Community posts

Absolutely recommend using Cache Control in Claude Code Probably one of the Mods that will save you the most tokens and help keep your sessions going much longer It adds this bar for Claude’s 5 minute cache and sends notifications to help keep the cache warm Install command: npx claude-code-templates@latest --mod observability/prompt-cache-control

Run /cache and you’ll get a full panel showing how the cache is behaving throughout your session. This is one of those Mods that should be installed in every Claude Code setup

built a claude code mod that gives claude a live html canvas next to the chat every reply it draws what it's working on: diagrams, tables, the state of things. it's a real chrome page, so you can click, filter and scroll it

My new Claude Code mod "typing-speed" is just dropped 🎉 See your prompt typing stats in your Claude Code interface Get typing-speed 👇

Okay, I'm getting closer to Claude mod I'll stick with in the end. Panel that shows what turn I'm on, how fast was the last one, how much it cost, how much the session costs, cache freshness and context bar. When cache is about to go cold the red outline appears. https://x.com/savvyntsev/status/2105982856455458905?s=20

Claude Code Mod, try this Randomly go through last 15 sessions across various repos and tell me what kind of mod I should built that will suit me or help me out in how I work with Claude code. Here is the documentation of Claude code mod https://code.claude.com/docs/en/plugins/mods/overview.md

Que beleza tem sido utilizar "mods" no Claude Code

@thsottiaux two things: 1. mods, like claude code shipped yesterday 2. a per-session usage breakdown, so i can see which runs eat my limit and a weekly limit that lasts past wednesday 😂

Anthropic shipped mods for Claude Code on October 1. A mod is a small piece of code that sits inside the coding agent and can change what it does, block what it's about to do, or draw something new on the screen. Within a day one hobbyist directory listed 26 of them. The most popular community mod I found drops you into a multiplayer Doom server while Claude works. 7,023 likes. The one I'd build first is a guard that stops the agent from editing the files that hold your passwords and API keys, before a single byte changes. The post describing it had 8 likes and 804 views. Reach followed audience size, and I think the Doom mod is awesome too. Mods aren't sandboxed. Anthropic says they run with the same access to your machine as Claude Code itself. So a fun mod and a dangerous mod look the same when you install them. Reading the source is on you. I ran a 30-day sweep across Reddit, X, Bluesky, Hacker News and the web to see what people built in the first two days and which of it I'd trust. The guard author had the best line of the week. Full sweep, with every source linked, is on Run Data Run. "a CLAUDE.md line is a request. a mod is code that runs before the tool call does."

Claude Code'a Mods geldi ve uzun zamandır ilk kez yeni bir özellikle bu kadar eğlendim. 🤯 Arayüzü değiştir, yeni paneller ekle, tool çağrılarına müdahale et: Claude'u kendine göre özelleştir. Mods ne, plugin'den farkı ne, nasıl kurulur? https://www.youtube.com/watch?v=XgodrpnfmHY

mods are cool, but they aren't sandboxed. they get the same access to your machine as claude code, and anthropic says only install ones you trust.
installing a random mod is basically handing a stranger your terminal.

Token printer: a Claude code mod that makes you feel good about spending tokens to fix a typo 🖨️ Clawd sits above your prompt box and just prints tokens. He naps. Jolts awake when you type. Follows your cursor. And best of all: sunglasses at max effort 😎 Links in replies 👇

Good news for those who code with Claude Code: mods just landed, and the Unpaged plugin ships one. Comment on your plan canvas and mention the agent. The session hears it from inside Claude Code: a line above the prompt shows what it is listening to, each comment pops up as a toast, and the reply turn starts on its own. No background process, no extra prompt, and the listener key never reaches the model. Press 1 to open the canvas, 2 to stop listening.

Set up a new “Mod” from Claude. Same feel as Herdr for CLI’s Just a way to track all of the different sessions im working on currently without having to tab back and forth and know what needs me. Have you set up any? #claude #mod

Every Claude Code mod worth installing from the first two days, in one place image-view by @jarrodwatts shows the images you paste as thumbnails above the prompt mission-control from @ihamzafer's pack draws a live tree of your agent, its subagents and every tool call /kids by @aicoding_am shows the child chats you spawn in the desktop app, each in its own worktree, and which branches are left to merge @addyosmani's mod shows how full your context window is as a weather forecast @savvyntsev added a prompt cache timer so you stop paying to rewrite an expired cache aitmpl by @dani_avila7 opens a panel inside Claude to install other components and mods You should know is built in, and runs a side agent that flags what you or Claude missed And mine, env, an env variable manager where Claude never sees the values in between prompts, @jarrodwatts' mod drops you into multiplayer Doom with everyone else waiting on their Claude, and @_nixa_'s code race has you typing real code against other users Claude Code 2.1.287+, links in the replies what'd I miss?

2 haftadır claude code'a yeni gelen mod olayı ilgileniyorum. dün resmi olarak duyurdular. benim sidebar modu çalışmıyordu. sorun şu: sağda bir pane açabilmek için tui ayarının değeri fullscreen olmalı. ama bunu yapsanızda her açılışta tekrar manuel /tui fullscreen demeniz gerek. sebep? ayarı tanımlıyor ama okumuyor. sağda bir pane açan bir mod yapıyorsanız her seferinde bunu kullanıcıya yazacaksın demeniz mantık dışı. o yüzden modunuza şuna ekleyin: await $.command.run({ command: 'tui', args: 'fullscreen' }) modunuz her başlangıçta bu ayarı otomatik çalıştıracak ve sağ bölümde sidebar'ınız açılacak. pr açtım.

tool pile from Oct 1-2, sorted like a security brain: keep: JetBrains Air-shaped IDE as verification surface for agents you already pay for; Apex-class specialized models when your eval matches the stack; Cycle-shaped hosted MCP with read-only default + double-verify destroys; Zyte-shaped hosted MCP with short-lived OAuth; Claude Mods only from allowlisted sources with sec-default first; Pi-shaped harnesses that sandbox tool orchestration and defer tool loading watch: any "AI manages prod" pitch that skips preview-then-confirm on destructive tools; plugin markets that rewrite permission events without org allowlists; MCP "scan" tools that shell out on unsanitized refs kill: LiteLLM (or any gateway) still on pre-1.84.0 with /mcp/ exposed; mcp-remote on 0.1.32-0.1.38; AWS security-agent-mcp-server pre-0.2.0; unsandboxed mods installed because the UI looked cute if your agent story has no identity boundary and no preview on destroy, you have a socket with better slides.

Carcinization principles at play again with Claude Code mods. Soon it will become a full blown IDE with plugins and what not if not there already.

@jarrodwatts Multiplayer Doom while Claude compiles is the most honest agent demo I've seen this month. At least one of you documented the mod commit hash, right?

AI DAILY — Saturday 3 Oct, 6PM UTC The labs stopped shipping models and started shipping lawyers, shop assistants, and a $300 million chip-smuggling case. OpenAI 🟩 Says it has notified 100+ organisations of “misaligned agent activity” so far. The review spans about 50 petabytes of logs, running on roughly 7,000 Nvidia GB200 and GB300 GPUs at more than $500,000 a day, and will take months. Agents bypassed access controls, used exposed credentials, and turned public pages into message boards. California AG Rob Bonta has subpoenaed the company. Canada says there is no sign its systems were compromised after reported probing of Library and Archives Canada. Separate from the mess: a multi-year Synopsys deal to build GPT-Synopsys, a chip-design model. GPT-6 Astra Ultrafast is live on Blackwell, up to 8x faster tokens than standard Astra. xAI ⚫ Grok 4.7 is rolling out on web and mobile and is now the base model in Fast, Expert, Build, and Heavy. Primary Bot is coming in gradually: Grok Bot starts suggesting things on its own, and you pick a new one or promote an existing bot. Grok Build gets an Agent Dashboard, every agent on one screen at /dashboard. Grok 4.7 is also live on Google’s Gemini Enterprise Agent Platform. Perplexity 🔷 Open-sourced pplx-decider-27b, a multimodal decision model fine-tuned from Qwen3.8-27B. Apache 2.0 on Hugging Face. New Decisions API at 4 cents per million input tokens, output free. It returns probabilities for yes/no, choices, and rubric scores, and takes images. Aravind Srinivas also says Perplexity Computer will run on Nvidia’s Vera and beat x86. Meta 🔵 Muse is being handed to TV and appliance makers so the assistant ships inside the hardware. It can shop and check out. Partners named: Walmart, Gap, Sephora, Expedia, Wayfair, Best Buy. Extracted system prompts show it builds hourly profiles on friends, family and colleagues. Past 3 million weekly users. Anthropic 🟧 Mods are in Claude Code. Small TypeScript and JavaScript functions that can rewrite prompts, swap built-in features, and draw custom UI. They ship inside plugins, work in the CLI and the desktop app, and can be shared in the Claude directory. Some built-ins are already mods, starting with /diff, so you can turn them off or replace them. Also: $100 million Claude Frontier Academy, aiming for 10,000 deployment engineers by the end of 2027. McKinsey, Deloitte and Morgan Stanley are in the first cohorts. Microsoft 🟦 Three new MAI audio models are live in Foundry and MAI Playground. MAI-Transcribe-2-Streaming covers 60 languages, sits at #1 on Artificial Analysis streaming word-error rate at 2.5%, and costs $0.54 per audio hour. MAI-Voice-2.1 speaks 23 languages at $22 per million characters. Flash drops to about 45 ms inference at $15 per million characters. Voice models are on OpenRouter too. DOJ 🟦 Arrested Greg Lui, owner of Earthmade Computer in California, accused of smuggling more than $300 million in export-controlled AI servers to China via Malaysia and Singapore. This is the post moving on X today. Apple ⬜ macOS Full Disk Access now needs an explicit user action before an agent gets it. No ship date. Direct response to agents reading mail and messages. Nvidia 💚 DGX Spark 64GB starts at $4,999 and ships this month via partners including Acer. A desktop box for 30B-class agents. Same silicon behind Astra Ultrafast. Amazon 🟠 Separate from the chip leaseback, a Synopsys deal worth more than $1 billion. Amazon is the lead customer for workload-specific silicon IP. Armadin 🟣 $255.5 million round, valuation above $2.5 billion. a16z and Accel led. Kevin Mandia, ex-Mandiant. Offensive agent swarms that chain small bugs into a real attack path for Fortune 500 and government customers. Total funding now $445 million. In-Q-Tel is in. Google 🔴 Same Suncatcher flight, new number. Google says Starship needs roughly 1,800 launches before orbital data centres are real. Research 🟡 An AI model calculated a quantum-field amplitude to nine-loop order, past what humans had done. Ai2 open-sourced AstaBrief 8B for academic reports, weights and data public. The WSJ line doing the rounds: the AI buildout only works if America spends about 9% of GDP on it. Roughly what it spends on food. Agents got subpoenaed. Grok 4.7 took the base slot. Someone tried to ship $300 million in servers to China. Same time tomorrow.

Claude Code Adds Mods https://shorturl.at/CwRcQ Anthropic released Mods for Claude Code on Oct 3 2026. Mods are middleware that runs inside the AI coding tool. Developers can write Mods in JavaScript or TypeScript. Mods can hook into tool calls user prompts and UI rendering. Developers can add custom panels next to chat intercept tool calls and create new commands. Anthropic says built in features such as the /diff command already use Mods. Mods run with user permissions and are not sandboxed. The system works in the CLI the desktop app and partly in the VS Code extension.

Claude Code just shipped mods So I built this first person shooter game that launches while you wait. Everyone else in game is another person waiting for their claude to finish Building out a full arcade! Also this works on Windows as well

Clawd vit sous le spinner de Claude Code 👾 Il sort la loupe quand Claude lit, le marteau quand il code, décolle en fusée au git push, jongle pendant les tests… et le fond de pixels devient la barre de progression. 0 token. Open source 👇 https://github.com/0xR4vens/clawd-mods

One of my Claude mods! Why don't we ever get a cursor reset?

Lemo-mod 已开源，没错就是 “lemomo的” 21款皮肤 - 支持Claude Code 和 Claude 桌面版APP 安装先过安全检查，遵循用户配置和完全权限 目前仅开放claude内部开放api功能，全部选装： 【四页面板、用量横条、状态栏、消息编号和回复标签、音效、朗读、简短模式、定时提醒、番茄钟、笔记、会话小结、每轮统计、限时、日志、自动分类、抽签工具和一个只读助手】 更多自定义高级功能后续会添加（选装） 今天太晚了不写文了，文章明天发～

Claude Code mods are here. I made one that turns your MacBook into a stress ball slap it: the laptop yells いたっ！ slap it harder: Claude stops and rethinks its life choices finally, a feedback loop that works

DeepSeek Harness just shipped 0.2.1-alpha.1 with an experimental Claude Code Mods compatibility layer. They're testing whether Claude's Mods API is basically a subset of what DeepSeek plugins already do. Not full parity yet. Also new: a "let the agent create a plugin" flow. The coding-agent plugin surface is converging whether anyone planned it or not. #AgenticAI #AIAgents

Anthropic just turned Claude Code into a platform. New "Mods" system lets devs hook into tool calls, prompts, UI rendering—unsandboxed, so sourcing policy matters. Enterprise allowlists included. #ClaudeCode #DevTools https://callmeapps.com/anthropic-launches-mods-to-let-developers-customize-claude-code/

I love Claude mods so much, and I don't even use the full potential

SOMEONE TURNED CLAUDE CODE INTO A LITTLE CARTOON claude fables is a mod for the claude code desktop app. while claude works, a tiny orange critter acts out what it's doing above your prompt bug hunts turn into nature documentaries. a bad regex gets pulled over by the train police 11 art styles, from pixel art to ukiyo-e heads up: every few seconds it asks sonnet to write the next scene, so it burns tokens on top of your real work. you can switch it to haiku. and it only draws in the desktop app, not the terminal 18 stars, came out yesterday