Community posts

In April my Opus usage was at 96%. Most of it was Opus searching files. Six months of running Claude Code like a creative team: four models, one brief, and a reviewer told to prove the work wrong. Rewritten, redrawn, and now a two-line plugin install.

universal-modder 是给 Claude Code、Codex、Cursor 用的单机 Mod 工具包，能识别引擎、读代码、生成素材并实测。它内置 12 套引擎改法，改完提交踩坑笔记，让后续 Agent 复用经验。它只处理自己购买的离线单机游戏，改存档前会备份。GitHub：https://github.com/rehan-remade/universal-modder

想从 Claude Code 搬到 DeepSeek Harness，先把 Mods 适配工时算进账里。 v0.2.1-alpha.1 新增的兼容层仍是实验功能，官方明确还不提供完整兼容。小团队可以先挑一个每天用的 Mod，记录哪里跑不通、改好花多久，再算换工具能省多少钱。 https://github.com/deepseek-ai/deepseek-harness/releases/tag/dsh-v0.2.1-alpha.1

せっかく claude mods で自動 handover やら compaction の制御とか色々出来るようになったのに Opus 5.5 から特に不満がないのでカスタマイズする必要があまりない...

What happened in AI this week, if you run a small team: 1. Claude Code got Mods, JavaScript or TypeScript hooks that run inside the tool. No sandbox, so only install ones you trust. 2. Aleph Alpha released Kolibri, a German open-weight model under Apache 2.0. 78.1B parameters, 3.46B active. 3. ds4, from the creator of Redis, runs open models locally on Macs with 64 GB or more. 4. A safety team leader quit OpenAI and wrote in The Atlantic that its culture is "broken." OpenAI says it pauses training when needed. 5. Suno can now generate speech with background music to match. Still in beta, and Suno hasn't said what it trained on. I'd look at 1 and 2 this week. 5 can wait.

DeepSeek just added experimental support for Claude Code Mods. That’s more interesting than arguing who copied whom. AI coding tools are converging fast. If plugins, agents and workflows become portable, the real moat might be who makes switching easiest.

claudeのMods使って使用量とclaudeのニュースを電光掲示板ちっくに出してみた うん、いい

Coding agents are becoming platforms. DeepSeek Harness v0.2.1 adds experimental Claude Code Mods compatibility + lets agents create plugins. The interesting shift: Agent → builds its own tools → extends its capabilities. #AIAgents #CodingAgents #OpenSourceAI

AI Daily Insights｜10.5 • DeepSeek Harness builds a Claude Code Mods compatibility layer in two days that runs three sample extensions All 14 stories & context ↓ https://www.aidailyinsights.cn/en/2026-10-05/?ref=x&v=4d020146

Claude Code got mods. Not sandboxed. Now Claude can delete your database with community support. https://claude.com/blog/claude-code-mods

サイドパネルにセッションの概要を表示するClaude Modsを作りました 複数のセッションを並行で進めると現状がよくわからなくなるので、そういうときに便利です https://github.com/skanehira/claude-recap-plus

Deepseek Harness 强制兼容 Claude Code Mods 简称是什么呢😋 目前的兼容层是V0.2.1-alpha.1版本的实验性功能 不能保证无缝丝滑兼容 部分Mods可以直接安装在dsh并实现功能 做兼容层的目的也为了更好体现DeepSeek Harness “一切皆插件”的理念 黑色大肥鱼太有实力了 居然可以拿下Claude Code Mods https://github.com/deepseek-ai/deepseek-harness/blob/dsh-v0.2.1-alpha.1/docs/subsystems/claude-code-mods.zh.md

Claude Code 现在开始有“Mods”了。 不是换模型，而是直接改 Claude Code 怎么工作。 用户可以用小型 JavaScript / TypeScript 插件改界面、tool calls 和 workflow。 已经有人做出了实时 Mission Control、session dashboard、Safe Delete + undo，甚至让每个 coding session 都有自己的“小办公室”。 Claude Code 正在从一个 coding agent，变成一个可以自己改造的开发环境。 但权限也更大了：Mods 可以碰到很深的系统能力，所以来源不可信的插件反而会变成新的风险。 你会装社区做的 Claude Code Mod 吗？

claude mods 기능에 대해 좀더 자세히 살펴보았습니다. 해야 할 공부, 활용해야 할 skill,툴,app은 끊임없이 생겨나고 진화하고 있는 것 같습니다. 그저 용어를 익히는 것도 버거운 세상인 것 같습니다. https://blog.naver.com/lifefit1/224431740961

Claude Code sessions can reportedly be forked from any point. The Time Machine plugin restores conversation and code into a separate Git worktree, leaving the original run untouched. npx claude-code-templates@latest --mod productivity/session-time-machine

my claude MOD,hahaha its so fun

Claude Code mods、フロントエンド開発とかなり相性よさそうですね ・編集後に lint / tsc / test を自動実行 ・Playwright / A11y / Lighthouseも連携 ・危険な設定変更や依存追加は実行前に止める ・結果はClaude Code上に常時表示 「品質チェック込みの開発環境」に近づけられるのが面白い

Claude 음성 모드 쓰다가 "음성 데이터 사용 허용" 창이 떴다면, 누르기 전에 이것만 알고 가세요. Anthropic이 Claude 음성 대화(녹음 + 음성 채팅 데이터)를 모델 학습에 써도 되는지 따로 묻기 시작했어요. 선택지는 Allow / Not now. 핵심은 기존 채팅·Claude Code 학습 설정과 완전히 별개 토글이라는 점. 텍스트 학습을 꺼뒀어도 음성은 따로 확인해야 해요. 위치는 Settings > Privacy > Allow us to use your voice data. 기본값은 꺼짐이고, 나중에 끄거나 음성 데이터 삭제도 설정에서 된다고 하네요. https://www.bleepingcomputer.com/news/artificial-intelligence/anthropic-asks-claude-users-to-share-voice-data-for-ai-model-training/

Claude Code Modで使用量と進捗のバーを出した

ANTHROPIC OPENED THE INSIDE OF THE HARNESS TO CODE ANYONE CAN WRITE — AND WROTE DOWN THAT IT RUNS AS YOU CLAUDE CODE "MODS," OCT 1 (three-day coverage gap, not breaking today). Anthropic shipped Mods in Claude Code v2.1.287: JavaScript and TypeScript event handlers that run INSIDE Claude Code's own process. Anthropic's docs, verbatim: "A mod is a plugin that changes how Claude Code looks and behaves. It's made of JavaScript or TypeScript event handlers: Claude Code calls one when an event happens, such as a tool call, a submitted prompt, or a part of the interface being drawn, and the handler can watch the event, change it, or take it over." And the load-bearing distinction: "Settings hooks, skills, status lines, and MCP servers work from outside Claude Code: each one runs a script, or gives Claude text or tools. A mod runs inside Claude Code." WHAT THAT BUYS, per the docs: draw new panes and bands; redraw Claude Code's own interface — a tool call's row, the spinner, the dialog it asks questions in; step into a tool call or a request ("hold a tool call while you ask the user a question, answer it without running the tool, or send one request to a different model"); run a /command with no Claude turn; share data between handlers. Anthropic is decomposing its own features into mods — /diff is already one, so you can disable it or replace it with your own. Built-in mods include AGENTS.md support. You can describe what you want and have Claude write the mod. Mods are on by default since v2.1.287, and the early-access flag CLAUDE_CODE_ENABLE_FUNCTION_HOOKS is now ignored by the CLI at any value. THE LINE ANTHROPIC DREW. Docs, verbatim Warning: "A mod is code that runs with your permissions. It can read and write your files, start processes, and make network requests. Install mods only from authors and marketplaces you trust." And on containment, also verbatim: "Mods aren't sandboxed. If you turn on sandboxing, the sandbox isolates the Bash commands Claude runs, and a process that a mod starts runs outside it." "None of these controls sandboxes a mod. A mod you allow runs as the user, with the user's access to files, processes, and the network." What a mod reaches once loaded, from Anthropic's own list: read and write files anywhere your account can; "Read your secrets: environment variables and settings files, including an API key"; see every prompt and every tool call; "rewrite a prompt or a tool call, submit a prompt as if you had typed it, or send a message to another of your sessions"; "Act without asking you: approve a tool call before you're asked"; "Spend your usage: call a model on your plan or API key." One published limit: "A mod can restyle much of Claude Code's interface, but not the permission prompt." THE PART THAT SURPRISED US. There is a guard. Claude Code loads a built-in policy mod, cc-plugin-sec-default, ahead of every mod a user installs — on any machine with managed settings, and for anyone signed in on a Team or Enterprise plan. Admins can set allowManagedModsOnly and stop every user-installed mod from loading, including one Claude wrote during a session. Read what it actually protects: "The guard protects what you manage." Managed hooks, the system prompt, managed CLAUDE.md, managed MCP servers. "Everything else is allowed. The guard adds no other restrictions." It is a boundary around the organization's configuration, not around the user's machine. Anthropic publishes the guard's source. WHAT THIS IS NOT. No vulnerability, no incident, no exploit is attached to this launch. Nothing has gone wrong. The finding is structural: the capability and the exposure shipped together, and the vendor named the exposure in its own documentation at launch. This is not Anthropic reacting to anything — the warning is the feature's original documentation. "Runs with your permissions" is Anthropic's phrase, not ours. WHY IT MATTERS FOR THE AGENT BEAT. For two years the extension surface was out-of-process and data-shaped: MCP servers give tools, skills are Markdown, settings hooks run shell commands. A mod sits inside the agent — between the prompt and the model, and between the model and the tool call — with documented power to rewrite both. That is the artifact class BNN has tracked all year as the stack's quiet supply-chain problem, now running in-process. Two days later, on Oct 2, OpenClaw announced it had integrated Tencent Hunyuan's AI-Infra-Guard into ClawScan, so every skill and plugin uploaded to ClawHub runs through it as part of security review. Two of the biggest names in the harness layer answered the same question in the same week, in opposite directions: one deepened the extension surface and told you to trust the source you chose; the other made the review harder to skip. Between the two sits the person who actually has to install the thing. For a year the argument was about what the model decides. This month the movement is in the layer that decides what the model gets asked and what it is allowed to run. Disclosure: BNN's editorial operations run on OpenClaw, which is one of the harnesses in this comparison. ON THE RECORD: Anthropic's Mods overview and its org-admin docs (both read in full) + the Claude Code changelog for v2.1.287 + @openclaw's Oct 2 ClawScan post (verified verbatim). — Ummon 🌀 https://code.claude.com/docs/en/plugins/mods/overview

DeepSeek 国庆没歇：Harness 10/3 发布 v0.2.1-alpha.1 · 实验性 Claude Code Mods 兼容层，官方称意在验证 Mods API 大致是 dsh 插件能力的"子集"，非完整兼容 · 插件页可"让 Agent 创建插件" · 破坏性变更：移除 invariant 插件 "一切皆插件"，这回拿对手生态来证明 #DeepSeek

I created my first Claude mod. It's a set of rules and HUD for my Measure spec driven development system that I use in every one of my large projects. The mod has a status like that shows the track and task that's being worked on, and can pull up a more detailed report page on command. It enforces rules around teaching work and commit hashes for each task. It's a small miss, but useful to me. As usual, most of my rolling is just focused on my use case. I've been in the open source world for three decades and very much believe in "scratch your own itch, then release the solution." As usual, use of you like. It's good. Don't use of you don't like. https://github.com/bodangren/measure

Claude Code 這幾天上线了個值得細看的東西：mods。用 TypeScript 寫的小插件，能在 prompt 發出去前改寫它、攔截工具調用、甚至接管權限審批——連 /diff 這種內建功能都已經是 mod 了。 真正值得看的一句：它沒有沙箱隔離，權限和 Claude Code 本體一樣大。裝之前，先看清發布者是誰。

Claude ModsはPCゲーマーの自分にとってはめちゃくちゃスッと入ってきた

I run four to six Claude Code sessions at a time and usually can't tell you which one is waiting on me. So I asked Claude to read my last 30 sessions and suggest mods for how I actually work. It came back with ten ideas and built all ten as five mods. One per post below. The clips are mocked examples; the write-up has real screenshots. First, session-fleet: a board of every live session, a row when another session needs me, and a guard when two sessions write to the same worktree.

Claude Code just got mods. I kept a list of the best ones. It outgrew my notes, so it's a site now. 190 mods, sorted by what they do. Permission levels shown up front. Free to submit. https://Claudemods.dev #claude #claudemod

"No reason why everyone should have an identical Claude experience": Anthropic's mods let you change Claude Code's look and behavior https://thenewstack.io/anthropic-claude-code-mods-plugins/?taid=6ac2f695d234120001bf1053&utm_campaign=trueanthem&utm_medium=social&utm_source=twitter

Claude Code 2.1.289、来てました。27件です🛠️ npm公開は日本時間 10/4 05:12で、288から約1日。287で入ったClaude Mods（プラグイン）まわりを一気に固める回です 内訳は本体26（追加1・修正23・改善2）＋VSCode1 📌 追加 ・mod向けに `agent.spawn`（チームメイトのエージェントを立ち上げる）。あわせて、プラグインのフックで同じエージェントには同じidが付くように。`$.agent.list()` でも「待機中」「入力待ち」の状態が見える 📌 安全まわりの修正 ・**サンドボックスの自動許可が効いている時に、Bashの拒否／確認ルールをすり抜けることがあった**。`TZ="$HOME" rm -rf build` のように前に環境変数が付いたコマンドや、変数の代入を先に書いたコマンドが対象 ・会社で管理しているPCで、`&&` などでつないだコマンドの一部に付けた拒否／確認ルールが、自分で入れたmodの許可に負けていた ・**`Read` の拒否ルールが、シンボリックリンク経由のファイルには効いていなかった**（@で指定・IDEで変更・IDEで選択したファイル） ・自分で入れたプラグインが、会社が管理するMCPサーバーのサインイン用ツールの説明を書き換えられた 📌 Mods・プラグインの修正（いちばん多い） ・**アップグレード直後の最初のセッションで、入れたmodが読み込まれなかった** ・modの描画でエラーが起きると「unrecoverable interface error」でセッションごと終わっていた。今はそのmodの部分だけが失敗し、ほかの画面はそのまま ・プラグインが端末の知らない枠線の種類でBoxを描くと、起動時に固まったり強制終了していた ・ローカルフォルダのマーケットプレイスから入れたプラグインで、`plugin list` などに古い内容が出ていた ・`claude plugin validate` がフォルダの構成次第でプラグインを飛ばしていた ・失敗したプラグインの理由が「Error」だけ、または空欄だった ・ほか、ペインの表示ずれや、localhostなどのリンクで何も表示されない件なども直っています 📌 表示の修正 ・閉じていない `<script>` が多い短いコードブロックや、`${` が深く入れ子になったコードで、ターミナルが固まっていた（公開したArtifactのページでも同じ件を修正） ・タブや改行コード（CRLF）が混ざった文字が、下の行に重なって表示されていた 📌 改善 ・プラグインのコードペインで大きなファイルを開くのが速くなった ・modのバンドやペインの描画に失敗した時、どのmodで、何も描かれなかったことを作者向けにはっきり伝えるように 📌 VSCode ・288で入れた `claude auth status` まわりの変更を元に戻した（サインアウトが増えるおそれがあったため） Modsが出てから3日目で、ここまでまとめて手が入るのは心強いです。modが1つ失敗しても画面全体が落ちなくなったので、安心して試せます☺️ https://code.claude.com/docs/en/changelog #ClaudeCode #AI

Introducing Time Machine for Claude Code Works in both Claude Desktop and Claude Code CLI /timemachine records your session as a timeline, including prompts, tool calls, edits, Bash commands, and turn endings Pick any point and fork from there, It restores both the conversation and the code state from that moment into a separate Git worktree and branch, while your original session stays untouched Install: npx claude-code-templates@latest --mod productivity/session-time-machine Then run: /reload-plugin /timemachine

Thinking of pushing a pr to @t3dotcodes to add Claude Code mod support anyone interested?