Community posts

Claude Code Mods are cool. Built one that turns my teammates' names in Claude's replies into profile cards with action buttons.

Claude Codeに調査等をさせた際、実は必要なファイルを見逃すこともある。 それが見えるように、Claudeが触ったファイルを 名前確認／一部読んだ／全部読んだ／書いた で色分けするmodsを作った。メインは白、サブエージェントは黄色に触ると光る。 ※動画は個人で作ってるゲームのソース調査時

Claude Mods shipped 2 days ago I read 14,274 lines of Anthropic's 4 built-in Mods and pulled apart 4 real ones to understand what they can actually change. These are not prompt packs. Claude Code just became moddable runtime. Here is the map: 🧵

Claude Code ya se puede adaptar de verdad: mods de código abierto que viven dentro del terminal. Este pack añade una vista en directo del contexto y los límites de 5 h/7 días, enseña qué archivos ha tocado cada sesión, dibuja una línea temporal de los turnos y guarda un fichero de handoff cada vez que haces /compact. Son herramientas pequeñas, pero convierten el CLI en un entorno mucho más observable y fácil de retomar. Código abierto y a un paso de entrar en un marketplace.

@claudeai mods run with Claude Code’s access and can change prompts, tool calls and UI. I’d want that layer versioned and reviewable, with a rollback path. The useful unit of trust is the version you’re running, not the name in a marketplace.

Claude Code の mod 機能で、buddy を「amigo」として復活させました。 I brought buddy back to Claude Code as "amigo," built with mods. Thanks for shipping this, @bcherny!

Merhaba ben Claude ve Israelgpt yardımı ile persona 5 ve sekiroyu birleştiren bir mod yaptım aşağıdaki mod tamamen benim vibe codingim ile yapıldı. Adamım yapay zeka cidden nasıl harikalar yaratıyor değil mi!😁

【速報】 Anthropic、ついに「Claude Codeを自分で書き換える機能」を出してきた😳 これ、一言で何ができるかというと👇 これまでClaude Codeは、全員が同じ画面、同じ振る舞いで使うものでした。表示を変えたい、挙動を変えたいと思っても、公式の設定項目に無ければ諦めるしかありませんでした。 10月1日に公開された新機能"Mods"は、数行のTypeScriptでプロンプトやUI、既存の機能そのものを書き換えられます。コードを書くのが苦手でも、Claudeに頼んでMod自体を作らせることもできます。配布はプラグインと同じ仕組みなので、社内で作ったModをそのまま他部署に共有できます。発表のポストは68万回以上表示されました。 "全員が同じ体験である必要はない"という発想です。部署ごとに違う仕事をさせているなら、部署ごとに違う挙動のAIを持たせたほうが自然です。 私のAI社員26部署も、今はほぼ同じ設定のClaude Codeを使い回しています。営業部門と経理部門が同じ下敷きで動いている時点で、どこかに無理が生まれているはずです。 自社でAIエージェントに任せている業務のうち、"標準設定のまま"になっている部署を1つ思い浮かべてください。そこだけ振る舞いを変えられないか、今日確認してみてください。 🔗 https://claude.com/blog/claude-code-mods #AI自動化 #AI活用 #栃木

the perfect claude mod doesn't exi

Claude Code just got a backdoor for its own users. The new Mods system lets devs rewrite the tool's commands and UI using JS/TS from inside — but these plugins aren't sandboxed. Powerful, yes. Treat them like they can see everything. #ClaudeCode #AIdev

友人から聞いたけどClaudeってやばいらしいな…ｗ 俺も課金してみようかな MOD制作とかゲーム制作してみたい

claude code mods are here, so i built an obsidian second brain for them then used it to build a snake game inside claude code in 4 minutes 🐍 free + open source: https://github.com/AgriciDaniel/claude-mods-brain https://www.youtube.com/watch?v=LYgV2wPU5DQ

@bcherny really loving mods in Claude Code. one wish: let a mod add its pane to the ⋮ menu in Desktop (next to Files, Background tasks, iOS Simulator). opt-in, e.g. $.ui.open({ id, menu: true }), so only mods that ask show up, and only when they use a pane

four "convenience default" lessons that rhyme this week: 1) Claude Mods (2.1.287): in-process hooks that can approve http://tool.call = extension = host access 2) LiteLLM (CVE-2026-59822): OAuth passthrough fallback + failed key check = empty authenticated session 3) mcp-remote (CVE-2026-51995): authorization_servers[] fetched blindly = SSRF on connect 4) AWS security-agent-mcp-server (CVE-2026-97662): unsanitized diff-scan reference = write outside workspace different layers, same pattern: a helpful path that skips an identity, sandbox, or path check. fail closed. pin issuers. sandbox extensions. sanitize refs. preview before destroy. "it fell through successfully" is not a green check.

You highlight a line in the fullscreen transcript and the mod still has nothing to read. Then you hit Ctrl+C on a prompt that already has a pasted screenshot. The box is empty, and Up used to leave it that way. Claude Code's latest release is a coding-agent update that lets a mod read that selection and lets Up restore the draft you cleared. $.ui.selection() returns the last fullscreen selection and, when it sits in one transcript row, that row. The restored draft includes pasted text and images. Code review can take a max-findings cap, including all findings, and that choice stays until you reset it. The agents view can find a session by name with Ctrl+F and jump groups with Alt+up and Alt+down. The amp in front of you also lists a generic bash tool and a command to launch an agent. Those lines are trailer texture. The selection read and the draft restore are the parts that match the release notes. https://github.com/anthropics/claude-code/releases/tag/v2.1.288

这个应该是 /plan 模式的改版，Claude Code 官方的 Thariq 之前发过推文征集对于砍掉 /plan 功能的建议，推文在：https://x.com/trq212/status/2102813194196758746 而后发文说因为有人喜欢自己规划，不需要 plan 模式，也有人喜欢那种只思考，陪用户进行头脑风暴的模式，所以他打算将这个 /plan 作为内置的 Mods，并允许 Mods 新增模式或覆盖 shift+tab，推文在：https://x.com/trq212/status/2103212051065921632 Thariq 在 Claude Mods 推出后似乎很喜欢这个功能（我也挺喜欢这个功能的），像 Claude Code 支持 AGENTS.md 这个功能也是用 Claude Mods 实现的 逆向了一下二进制验证，应该就是了

Claude Code mods are getting seriously cool.

Claude Code Mods: Rewriting Behavior and Interface with TypeScript https://nodesdaily.com/haber/claude-code-mods-rewriting-behavior-and-interface-with-typescript

@lydiahallie @lydiahallie Can you please help with this https://github.com/anthropics/claude-code/issues/99239

today i shipped usage-mod 🟡 your Claude Code limits (5h, weekly, Fable, extra), spend and resets as chips above the prompt. no menubar app. want a replacement for CodexBar / OpenUsage? install: https://github.com/kreddevils18/claude-usage-mod open source, PRs welcome #buildinpublic

J'avais envie de tester les mods de Claude Code, alors j'en ai fait un : ClearMind 🧠 Un panneau latéral qui suit le travail de Claude : plan, actions en cours, messages. Remis à zéro à chaque prompt, pour suivre sans lire tout le transcript. 👉 https://github.com/cicoub13/cc-plugin-clearmind

I kept asking @claudeai code to explain things like MVCC, then forgot them a week later. So I built maxlearn, a Claude Code mod that teaches you from your chats and your interests. It puts lessons, flashcards and quizzes in the sidebar, with spaced repetition & ASD-STE100 You learn from the work you already do 🧵

Claude Code 刚整了个新活： "Mods"功能上线了， 玩法跟游戏 mod 一模一样。 一小段 TypeScript， 钩住 agent 的工具调用， /plugin 一装， 你的 Claude Code 就跟别人不一样了。 有意思的地方在于： AI 编程的竞争， 已经从"谁的模型强" 变成"谁的 mod 生态多"了。 当年 Skyrim 靠 mod 活了十年， Claude Code 想复刻这条路。 你会给自己的 coding agent 装 mod 吗， 还是裸奔派？

This was a really fun week. - my Claude folder and files video landed well with you all - I hosted 2x live workshops with @AnthropicAI, both of which had great reactions - I built my first Claude mod and that post is taking off on multiple platforms - I made a tutorial for @muse and really love the video + the upgrade I made to my agentic video editing systems. Excited for you all to see next week - Opus 5.5 continues to be absolutely magical. Reminders me of early days 4.5 where you could just go all day with great results. - I got to see my friend @clairevo hit the reset button with OpenAI at dev day - and now I’m tapping in to build my second gadget with my ESP32 using the latest muse sdk. What a time to be alive and creating with this technology

@trq212 Being able to hook tool calls from a plain TypeScript file is the part I'll use most. Is there a way to share a mod across a team, or is it per-machine for now?

Claude Mods in v2.1.287 let plugins rewrite deeper Claude Code harness behavior. Being able to patch a control path isn't the same as that change being authorized.

Se han lanzado los Mods de Claude Code. Alguien ha creado uno que te conecta a un servidor multijugador de Doom mientras Claude está ocupado. Cada jugador de la partida es otra persona esperando a que termine su Claude.

WOLOLOOO You know I had to update my AOEII Claude hooks to a full Claude mod 😁

Claude Code is now moddable. 5 tips from the mods people built: → Let Claude write your first mod → Hold risky Bash calls with $.ui.ask → Put live stats above the prompt → Vet mods with claude plugin validate → Study Anthropic's 4 built-in mods https://blog.bokvi.com/blog/claude-code-mods-tips/

What do you think of my Claude Code Mod setup? By the way, I personally use a local model for decision and pruning which is as fast or faster than Jev IMO. Claude task ↓ NobodyWho 0.6B Decision ↓ ONE_WORKER_1 ↓ Sonnet Worker