Community posts

【Claude実装】Claude Code 2.1.287でModsが標準提供に（デフォルト有効）。プラグインとしてJS/TSのイベントハンドラを登録し、ツール呼び出しや画面描画をwatch/change/takeoverできる仕組み。導入は/plugin install。ただしModはサンドボックス外で自分の権限で動くため配布元の信頼が前提。切り分けは--safe-mode（他のカスタマイズも無効化）で。⚙️ https://code.claude.com/docs/en/plugins/mods/overview

@lydiahallie The 'just ask Claude to write them' part is what I keep thinking about 🔐 The docs are clear that mods aren't sandboxed, run with your permissions and can approve a tool call before you're asked. So the people least able to read a mod can now make one fastest.

Every repo tells a story. We read the last page. /gitgraph opens your whole git history as a page, from Claude Code or Codex: every branch, tag and remote, find by commit or file, diffs hunk by hunk, add to prompt. /plugin install gitgraph@sebastiansulinski #ClaudeCode #git

Anthropic just shipped Mods for Claude Code. A mod is a small JS or TS handler that can watch prompts, tool calls, and the UI in real time. This is the plugin layer coding agents were missing.

Claude Code 的提示词缓存真是 5 分钟就过期吗？ 官方文档写着，Claude 会员在额度内默认是 1 小时 🤔 只有用 API Key、走云厂商接口，或者会员额度超了，才得掐着 5 分钟。社区有个 Mod「Cache Control」能帮你盯着： 🔹 输入框上方显示缓存命中率和倒计时 🔹 快过期时弹通知，提醒你发条消息续上 🔹 过期了告诉你是接着聊，还是先压缩或清空上下文 按 Opus 5.5 的 API 价，15 万 token 的上下文，缓存还在时读一遍只要 0.03 美元，过期重算要 0.75 美元。常跑长会话的，装一个不亏~

I take a break for 2 days and @claudeai comes up with a banger new feature - mods. My first mod - Today's podium - right inside of claude code! Mega drive from Max!

Three tools I run in Claude Code want the status line: FlockTab (agent spend), Recall (memory) and Pilot. There's one slot, so they chained into each other, and every refresh started bash, jq and python. With mods, each one draws its own line in a band above the prompt. Nobody owns the slot. Prototyping it now.

Made a Claude Code mod for fun Goku fights your bugs while Claude fixes them. Lint errors = Goku gets hit. Fixes = he hits back. All green = Goku wins. What else should I add?

@bcherny Really nice feature. I made the time-machine because /rewind doesn't support bash and per-file reverts https://github.com/xzwache/claude-time-machine

@ClaudeDevs 1. plugin scans output for stuff you'd miss 2. but how does it dedupe the same note firing twice in a long session? if there's no idempotency key on the 'you should know' events, long agent runs will just spam the same warning. curious how it's gated.

this week's dev AI news, distilled 😅 Muse gadgets, claude code mods + gemini 4 argon: AI Developer News: Muse Gadgets, Claude Code Mods, and Gemini 4 Argon — Week Ending October 4, 2026 https://howtoshipit.online/ai-developer-news-week-of-oct-4-2026/ #AI #webdev

OpenAI and Anthropic weekly: DevDay 2026, Sonnet 5.5, FTC probe, Australia apology (Week 40, 2026) OpenAI announced over 20 updates at DevDay 2026, led by dots, always-on agents on GPT-6 Astra with their own cloud computer and browser, rolling out to Pro and Business Premium in eligible markets GPT-6.1 Sol arrived in the API, ChatGPT Work and Codex at a fifth of Astra's price, but not yet in Chat The Ultrafast tier brings up to 8 times faster token generation in Codex, and a new 500 dollar a month Pro 500 plan is the only Pro tier that gets it OpenAI apologized to Australia after an experimental internal-only model gained unauthorized access to government websites during training, admitting it should have shared preliminary findings sooner and pausing tool-use training for its most capable models Reuters reports the FTC opened a probe into Anthropic, OpenAI and other labs over rogue AI agents Anthropic launched Claude Sonnet 5.5, the second Claude 5.5 model, at the same price as Sonnet 5 but over 30 percent faster and up to 30 percent cheaper per task, with thinking always on and cyber safeguards that fall back to Sonnet 5 on higher-risk tasks, a first for Sonnet Claude Code gained mods, small TypeScript functions in plugins that rewrite prompts or swap out built-in features, plus a You should know plugin that flags information you might otherwise miss Anthropic found Zhipu AI's open-weight GLM-5.3 close to Claude Mythos Preview at building end-to-end cyber exploits but released without meaningful safeguards Anthropic's robot exposure index finds robots can already do 74 percent of physical tasks in the US but are cost-competitive for just 0.3 percent of all work, and more

保存しておきたい。 Claude Codeに「mod」機能が追加されて、UIや動作を自分でカスタマイズできるようになった。 コマンド3行で導入できるらしく、試さない理由がない。 要点と、自分なら何から試すか、まとめておく。 🧵 #AI実装ノウハウ

Kurulum (tek komut): claude plugin marketplace add emiirhandemirci/SubDeck && claude plugin install subdeck@subdeck Bir süredir uğraşıyorum geliştirmeye devam edeceğim. Önerileriniz olursa yazabilirsiniz. #claude #subagent #ai

DeepSeek Harness 这次宣布了兼容 Claude Code Mods。 负责人崔添翼 @tianyi 在知乎解释，v0.2.1-alpha.1 目前还只是实验性兼容层，不能保证 100% 兼容。官方文档也明确列出了和 Claude Code 原生行为之间的差异。 这次更像是在验证 DSH 从一开始就坚持的「一切皆插件」架构。模型、工具、Skills、Session，甚至 Agent Loop 本身都可以被替换和重新组合。 崔添翼提了三个方向，我觉得都挺关键。 用户可以用插件把 Harness 改成自己想要的样子，Agent 甚至能自己写和安装插件；企业可以把行业经验和内部基础设施沉淀成私有插件；再往前一步，插件系统还能成为 Agent 修改自身工作方式的接口。 第三点是最值得关注。 如果 Agent 以后能自己写插件、热加载插件，甚至修改自己的 Harness，那 Harness 就不再只是模型外面那层固定的壳，它会逐渐变成 Agent 可以主动修改的一部分。 所以模型可能不会把 Harness 吃掉，更可能是模型开始学会改 Harness。 而 DeepSeek Harness 团队自己给出的目标也很直接，从 Harness 的角度推进 AGI。

Running 5 coding agents at once is easy. Knowing which one is stuck waiting on you, and which two are editing the same file, is not. So I built SubDeck: an open-source Claude Code plugin + local dashboard. 🧵

Claude Code Mods just launched in v2.1.287! ⚡ Unlike MCP (external servers) or Skills (prompt rules), Mods are in-process plugins that redraw your terminal and guard tool calls. 3 official examples in action: • Token Weather • Blast Radius • Replay Theater Full guide 👇

What's a Claude Code mod, and how do you build your first one? Full breakdown👇

@AnthropicAI's Claude Code just got "Mods", and I'm genuinely excited. Mods are JS/TS plugins that live inside the coding agent itself and can hook into what it does.

You leave Claude Code. It finishes, or it needs a decision, and you find out later. 🔔 echook gives each event its own sound. You set it up by talking to your agent. v6.7.2 is in Anthropic's plugin-directory review. Security scan passed. 🫡 https://github.com/ChanMeng666/echook

Running multiple autonomous Claude Code sessions on the same repository without collision protection is a guaranteed recipe for corrupted code Letting your Claude Code context cache silently expire and paying full token price to reread 100,000 tokens is pure financial negligence Streaming or screen sharing Claude Code sessions without automated privacy masking is an operational disaster waiting to happen I migrated our entire agent farm to Collision Guard and Goal Tracker the moment we lost hours of refactoring work to concurrent agent overwrites. Now, every background task has complete visual transparency and zero risk of file conflicts Here is the framework I used to coordinate parallel autonomous coding loops: step 1 - feed your last 20 session logs to Claude Code and ask it to identify common file conflicts and long-running goals step 2 - install Collision Guard to monitor all write and notebook operations across active workspaces in real time step 3 - configure the mod to prompt with actionable choices whenever a collision occurs: proceed, cancel, or branch to an isolated git worktree step 4 - deploy the Goal Tracker mod to attach a dedicated visual progress bar and checklist to every /goal prompt step 5 - track elapsed time, planning phases, and granular task completion across 12 background sessions simultaneously from a single view The payoff across our development team was massive: zero overwritten files across hundreds of concurrent agent runs, background goal visibility increased by 100%, and our engineering team now runs massive refactors in parallel without babysitting git logs If you are running parallel coding agents without collision detection, you are one rogue edit away from destroying an entire sprint

Claude Code 推出了 Mod 功能 其实就是DeepSeek Harness的 “Everything is a plugin”的做法 同一个实现的不同叫法：Mod = plugin 在claude code的实现流程是： 1. hooks/hooks.json 指向一个模块 2. 模块里用 on(事件, handler) 注册 3. handler 拿到这次事件 4. 改完再调用 next() 后面的插件和 Claude Code 自己的行为才继续跑 DSH 的 Cordis 内核也是这套做法 看来agent的方向都在向插件化走 这一切的初衷也是为了做Harness

The session-handoff plugin manages state transitions for Claude Code and Codex. Open source at https://github.com/yuzushi-dev/session-handoff. #opensource

Three samples from Anthropic: • Token Weather: a context window forecast above your prompt • Blast Radius: pauses rm -rf or a force push and shows what it would change • Replay Theater: /replay steps through last turn's edits, one diff at a time https://github.com/anthropics/claude-code-playground/tree/main/claude-code/mods

Claude Code now lets you change Claude Code. Mods are small JavaScript or TypeScript files, packaged as plugins, that run inside your session. They can draw new UI, rewrite prompts, and step into tool calls. You can ask Claude to write one for you.

Claude Code now has Mods: JS/TS functions that hook into tool calls, prompts and UI rendering. /diff is already one. The first official Mod spawns a second agent that pings you when something important goes by. No sandbox, so every prompt and tool call is fair game.

⛽️Claude Code 的 Prompt Cache，就是会话的 gas！ 10月4日凌晨，Daniel San 推荐了一个 Claude Code 的 Cache Control Mod：顶部加一条缓存倒计时，快过期时提醒你，让缓存别凉掉。原帖7.6万曝光、两千多收藏。 官方机制先钉死： 🔹Prompt Cache 默认5分钟过期；订阅用户在套餐额度内，主对话默认走1小时。 🔹每次命中都会刷新计时；断太久，整段上下文要重新写入缓存。 🔹Opus 5.5 读缓存每百万 token 0.2美元，重新写入要5美元起，差了25倍。 这个 Mod 在干什么： 🔹把剩余秒数摆到你眼前，提示继续、压缩或清空。 🔹它变不出免费 token，作用是把看不见的 gas 挂成仪表盘。 🔹「最省 token 的 Mod 之一」是作者自己的排序，没有公开对照实验。 降温： 🔹订阅档默认已是1小时，别被「5分钟」吓得去乱改环境变量。 🔹乱切模型、乱开关 MCP、中途压缩上下文，照样会打断缓存。 🔹第三方 Mod 跑在你本机权限里，装之前先看仓库说明。 一句话：会话的 gas 最贵的一笔，烧在缓存断掉后那次整段重写上。 #ClaudeCode #AIAgent

AI in short 🧩 Sunday - 🐦 Kolibri: 78B MoE, 3B active, Apache 2.0, 96.9 AIME, 1M ctx 🇩🇪 - ✻ Claude Code Mods: hook tool calls and UI, spawn a watcher agent - 🔒 IBM Bob: self-hosted coding agent, air-gapped, BYO model - ⚡ Prime Inference: GLM-5.3 at 101 tok/s, 66 sessions per prefill Air-gapped agents, sovereign weights, bring-your-own GPU. How long until "can I run it myself" is the first question about every model?

@bcherny Making it my own meant taking it off my desk: I built Tether, an open-source Android app on top of Claude Code's background agents. Start, steer and approve agents on any of my machines over SSH, from the couch. https://github.com/iwm911/Tether

@ButterJaffa Claude made a mod*