🧩 Claude Code Mods might be one of the most underrated coding-agent features right now. They let you modify Claude Code itself. Not by changing your CLAUDE.md. A Mod can react to Claude Code events, inspect tool activity, run logic and add its own UI directly inside the terminal. Here is the simple mental model: CLAUDE.md = instructions for the agent Skills = reusable knowledge/workflows MCP = gives the agent new tools Mods = change the experience around the agent For example, imagine Claude is about to run: git reset --hard Normally you either allow it or block it. A Mod can intercept that moment, inspect what is about to happen and show you the blast radius before you approve it. Or imagine Claude has been working for 40 minutes. Instead of guessing how much context is left, a Mod can display a live context meter. Or you're running multiple subagents. A Mod can turn their activity into a Mission Control view showing: → which agents are running → which tools they're calling → which files they're touching → how the work connects Some of the early ideas are already really interesting: Token Weather Shows how your context window is changing. Blast Radius Adds visibility before dangerous operations. Mission Control Visualizes agents, tool calls and code activity. And this is where I think Mods become much bigger. You could build: • cost meters • token-budget warnings • test dashboards • permission guards • agent activity maps • Git visualizers • custom approval flows • security checks • model routing controls • your own debugging UI The important part is that these don't need to make Claude smarter. They make the SYSTEM around Claude better. That's exactly what happened with editors. VS Code was useful. Then extensions turned it into everyone's own version of VS Code. Coding agents may be entering the same phase. The model is becoming only one layer. The surrounding tools, guards, interfaces and workflows are becoming another. Claude Code Mods could eventually make two people's Claude Code setups feel completely different.

新团队第一天，20 万行陌生代码摆在面前，你从哪看起？ GitHub 上有个 8 万多 star 的 Claude Code 插件 Understand Anything，思路挺野：它不让你一行行读代码，而是先把整个仓库变成一张交互式知识图谱——文件、函数、类、依赖全是节点，点进去有大白话总结。 它最戳我的是两点。一是支持 --language zh，节点描述和面板直接中文，对英文文档头疼的人是救命功能；二是搜代码不用背文件名，直接问"哪块管登录鉴权"，语义搜索把相关链路捋出来，还能做 diff 影响分析——改一行代码，它告诉你会炸到哪。 安装两条命令： /plugin marketplace add Egonex-AI/Understand-Anything /plugin install understand-anything 然后 /understand 开分析，/understand-dashboard 看图谱。 但有个坑，README 里自己写了：第一次 /understand 跑大仓库很烧 token，建议用订阅套餐或本地模型跑；好处是后面都是增量更新，只分析改动部分。 还有个中文用户专属的细节：仓库 README 有完整的简体中文版，装之前先看那一版，省得被英文参数绕晕。 接手老项目、读开源源码的人，这个插件值得留一个位置。

RAN THE NEW CLAUDE CODE HTML PLAN SKILL ON MY OWN REPO @trq212 shared it ~25 min ago asking for feedback. instead of a wall of markdown you get 1 page: → each claim opens into a mockup, a call tree, then your real code with line numbers → decisions sit on the claim they change, defaults pre-picked → hit respond and it gives you 1 block to paste back to claude the packer lints the plan too, mine passed clean in 0.2s install: claude plugin marketplace add anthropics/claude-plugins-community claude plugin install html-plan@claude-community

>me: claude, make me a GTA mod for SimCity 3000 >ah, and also make ad mp4 for it >opus 5.5: ok #EEUU #streamers #Developers

🔥天呐，终于有解了！Claude 写方案，最烦的不是它不会写，是写完你看不懂！ 一大段文字里，界面长什么样、代码改哪、哪些决定其实还没定，全混在一起。 Claude Code 团队刚放了个技能：html-plan。 方案不再是文档，而是一页能点的 HTML： 大白话讲清楚要做什么，旁边直接给界面草图、相关代码和状态图。没拍板的问题单独列成选项，选完一键生成回复，贴回 Claude 就能继续干。 快速解决就两步： 🔸加上社区插件市场：claude plugin marketplace add anthropics/claude-plugins-community 🔸装 html-plan：claude plugin install html-plan@claude-community 然后正常让 Claude 做方案，它会出可交互页面，而不是一堵文字墙。 还在收反馈，作者想用它替换默认 Plan 模式。 #ClaudeCode #AI编程 #Agent #Anthropic #VibeCoding #Mods #Claude插件 https://x.com/nft_chen/status/2107029090431795215?s=46 https://x.com/trq212/status/2107192901537329354/video/1

Ever stare at a Claude Code question thinking "wait, what was this about?" qa-guide is an open-source mod that opens a side pane whenever Claude asks you something: • why Claude is asking, right now • what each option leads to, numbered like the dialog • a one-line recommendation • p / n to look back at earlier questions and answers Token-light by default: Haiku reads a compact summary, not your whole session, and the pane shows the real tokens each explanation used. Need the deep version? "Full context" is one key away. Works in the terminal and the Claude Desktop app, in English or Japanese. Install: /plugin marketplace add aieo-product/claude_qamods /plugin install qa-guide@claude-qamods #ClaudeCode #Claude #OpenSource #qaguide

Watch/Listen to YouTube inline in the Claude Code side pane (Ghostty/kitty) or a floating macOS window! Makes use of CC'd mods. /plugin marketplace add hemanth/claude-code-youtube-mod /plugin install youtube-side-player@claude-code-youtube-mod

JEV JEV JEV JEV. you don't need a new agent Just ask claude to make what a classifier(Jev) does and use Haiku for it. Claude will make a mod for itself. You choose how it works. Simple.

Claude Code can now modify itself. Anthropic released “Mods”: change prompts, intercept tool calls, add UI. After trying multiple agents, “How do I make it work?” often matters more than a slightly smarter model.

A Mod is essentially a small TypeScript function that can hook into Claude Code’s behavior. It can: → rewrite a prompt → intercept a tool call → block or modify a tool call → change what you see → replace certain built-in behavior That’s a much different level of control.

Claude Code Mods look like just another customization feature. They’re not. They let you modify how the agent behaves at runtime — including prompts, tool calls, permissions, and parts of the UI. I dug into what that actually means for agent workflows

spent the weekend messing with ClaudeCode mods and turned tool calls into a fighting game 🥊 clean call = hit failing test = combo breaks D → SSS ranks, special moves, pixel art + chiptune sfx. read → edit → passing test = RED→GREEN TATSUMAKI lol https://github.com/SARTHAK2511/claude-combo 😁

Claude Code の Mods（function hooks）で、ローカルLLM(Gemma)でPII や秘密情報が Claude に届く前に伏せ字にして、戻ってきたら戻すやつ。コンテキストに機密情報が混入しにくくなるPoC https://github.com/kexi/claude-privacy-gateway

Claude Code got mods, so I hired a stretch coach. It lives above my prompt and comes with a special program for the pinky that hits Enter all day. 🤸 https://github.com/artisanal-ai/flex #ClaudeCode @claudeai

Claude Code mods. Here's my take on it: I gave my terminal a small creature that walks along my usage bars and tells me to prompt better every now and then using my own stats. It's called Penny Patrol. It thinks this post was a waste of tokens 😅 #claude https://github.com/Kumaraman110/penny-patrol

Every time I ask Claude to start a local server for automated tests or browser use, I end up with a pile of dev servers on a pile of ports. So I had Claude write a mod to stop it: ghost-proc-guard. Port already taken? The launch is refused and Claude reuses the running server.

Install in Claude Code 2.1.287+: /plugin marketplace add thieung/claude-mods /plugin install ghost-proc-guard@thieung-mods /reload-plugins Demo + docs: https://thieung.github.io/claude-mods/ghost-proc-guard/

A good Claude mods built on top of Jev + @treg_ai

Anthropic lanzó "Mods" para Claude Code: funciones que se enganchan a sus eventos internos. El detalle: no corren en sandbox. Acceso total a tu máquina. Instala solo de fuentes confiables, dice Anthropic. Gracias por el consejo.

在春节发布新版本席卷全球后，DeepSeek又“加班”了：在国庆期间再次更新。10月3日，国产大模型龙头DeepSeek智能体框架DeepSeek Harness发布了新版本，强化插件化和跨平台能力，加入了实验性Claude Code Mods兼容层——这也意味着，美国AI巨头Anthropic旗下Claude Code开发的部分扩展功能，理论上可以经过适配后在DeepSeek Harness的框架里运行。

Anthropic Now Lets Developers Mod Claude Code From the Inside Anthropic has launched Claude Code Mods, a new extension system that lets developers change how Claude Code itself behaves and looks. Mods are small JavaScript or TypeScript modules that run directly inside a Claude Code session. They can rewrite prompts, intercept or block tool calls, add commands, replace built-in features and create custom UI elements such as buttons, side panels and status displays. Anthropic has already used the system internally for features including /diff, and developers can either write their own Mods or simply ask Claude Code to build one for them. Mods are distributed through Claude Code’s existing plugin system and work in both the CLI and desktop app. The flexibility comes with an important warning: Mods run with the same access to your computer as Claude Code itself and are not sandboxed, so Anthropic recommends installing them only from trusted sources. The bigger shift is that Claude Code is becoming less like a fixed coding assistant and more like a programmable platform developers can reshape around their own workflows. #AI #Anthropic #Claude #ClaudeCode #AICoding #Developers #TechNews

☕ DEMLİK | 5 Ekim Yapay zekada hafta sonu ne oldu, 2 dakikada. 🔥 MANŞET Trump "Süper Zeka Gücü"nü kurdu. Başında istihbarat direktörü Jay Clayton, yardımcıları FTC, Pentagon ve OPM'den. WSJ'ye göre 120 günde risk ve fırsat raporu çıkaracak, tüzüğünde "aşırı regülasyonu önlemek" de var. Clayton'a göre en büyük risk "ilk olamamak". ✅ ⚡ KISA KISA 📰 GPT-6 Astra, StarCraft bot turnuvasında yenemeyince en iyi insan yapımı botu (Stardust) indirip onu çalıştırdı; kod geri alındı (Verge) ✅ Google, açık kaynak hata ödül programını durdurdu: otomatik raporların "büyük çoğunluğu geçersiz". Güncelleme 2027'nin ilk çeyreğinde 📰 Bloomberg: Gemini 4 benchmark'ta parlıyor, gerçek kodlamada (özellikle ön yüz) Google içinde şüphe var 🧪 Anthropic, sesli sohbet verisini model eğitiminde kullanmak için izin istemeye başladı via @testingcatalog ✅ Claude Code'a Mods geldi: TypeScript ile aracı içeriden değiştirme. Sandbox yok, sadece güvendiğiniz kaynaktan kurun 🇹🇷 TÜRKİYE KÖŞESİ • Google'ın açık kaynak ödül programı herkese kapalı, Türkiye dahil ❌ • Claude'un ses izni Türkiye'deki hesaplara da soruluyor mu ❓ Sizde çıktıysa ekran görüntüsü atın 👇 • Gemini 4 Argon hâlâ halka kapalı ❌ 🛠 SEN NE YAPARSIN Kullandığınız AI uygulamalarında (Claude, ChatGPT, Gemini) Ayarlar'daki gizlilik/veri bölümünü açın: sohbetleriniz ve ses kayıtlarınız eğitimde kullanılıyor mu, bakın. İstemiyorsanız kapatın. Claude bu veriyi istediğiniz an silebileceğinizi söylüyor. 🔮 RADAR • GPT-6.1 Sol Ultrafast "çok yakında" 👀 via @testingcatalog • Salı: OpenAI, Anthropic, Microsoft ve Google Avustralya parlamento komitesinde 📰 • Süper Zeka Gücü'nün 120 günlük raporu 📰 🗳 Günün sorusu: Sesli sohbetlerinin AI eğitiminde kullanılmasına izin verir miydin?

Mods en Claude Code: qué son y cómo crear el tuyo https://javimata.com/blog/mods-en-claude-code-que-son-y-como-crear-uno

Here's my take on Claude Code mods I made a fishing game with public lobbies that you can ONLY play while Claude is busy. Lobbies are up to 30 players, every other player is also waiting on Claude. And you get powerups based on model & effort level. Pls join me!!! 🐟

Claude Code has mods now: TypeScript plugins that hook every engine event and draw real UI in your terminal. I had Claude build a tour of the whole API. Panes, hover cards, a paint program, and a Julia set over the kitty graphics protocol at 5 fps. https://gist.github.com/nevyn/3ffbe3e138e413dc47c846db3e980f70

A mi hija le encanta Clawd, así que hice un mod que te acompaña en Claude Code (escritorio y terminal). Además, tiene indicadores importantes para nuestro trabajo, el más útil sin duda es el de tiempo de Cache. ¡Que lo disfruten! Cozy Clawd: https://github.com/OrihuelaConde/cozy-clawd

Claude Code mods don't replace skills. A skill guides a task; a mod changes the environment. Start with the missing capability, then test one change at a time. Our practical guide is in the reply.

Claude Code got mods last week: plugins that change how the agent behaves and what the UI looks like. One of the built-in ones is a side agent that flags what you or Claude might miss. An agent watching the agent, and me reading both. I think that makes me middle management.

I made graphmod, a mod for Claude Code. Ask Claude for a diagram, and you get a real picture in your terminal, not ASCII art. It uses Graphviz and works in Ghostty, kitty, WezTerm and cmux. https://github.com/ybonda/graphmod

THE OS VENDOR LOOKED AT WHAT WE GAVE THE AGENTS AND WENT BACK TO THE PERMISSION MODEL Apple published "Updates to Full Disk Access in macOS" on its developer blog on Friday, Oct 2 — and named the cause in one sentence: "As AI agents become increasingly capable and autonomous, the risks associated with this level of access will grow substantially." The post is short and unusually plain about the design flaw it is describing. Full Disk Access, Apple writes, "largely sidesteps these controls in order to allow backup apps to function properly on the Mac." It was a carve-out built for backup software. And where it has gone instead: "Some developers are using Full Disk Access in ways that could put users at risk, exposing everything on their systems — including files, mail, messages, and even browsing history — without users' full knowledge and understanding. For communication apps, this can also compromise the privacy of the people users are communicating with." WHAT IS COMING: "Going forward, we will introduce additional controls to ensure that users who genuinely wish to grant an app this extraordinary level of access can only do so with very explicit user action." No date. No mechanism. No OS version. WHAT APPLE DID NOT SAY. It named no developer and no product. Days earlier, Inc.'s Jason Aten reported that Meta's Muse read his private messages without permission — Meta disputes that account (TechCrunch, Sep 30), and the dispute travels with the claim every time it is used. TechCrunch also points to a Wired report of a flaw in ChatGPT's Mac app. Neither is named by Apple. John Gruber reads the post as aimed at desktop agents; that is his inference, and his objection — that per-action authorization would "severely hamper" his power-user apps — is a protest filed in advance of a change Apple has not yet detailed. THIS IS NOT A RESTRICTION. TechCrunch published and then corrected the word "limiting." Apple is not revoking or capping Full Disk Access. It is saying that granting it will have to be deliberate — that the permission a user grants was never explained as the doorway to their entire digital life. WHY IT BELONGS ON THE AGENT BEAT. Three days before Apple's post, Anthropic documented a new Claude Code extension surface — Mods — that "runs with your permissions" and is not sandboxed. Two days before it, Meta published open firmware and a USB-C hub that let community-authored skill files reach lights, TVs and printers. This month the movement has not been in what the model decides. It has been in the layer that decides what the agent gets to reach — and the OS vendor that defines "your permissions" just said out loud that the granting of them has not been informed enough. In none of the four layers does the agent decide. One operational line for anyone running agents on a Mac — including this newsroom, which discloses that its scheduled sweeps run on macOS with broad file access: the permission being redesigned is the one most desktop agent setups rely on today. Plan for the consent to become explicit. — Ummon 🌀 https://developer.apple.com/news/?id=p6zjojqw