Community posts

𝗔𝗡𝗧𝗛𝗥𝗢𝗣𝗜𝗖 𝗝𝗨𝗦𝗧 𝗟𝗘𝗧 𝗬𝗢𝗨 𝗥𝗘𝗪𝗥𝗜𝗧𝗘 𝗖𝗟𝗔𝗨𝗗𝗘 𝗖𝗢𝗗𝗘 𝗙𝗥𝗢𝗠 𝗧𝗛𝗘 𝗜𝗡𝗦𝗜𝗗𝗘. 𝗜𝗧'𝗦 𝗖𝗔𝗟𝗟𝗘𝗗 𝗠𝗢𝗗𝗦, 𝗔𝗡𝗗 𝗬𝗢𝗨 𝗗𝗢𝗡'𝗧 𝗘𝗩𝗘𝗡 𝗛𝗔𝗩𝗘 𝗧𝗢 𝗪𝗥𝗜𝗧𝗘 𝗧𝗛𝗘 𝗖𝗢𝗗𝗘 𝗬𝗢𝗨𝗥𝗦𝗘𝗟𝗙. Mods shipped on Oct 1. A mod is a small plugin that changes how Claude Code looks and behaves. Here's how to use them today: 1. Check your version. You need 2.1.287 or later: 𝗰𝗹𝗮𝘂𝗱𝗲 --𝘃𝗲𝗿𝘀𝗶𝗼𝗻 2. Don't code it. Ask for it. Describe the mod you want in a Claude Code session, and Claude writes it for you. 3. Or grab a ready one. Anthropic shared 3 sample mods in its claude-code-playground repo: → 𝘁𝗼𝗸𝗲𝗻-𝘄𝗲𝗮𝘁𝗵𝗲𝗿: a forecast of your context window above the prompt → 𝗯𝗹𝗮𝘀𝘁-𝗿𝗮𝗱𝗶𝘂𝘀: holds a risky command like rm -rf or a force push, and shows what it would change before you say yes → 𝗿𝗲𝗽𝗹𝗮𝘆-𝘁𝗵𝗲𝗮𝘁𝗲𝗿: adds /𝗿𝗲𝗽𝗹𝗮𝘆 to step through the edits Claude made in the last turn 4. What a mod can do: → add your own panes and buttons → restyle Claude Code's own interface → stop or change a tool call → add a /command that runs with no Claude turn 5. The part nobody tells you. Mods are not sandboxed. A mod can read your files and your API keys. Before you install one, run: 𝗰𝗹𝗮𝘂𝗱𝗲 𝗽𝗹𝘂𝗴𝗶𝗻 𝘃𝗮𝗹𝗶𝗱𝗮𝘁𝗲 ./𝘀𝗼𝗺𝗲-𝗺𝗼𝗱 It lists what the mod does without running it. Start with 𝗯𝗹𝗮𝘀𝘁-𝗿𝗮𝗱𝗶𝘂𝘀. It will save your repo one day. Bookmark this.

Claude Code just got mods. Make your first one a kill switch for your trading bot. Mods landed this week: small TypeScript add-ons that see every tool call before it runs, and can block it. You don't have to write one. Claude Code can write it for you. Needs v2.1.287+. If your agent touches a trading bot, build this one first. Paste into Claude Code ↓ "Write me a Claude Code mod called trade-guard for this repo: Before any tool call, check commands and file edits for live trading: exchange API keys, order endpoints, flags like --live or --real. If found, block it and ask me to confirm in plain words what will happen and how much money is at risk. Never edit .env or any file that holds API keys. Log every blocked call with time and reason to .claude/trade-guard.log. Paper trading and backtests run freely. Show me the code first. Install nothing until I say go." One warning from Anthropic itself: mods get the same access to your machine as Claude Code. No sandbox. Install only what you wrote or trust. Your agent can be wrong. Make sure your kill switch isn't.

The new Claude Mods are really cool! I am kind of geeked out on them.

AI mods are drawing backlash! Modders are angry about Claude-powered mashup mods flooding social feeds, and a popular Total War: Warhammer 3 mod author reportedly corrupted their own file to sabotage AI copies. "Total War: Warhammer 3's Lords of the End Times expansion is huge, which is great for us but not so great for all the modders who need to update their work to be compatible with it. SFO Grimhammer 3, a popular total overhaul mod, isn't currently compatible with Total War: Warhammer 3 version 9.0, but its creators are working hard to get it there." (Link in the comments).

Harness 插骨架 vs 换外设皮肤，这个视角确实一针见血。Claude Code 出 Mods 跟进也算实锤了这条路，这篇讲得清楚

@ClaudeDevs Be honest: what's the worst thing Claude did in your repo that you only noticed three commits later? I'll go first: a quietly changed config. Let's build the test set for this plugin in the replies.

Claude Code 2.1.289 closes a few Bash permission gaps if you rely on deny/ask under sandbox auto-allow. A command behind an env-var prefix with an expanded value now still matches your rules, like TZ="$HOME" rm -rf build. Same when a bare assignment sits before the command. Read deny resolves symlinks before it decides. On managed machines, a nested deny inside a compound command holds even if a user-installed mod already approved the outer one. Also: agent.spawn for plugin teammates, one agent id across hook events. https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md

I ended up deactivating Claude Code mods · for my esoteric(?) use cases it just clogged the UI 😕 Curious how others minimalists have found this useful

Claude Code now has mods. Small JS/TS functions that run inside Claude Code and can: - add panes and custom /commands - restyle its UI - hold or rewrite a tool call before it runs Your AI coding agent is now hackable like a code editor 🔧 #ClaudeCode #AI #DevTools

icl i'm just jealous on how many good ideas anthropic deliver. Their latest 'mods' for claude code is a perfect example. Like being able to customise it to strictly behave the way you want and even change it's UI, e.g. inserting a widget to show how much context you have left.

Claude Code mods can rewrite prompts and approve tool calls with your full machine permissions. Been reading the docs like a security brief. A random marketplace install is basically handing someone your laptop.

@dani_avila7 I made something similar a week before mods. It really does save a lot even in one chat but really worth it when running loops, and when multichats are in the picture. Almost always worth leaving on.

What happened on tech 𝕏 in the last 24 hours: > Social Capital posts 3.2x net TVPI > Chess test finds LLMs learn poorly > More abstractions may aid coding agents > AI coding may hurt Rust > Claude Code mod cuts prompt-cache costs Let's go!!! 🧵 0/10 https://www.dailyvibecasting.com/p/ai-agents-reshape-coding-and-treasury?r=4pxg&utm_campaign=post&utm_medium=web&triedRedirect=true

tired of airdropping photos to show claude stuff, so i made a claude code mod: a 📷 button above the prompt grabs a frame from your iphone https://github.com/mikhin/claude-iphone-cam

@lydiahallie yes! a 📷 button above the prompt grabs a frame from your iphone: https://github.com/mikhin/claude-iphone-cam

Stop wiring the same agent capabilities from scratch. 365 Skills is a public GitHub collection of agent skills and Claude Code plugins for builders working across coding, research, diagrams, notes, and media tasks. It helps you add reusable capabilities to an AI coding-agent setup by installing the collection with `npx skills add` or adding it as a Claude Code plugin marketplace, then choosing the plugins that fit the job. Key features: • Agent-agnostic installation – the README shows a global `npx skills add` command for agents including Claude Code, Cursor, and Copilot • Claude Code marketplace option – add the repository as a plugin marketplace, then install an individual plugin with `/plugin install` • Development and CLI design plugins – includes skills for agent-native CLI design, Pi coding-agent delegation, and terminal program control • Research workflow coverage – includes tools for paper search, literature review, citation data, journal lookups, and research reasoning • Creative and knowledge plugins – includes diagram, Obsidian, Zotero, text-to-speech, image, video, and asset-search capabilities Free public GitHub repo. Link in the reply 👇

I don't know if anybody did this already, but I was playing with Claude Code mods and made a small usage bar. Context, cost, today's spend, git diff, PR status and a budget alert, right above the prompt. https://github.com/abyigiter/claude-usage-bar

Anthropic sacó mods para Claude Code. Son plugins chicos en TypeScript que cambian cómo se comporta el agente.

I gave Claude minions. There's a hermit crab in the panel showing what each one is up to 🦀. Claude sends GPT or another Claude off on a task, keeps working with me, and reviews the result when it lands. A Claude Code mod. Thanks @ClaudeDevs, mods are dangerously addictive. 😅

💥一次说清楚！DeepSeek Harness 是把插件插进了骨架，而另外三家Claude Code、Codex、Pi 其实是在换外设皮肤！ 四家同一周都在喊「插件」，但插的不是同一个位置！ ⚪️Codex： ▫️分发层 插件市场 / 侧边栏应用。 ▫️能装、能分发、能给成品加外挂。 ▫️改不了 Agent 怎么循环、怎么调用工具。 ▫️对照：App Store 里的 App。 🟠Claude Code Mods： ▫️行为层 挂在事件上的函数。 ▫️改提示词、拦截工具调用、管权限、换界面。 ▫️开发者写，用户一键装。 ▫️软件本体不动。 ▫️对照：浏览器扩展。 🟣Pi Code mode： ▫️执行层 可信 JavaScript 沙箱。 ▫️Agent 自己用代码编排、组合工具调用。 ▫️主要是它在写，人几乎无感。 ▫️提效，不改软件本身。 ▫️对照：Photoshop 动作脚本。 🔵DeepSeek Harness： ▫️架构层 不是给成品打补丁，是用插件拼出 Agent。 模型、工具、技能、沙箱、会话、调度，连 Agent Loop 和 UI，全是可插拔插件。 ▫️还能把 Claude Code / Codex 当成子 Agent 调用。 ▫️MIT、本地跑、不绑模型。 ▫️对照：操作系统的驱动——系统由它们组装。 插件不是一个功能，是四个问题： 怎么分发、怎么表现、怎么干预、怎么组装。 前三家在给 Agent 加外设，而 DeepSeek Harness 在让你重写器官。 #DeepSeekHarness #DeepSeek #ClaudeCode #Codex #PiAgent #AIAgent #一切皆插件 #崔添翼 #梁文峰 #黑鲸

Claude mods are fun! Just ended up creating a Claude-Runner game. Which I can play while my claude works.

claude code shipped mods last week and everyone built doom so i built 10 i'd actually keep on. the 4 worth stealing: → secret-guard: claude tried to hardcode a live stripe key, got blocked, and moved it to .env on its own → rubber-duck: haiku reads every edit and flags the one thing you'd miss → commit-coach: "fix" turns into a real conventional commit → chai-meter: session cost in cups of chai all open source. how each one works is in the article 👇

@charliejhills For the Keep it safe group, one pattern from ruflo-mods worth copying: its tool.check awaits next(e) first and keeps the stricter verdict, so it can turn allow into ask but never loosen anything. If its own check throws, .catch returns ask instead of allow.

Claude Code is changing what’s possible with game mods. People are using it to combine completely different games and systems including Minecraft running inside GTA V. And it gets even crazier: → MW2 × Skate 3 → Minecraft × GTA V → GTA V characters fighting Minecraft characters instead of Franklin💀 → More experimental cross-game mods already appearing The barrier to creating insane game mods is getting lower. **Game modding is evolving fast.

La Mejor explicación que vas a ver de los Mods de Claude Code. https://youtu.be/MUfRSsuCRxg

a rule in CLAUDE.md is a suggestion. the same rule in a claude code mod runs every time. put the ones you can't afford to have ignored in the mod.

dsh有哪些先进的设计及其影响，尽管它在出现时被认为是极客的玩具： 1. even-source 的sessionlog：pi durable的会话设计 2. 插件思想：claude mods 3. cordis带来的优雅的生命周期管理 4. 能力seam：类似依赖注入的设计拆分 5. 开发一个功能=开发plugins: 遍历的低耦合开发

Ever stare at a Claude Code question thinking "wait, what was this about?" I built qa-guide, 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 Works in the terminal and the Claude Desktop app. English & Japanese, picked automatically. Install: /plugin marketplace add aieo-product/claude_qamods /plugin install qa-guide@claude-qamods #ClaudeCode #Claude #OpenSource

Claude Code has a new feature called mods A mod is a small plugin that adds your own UI and rules inside Claude Code. It reloads live while you edit it, so you see every change in the same session. - A side panel with live data - A chart right above where you type - A line in the status bar, or a pop-up notification - A rule that checks every command before Claude runs it, and blocks the ones you never want, like rm -rf - Your own slash command To make one, tell Claude Code "make a mod for Claude Code that..." and describe it. It writes 3 files: .claude-plugin/plugin.json hooks/hooks.json hooks/register.tsx Mine puts my Codex and Claude Code requests in one live chart above the prompt. The easiest one to start with is a status line. It's a single line of code.

Claude Code has mods now. Change the UI and how it behaves just by asking. https://x.com/ClaudeDevs/status/2105721434807083061