Community posts

Claude Codeに、Chromeの拡張機能みたいなものが来た もともとあったHooksは、外側でコマンドを走らせる仕組み ModsはClaude Codeの中に住んで、動きも画面も変えられる 入れると、たとえばこんなことができる： ・rm -rf や git reset --hard の前に、消える範囲を出して一度止めさせる ・AIが一度に覚えられる容量を、どれくらい使ったかを入力欄の上に表示する ・さっきAIが直した差分を、あとから /replay で見返す 作り方は簡単で、Claude Codeに「危険なコマンドの前に確認させて」と日本語で頼むだけ 他人が公開しているModを入れるときは、この3つ： /plugin marketplace add 作者/リポ /plugin install 名前@マーケット /reload-plugins 注意するべきは、ModはあなたのPCの中身を普通に触る 知らない人のものは、入れる前に claude plugin validate ./そのフォルダ を通すか、中身をClaudeに読ませたほうがいい 前提は claude --version が 2.1.287 以上

Claude Code CLI + Plugin Mods + Highlight JS =

Claude Code Mods are JavaScript or TypeScript hooks for prompts, tool calls, and UI rendering. They can add panels, intercept calls, or replace built-ins like /diff. I'd try one for repo-specific behavior. What would you hook first?

こんなClaude Modsもできたぞ！ ↓claude codeの画面でこれでインストールできます /plugin marketplace add jnk0vc/quota-reactor /plugin install quota-reactor@quota-reactor #Claude

Claude AI destekli, farklı oyunları birleştiren hızlı modların sosyal medyada çoğalması; emek, izin, kalite ve hak sahipliği tartışmasını büyüttü. Bazı deneyimli mod geliştiricileri bu duruma tepkili. #Modding #YapayZeka #AI #GameDevelopment #OyunGeliştirme

A CLAUDE SKILL THAT MAKES YOUR AI LAZIER JUST PASSED 155,000 STARS ON GITHUB. it makes Claude write 73% less code, 46% cheaper and 51% faster. bookmark this before your next Claude Code session. it's called ponytail, and it makes Claude think like the laziest senior dev in the room. before writing a single line, Claude has to climb a ladder: does this need to exist? is it already in the codebase? does the standard library do it? is there a native feature for it? is the dependency already installed? can it be one line? only then does it write the minimum that works. the author ran it on 12 real feature tasks in Claude Code: 54% less code, 22% fewer tokens, 20% cheaper, 27% faster and it didn't drop a single safety check. the funniest part is the control group. a "caveman" prompt that only tells Claude to talk less made it 3% MORE expensive. talking less doesn't save money. writing less does. install takes 2 commands: /plugin marketplace add DietrichGebert/ponytail /plugin install ponytail@ponytail

Public service announcement: Claude-mods is absolutely cracked 😂 More to follow https://code.claude.com/docs/en/plugins/mods/overview

ok so this exploded 😭 so i shipped context-bar 📊 see where your tokens go. https://github.com/hamzafer/claude-code-mods

マイクラのプラグイン開発用にサーバー用のMCPプラグインとクライアント用のMCP MODを作ったら、claude codeがプラグインの動作確認やUIのデザイン検証まで全自動でやってくれるようになった 最終的には自分でも確認は必要ですが、だいぶ楽に #マイクラ

added some claude mods to my dev workflow git-gate: Shows an Allow or Deny dialog before git push, branch changes, resets, rebases and similar commands. session-pane: /session pane with session age, context fill, 5-hour and 7-day limits with reset times, and cost. secret-redactor: Replaces API keys, emails and public IPs with placeholders before I read them and restores them when a tool runs receipt: After each turn, a line above the prompt showing files changed, commands run, failures, plus /receipt for the full breakdown. agentpane: Side pane of live subagents with their tool calls, tokens and a Stop button.

Pacing your Claude usage when you're shipping fast. What actually stretches a plan: 1) Fresh session per task instead of one endless thread 2) A token tracker so you see which jobs burn the week 3) Big-context runs saved for judgment calls, not boilerplate Claude Code mods help. Habits do most of the work. Bookmark it for the week you hit the wall mid-launch.

Claude CodeのModsでAgentたちのコミュニケーションの場を準備中（開発途中）。絵文字でちゃんとリアクションつけてくれるのが、なんかおもろい(´･_･`)

AI 코딩 도구 Claude Code에 모드(mods)라는 기능이 생겼어 Claude가 파일을 통째로 지우려는 순간 내 코드로 먼저 막을 수 있어 Claude한테 하지 말라고 적어 둔 걸 코드로 막아 볼 생각이야 #AI #바이브코딩

Samyak Jain shipped cache-warmer today, a small open-source Claude Code mod that pings your idle sessions every few minutes so coming back from a break doesn't force a full re-read of the conversation at full price. He's claiming sixteen times cheaper on the token bill for sessions that would otherwise go cold, same usage limit, same plan. The number that makes sixteen times plausible is Anthropic's cache window: 5 minutes. A cached prefix write costs 1.25x normal input price, and reads off that cache cost 90 percent less than fresh tokens, but only while the window is open, and every read resets the clock. Step away from Claude Code for ten or fifteen minutes between prompts and you likely get one read before the cache expires and the whole conversation gets billed fresh again. That's precisely the window cache-warmer's pings are built to keep open. Every major lab treats this differently, which says the industry hasn't settled on an answer. OpenAI keeps cached prefixes alive for 24 hours by default with no write surcharge on current models. Gemini's implicit caching is automatic and free to write, and its explicit option charges hourly storage instead of a write premium. Anthropic is the only one of the three charging extra to write a cache that goes cold in single digit minutes unless you pay double for the one-hour version. None of this is unique to coding agents. It's what happens any time a cost-saving default gets set somewhere upstream and the people relying on it only find out when the bill moves. A free mod that just pings a server every few minutes to keep a clock from expiring shouldn't be the state of the art for managing that, but for now, it is. https://x.com/Samyak0606/status/2106987472668545400

Claude Code Mods 플러그인 사용법, 켜기 전에 권한부터 확인하세요 (출처 : MAPS .. | https://blog.naver.com/mindmethod7/224432014442 블로그) https://naver.me/Gnt4UZOz

Tested 6 ways to route Claude Code work across models on a real repo. What paid off: Sonnet for the whole session. Same code checks and tests as Opus, about 40% of the cost. What didn't: switching models mid-session (+39%) and Haiku subagents for chores (+31%). The cache rebuild costs more than the cheap tokens save. I'm turning pick, my Claude Code mod, into exactly this: one model suggestion before your first prompt, then it stays out of the way.

Claude Code just let plugins change the product. Not sit next to it. v2.1.287, 1 Oct. The changelog line is one sentence: "Added Claude Mods: plugins may now modify deeper behavior." First built-in mod is called You should know. A side agent that flags what you or Claude missed. /plugin enable cc-plugin-you-should-know@builtin That switch is for first-party sessions with telemetry on. v2.1.288 followed about a day later (early 3 Oct GST) and kept building on mods. The moat is not the IDE. It is who gets to rewire the agent.

Claude Code mods from @AnthropicAI let you enforce a rule instead of asking for it: a TypeScript hook that rejects git push to main beats any line in CLAUDE.md. Flip side: a mod runs with your shell permissions, so review third party ones like an npm dependency.

Claude Code just closed a Bash deny and ask hole under sandbox auto-allow. A command wrapped in an environment variable prefix like TZ="$HOME" rm -rf build could skip the deny or ask rule. A bare assignment sitting in front of the command could skip it too. Both paths are caught now. Mods can also spawn teammates through agent.spawn, keep one agent id across plugin hook events, and see idle and waiting states in $.agent.list(). Claude Code is Anthropic's coding agent. The deny and ask rules hold when the sandbox would have waved the command through, and a mod can spin up a teammate without inventing its own spawn path. https://github.com/anthropics/claude-code/releases/tag/v2.1.289

Every Claude Code subagent inherits the main session's model unless Claude says otherwise. I watched it spin up an Opus subagent to count Swift files. So I built a mod that asks Jev which model each call actually needs. Write-up with the numbers: https://benenewton.com/blog/jev-agent-routing-claude-code?utm_source=twitter&utm_medium=social&utm_campaign=every-claude-code-subagent-inherits-the-main-session-s-model-1

My Claude Code mod lets Jev choose the agent and model for every call. It answers in about 0.3 seconds, before each subagent starts. On lookups it cut the cost 51%, and every fact in the answers still checked out. What I built, and the three rounds it took to get the rules right:

Claude Code's new mods come with a safety net, but only if your company manages your machine. Solo devs get the power without the net. Mods (shipped Oct 1) are small TypeScript functions inside plugins. They can rewrite your prompt, block tool calls, and answer permission prompts. Anthropic says they aren't sandboxed. The guardrail is a built-in mod, sec-default, that loads first and stops user mods from overriding your deny rules. Anthropic lists it for Team, Enterprise and managed machines. Anthropic's list doesn't mention personal setups. Even that path needed patching: two days later, 2.1.289 fixed a deny rule not holding over a mod's approval, and a plugin rewriting an org MCP server's sign-in tool descriptions. Treat every mod like a dependency with shell on your laptop. Read it, pin it, watch anything that hooks permission events. Installing other people's mods yet, or only your own?

claude code のmodsで何作ったらいいか分からない。という人は、 「自分のclaudecodeのセッションを分析して、modsで どういうものが作ったら良さそうか、効率化できそうかを教えて」という指示を出すんや

I made a claude mod that shows a summary of what the agent is working on. It's easy to get lost with several running.

Most AI apps do what they're built to do. This one lets you change what it's built to do. Why DeepSeek Harness stands out: ✔ Open source, so anyone can see how it works ✔ Plugin manager shows what each add-on does and where it came from ✔ Creator mode builds the extras that are specific to you ✔ Automation tasks add scheduled runs, run logs, and progress previews ✔ Oct 3 alpha even tests a compatibility layer for Claude Code Mods → Not full compatibility yet, but it shows how fast the project is moving The real edge isn't one feature. It's being able to shape your own workspace in plain English. 💡 Save this video, you'll see why extensible AI workspaces are worth learning now. Want the SOP? DM me.

Claude Code Mods: A Salesforce Developer's Guide to the Moddable Agent https://www.worldprogramming.org/posts/claude-code-mods-a-salesforce-developers-guide-to-the-moddable-agent-tofm9f

DeepSeek just made Claude Code mods portable Claude Code extensions can now start working inside DeepSeek Harness too. AI coding ecosystems are starting to merge.

Claude Mods、地味だけどlgtmしないとコミット不可にするやつ作ってる。あとこれとは別にiPhoneでも差分確認しやすいビューア作ってだいぶ楽になった

Recall logs what Claude does after each tool call. Today that's a bash script: about 290 ms and around 10 programs started, on every tool call. As a Claude Code mod it runs inside Claude Code. A queue and a timer. Zero programs started. Still a prototype, but this is the kind of win I like.

While Claude Code has mods, Codex has the "dancing ToC", did you know?