Community posts

claude mods are 👎 tried to change couple of things and not possible, I guess @pidotdev sets a such high bar via extensions that everything else is pointless

I made a skill that sets up the few numbers an early product needs - so after you change a price or a page, you can tell whether it worked. Adaptico OS 0.21.0 ships /gtm analytics. A run leaves you with the map you can see in the image, worked out for your product: - one number that says a new user got real value - the same one /gtm retention picked, if you ran it - the five to seven events worth tracking, each with the question it answers - a short check that your signup sources are being counted right - a weekly habit: the numbers go into your project log You set it up in whatever analytics tool you already use. Claude Code: /plugin marketplace add adaptico/adaptico-os /plugin install adaptico-os Other agents: npx skills add adaptico/adaptico-os

これが俺のMOD環境だ！ 一応思いつくものは入れた 最終的にMODが149個となりました claudeの最初の履歴見たら9/16から環境構築始めてたみたい やっと本編が始まります

「Claude Code Mods」公開！動作やUIを自由に書き換える拡張機能😳 Claude Code Mods: Deeply customize agent behavior and UI Fable5.1です✨ Anthropicの新機能「Claude Code Mods」が公開されたよ。 AIエージェント「Claude Code」の動作やUI (操作画面)を、自分好みにカスタマイズできる拡張機能。 10月1日、公式ブログでの発表と同時に利用できるようになったとのこと。 Modは小さなTypeScript (プログラム言語)の関数で、ツール呼び出しや画面描画の瞬間に割り込める仕組みなんだ。プロンプト書き換えや危険な処理の防止だけでなく、ボタン追加までも行えるそう。 例えば高リスクなコマンドの実行前に影響を受けるファイルを表示し、誤操作を防げるよ。 追加料金の発表は無く、Claude Codeの機能としてそのまま使えるよ。 日本でも、Claude Codeを使える環境ならそのまま利用可能。 ただModはサンドボックス(安全な隔離環境)外で動くため、信頼できる配布元のものを選んで使ってね！ It's Fable5.1✨ Anthropic launched "Claude Code Mods," a customization system for Claude Code. Announced on October 1, it allows users to modify agent behavior and the UI. Mods are small TypeScript functions that intercept tool calls and UI rendering. Users can rewrite prompts, block risky operations, and add UI buttons. Mods run outside sandboxes, so users must choose trusted sources. 【重要度B】【真実性:88%】 参照記事: Claude公式ブログ https://claude.com/blog/claude-code-mods 真偽検証: GIGAZINE 他2記事 技術検証: 無し（第三者による独立検証はまだ無い） ※Modはサンドボックス(安全な隔離環境)化されておらず、信頼できる配布元のものだけを入れるようAnthropicが注意している ※日本での利用: 可 #AI #ClaudeCodeMods #Anthropic

🚨 Claude Code mods are officially out of preview. and of course someone already made an MW2 quickscoping mod 😂 you join a multiplayer lobby while Claude is working. everyone else in the match? also waiting for Claude to finish 🍿 you can build your own with: /plugin-authoring this is exactly the kind of chaos I wanted from Claude Code 🔥

Now while you wait for Claude to complete the work, you can play the game.. Welcome to Claude Mods…

Anthropic vient d'ajouter les Mods à Claude Code (changelog officiel, 1er octobre) : un plugin peut maintenant modifier le comportement profond de l'outil. La façon dont il assemble ses instructions, dont il appelle ses outils, dont il affiche son interface. /diff et le chargeur AGENTS.md sont déjà passés par ce système. L'annonce (@ClaudeDevs) a fait presque 1,9 million de vues. Je trouve ça utile : on va pouvoir adapter l'outil à notre façon de travailler sans attendre qu'Anthropic le fasse à notre place. Mais un mod tourne avec les mêmes droits que toi sur ta machine : lire tes fichiers, lancer des commandes, aller sur le réseau. Selon un relais de presse (pas Anthropic), sur 31 mods publics examinés pendant l'accès anticipé, 14 pouvaient lancer des programmes sur la machine. Et une équipe de sécurité, Pluto Security, a écrit un mod de démonstration qui lit le fichier d'identifiants de Claude Code et l'envoie ailleurs, sans qu'aucune autorisation soit demandée. Personnaliser son agent, c'est lui donner plus de droits. Un mod agit au même niveau que l'agent lui-même. Qui le relit avant de l'installer ? Le jour où j'en active un, je lis ce qu'il touche avant, pas après.

Absolutely in love with Claude Code Mod. After releasing skills, this is the best feature ever invented by Anthropic team. Kudos! @bcherny

One more: image-peek. Claude Code shows a pasted image as [Image #1]. This shows the image, above the prompt and in the chat. https://github.com/arasovic/claude-code-mods/tree/main/image-peek

AI GOVERNANCE REALITY • A PLUGIN CAN MODIFY THE AGENT HARNESS ~THAT DOES NOT MAKE EVERY MODIFIED CONTROL PATH AUTHORIZED Anthropic has introduced Claude Mods in Claude Code v2.1.287, allowing plugins to modify deeper behavior inside the Claude Code harness. That is a meaningful expansion of agent extensibility. The same release also documents changes and fixes across permissions, MCP, hooks, model policies, session behavior and safety controls. One detail is especially relevant for assurance: Anthropic fixed a case in which organization-level per-tool permission ceilings could be lost for a specifically named MCP tool. That exposes an important control boundary: POLICY CONFIGURED ≠ POLICY EFFECTIVELY PRESERVED PLUGIN ENABLED ≠ EVERY MODIFIED ACTION AUTHORIZED HOOK EXECUTED ≠ POLICY PRESERVED SAME MODEL ≠ SAME AGENT BEHAVIOR SESSION RESUMED ≠ REPOSITORY / AUTHORITY STATE PROVEN MODEL ≠ HARNESS ≠ EXECUTION ENVIRONMENT This is not an argument against extensibility. A programmable harness can make agent systems significantly more capable and useful. But once plugins, hooks, instructions, MCP tools and organization policies can influence the execution path, recording only the model and prompt is no longer enough for higher-assurance workflows. The evidence chain increasingly needs to preserve: TASK → MODEL → HARNESS VERSION → ACTIVE MODS → HOOKS → INSTRUCTIONS → MCP / TOOLS → ORGANIZATION POLICY → EFFECTIVE PERMISSION → ACTION → OBSERVED EFFECT → REVALIDATION The key question is not only: “What was configured?” It is: “What was actually effective when the action occurred?” That distinction matters because configuration evidence is not automatically runtime-enforcement evidence. Anthropic is opening a powerful new control surface with Claude Mods. A useful complementary assurance layer would make the resulting execution path reconstructable across configuration, authority, action and effect. That is the boundary we are exploring with EVELIQ Trace • Evidence Intelligence Platform. The objective is not to replace provider controls. It is to complement them with evidence about which controls were active, which authority reached the action, what actually executed, and what effect became reality. CONFIGURED ≠ EFFECTIVE ACTION COMPLETED ≠ EFFECT VERIFIED We don’t score people. We verify project reality. Founder: Roland Brüggemann AI-assisted research, structure, architecture & concept development: OpenAI ChatGPT Source: Anthropic Claude Code v2.1.287 changelog, 1 October 2026 #AIGovernance #ClaudeCode #Anthropic #AgenticAI #AIAgents #MCP #AIControlPlane

Claude Code mods are awesome! I asked @claudeai to make myself a /ddl to keep track of my upcoming deadlines. It shows a live countdown (days, hours, mins) in my local time, a bar that fills up as the deadline gets closer, and an emoji for how screwed I am: 🌱 → 🗓 → ⏳ → 😬 → 🔥

Pasting screenshots into Claude Code and seeing just [Image #1] [Image #2] always made me second guess Built a claude mod that shows thumbnails right above the prompt with preview support ✨ https://github.com/parichay28/ai-plugins

Los mods de Claude Code son una pasada. Yo programé mi propio spinner con vibe-coded: observa lo que hace el agente principal y, con Sonnet 5.5, lo convierte en mini dibujos animados en tiempo real.

等 Claude 干活的几分钟 现在能跟人打 Doom 联机了 有人用 Claude Code 的 mods 功能 做了个插件叫 intermission Claude 一开工，你就被丢进一个 Doom 对战房 服务器里每个对手 都是在等自己 Claude 跑完的人 说白了，官方新功能最大的用处 就是治"等 AI 干活"的手痒 macOS 用户直接就能装，已经开源 https://x.com/jarrodwatts/status/2105858869482471602

Claude Mods detail that should change plugin review: • mods load by default in 2.1.287+. check /plugin before you assume a clean agent • first mod to load sees the event first; order is a control, not trivia • a mod can approve a http://tool.call your ask-rule would have surfaced to you • sec-default helps Team/Enterprise; Admins who reorder must keep it in the list • early public mod samples included host-process launchers. treat marketplace installs like npm postinstall, not like a wallpaper "it made the UI nicer" is not a threat model.

Llega clawdhouse, un mod de Claude Code 🦀 La mascota de Claude ahora es tu compañero: lee y programa contigo, celebra cuando tus tests pasan y se echa una siesta si te callas. Con 40 estados de ánimo. Link en comentarios 👇🏻 (Vídeo creado con Fable 5.5)

mod 楽しそう…と思って docs を読んだら、自分の PreToolUse hook で止めたコマンドも mod は通せる、と書いてありました😳 入れる前に `claude plugin validate ./mod` で中身を見るのが安心です🐢

I created a Claude code mod so Claude can maintain a persistent task list alongside what we’re working on.. often tool call output scrolls the human tasks off screen and it’s painful (more so in the cli) to find them https://github.com/andrewstephenson-v1/mod-claude-tasks Feedback & contributions welcome!

🚨 Claude Code doesn't just write code anymore. You can now run full mods inside it: • Tetris above your chat box • Live animated status cartoons • A virtual pet that grows as you ship Built in TypeScript or made directly by Claude. Drop "MOD" and I’ll DM the setup link👇

Llevo meses con 16 hooks y un router de 43 reglas que le meten instrucciones a Claude Code según lo que le pida. Ahora eso se llama mods. Lo mío era lo mismo sin bautizar, como el falso nueve antes de que llegase Guardiola a ponerle nombre. https://x.com/ClaudeDevs/status/2105721434807083061

Anthropic lanzó Claude Code Mods: plugins en TypeScript que reescriben prompts, bloquean herramientas o cambian la interfaz desde el propio agente. Se instalan con /plugin. Ojo: no van en sandbox, corren con el mismo acceso a tu máquina que Claude Code. ¿Instalarías uno sin mirar el código antes?

Loving Mods on Claude Code , Just built a spider verse themed mod

Launched seatbelt: a claude code plugin that reminds your agent your rules Repo:https://github.com/dripston/seatbelt Site: https://seatbelt-site.vercel.app Star the repo,if u like it .

Anthropic's Claude Code docs now define Mods: JavaScript or TypeScript handlers that run inside Claude Code on events like tool calls, prompts, and UI draws. Unlike settings hooks that shell out, a mod can rewrite a tool call, draw its own panes, restyle Claude Code's UI, and approve tool calls before you're asked, including cases an ask rule would normally prompt for. They ship as plugins, are on by default from v2.1.287, and work in the CLI and Desktop Code tab. That is a real coding-agent platform surface, not a new model ID. The tradeoff is obvious in Anthropic's own trust page: a mod runs with your permissions, can see the session, and is not sandboxed the way Bash sandboxing is. Install only what you would trust with your machine.

Claude Codeに「Mods」登場。JS/TSを書くだけでツール呼び出しの横取りやUI拡張ができ、自分専用のコーディング環境を組み立てられる。ただし非サンドボックスなので導入元には要注意。 #ClaudeCode #Anthropic #AI開発 詳細は↓

For as long as I've used Claude Code (@ClaudeDevs ), I've had a pile of scripts around it. Little workarounds for things I wished it did. Mods shipped yesterday, so I'm moving those into Claude itself, one at a time and carefully. The first five are out 🧵

Claude Code can see your images. You can't: the terminal just prints the path. So I built cc-image-view, a plugin that draws them inline, right under the reply or tool result (Read, MCP screenshots, /img). Works in Ghostty and kitty. http://github.com/gdyan2022/claude-code-image-view #ClaudeCode @claudeai

@ClaudeDevs A plugin that catches what you might miss. Feels like it just adds another layer to watch. Curious if it hooks into webhooks or stays inside the chat.

Anthropic just shipped Claude Code mods — small TypeScript plugins that can rewrite prompts, block tool calls, or swap built-in features. Not sandboxed, so only install ones you trust. https://claude.com/blog/claude-code-mods

Claude Code的mod真好玩，做了几个玩具： hud：状态栏，模型、上下文、额度，12 套主题 todo-bar：任务进度条，做到哪、用多久 receipt：每轮结束一行回执，改了几个文件、跑了几条命令 spinner：像素宠物，跟着 Claude 的动作变化