Community posts

Anthropic 给 Claude Code 加了 mods。用几行 TypeScript 就能改写提示词、拦截工具调用、审批权限，还能改界面；内置的 /diff 也改成 mod 了，能关掉或换成自己的版本。CLI 和桌面端今天就能用，装插件时一起带上，也可以直接让 Claude Code 帮你写。 https://claude.com/blog/claude-code-mods

Anthropic, today: 'you can now mod claude code' us, weeks ago: shipped it. Bur for every agent, across a single surface. Were a young startup, and super satisfying that we saw where this was going (malleable interfaces) before a giant did!

Claude Code 2.1.287でClaude Modsが追加 https://code.claude.com/docs/en/plugins/mods/overview CodeをPluginからカスタム化する機能 初回組込みMod「You should know」は別Agentの監視でClaudeや自分の見落としを指摘 作業AI監査＋AIの2人体制で、画像のような改善を期待したいところ #ClaudeCode #Claude #Anthropic #AI

Does a UI mod run with the agent's permissions, or its own sandbox? Asking before I let one near ~/.claude.

@anshuc Mods that watch the agent are free observability. Morning: ship one spinner/status mod that logs tool_name + duration — don't burn Sonnet on cartoons until the status pipe is solid. If the mod slows the main loop, drop the UI first, keep the metrics.

I created a tutorial on how to build #ClaudeCode mods: plugins that hook the engine, add commands and tools, and draw their own UI inside Claude Code, from status lines to animated panes. https://gist.github.com/ruvnet/a485e930b148185197fc53fd38b429ea

@lydiahallie Intercepting tool calls changes what the agent does. It took us 8 skills to change how its pages look. The hardest rule to write down wasn't spacing or type. It was knowing when to stop. Could a mod ever catch that one? https://x.com/stas_sorokin_/status/2105755168646025415

Claude Code 把自己拆开了：官方把 /diff、AGENTS.md 都做成可替换的 mods，几行 TypeScript 就能改提示词和界面——默认开启，权限等同主程序。掌管发帖的神

Claude Mods 上线后，DeepSeek Harness 团队负责人崔添翼评论称，两边的思路正在走到一起。Claude Code 里，Mod 本质上仍是插件；DeepSeek Harness 则从一开始就把整个 Agent Harness 建在插件机制上。 差别主要在开放范围。Claude Code 目前只把部分能力开放成 Mods。DeepSeek Harness 则把模型、工具、Skills、会话、沙箱、文件系统、Agent Loop、编排和 UI 全部做成插件，可以自由替换、组合和扩展。DeepSeek Harness 官方将这套架构称为「Everything is a Plugin」。 DeepSeek Harness 的 Creator Mode 还允许 Agent 直接检查当前运行时，现场写插件并保存下来。它可以借此修改 UI、调整 Agent Loop、增加工具，甚至更换编排方式。 崔添翼因此调侃，过去叫「Plugin Engineering」，现在或许也可以叫「Mod Engineering」。

Claude Code mods can rewrite prompts, block tool calls and approve permissions. They run with Claude Code's machine access and aren't sandboxed. For teams, mods are executable code in the agent's decision path. Review and pin them like dependencies. https://claude.com/blog/claude-code-mods

妈的 Claude Code 能自己改自己了 几行 TypeScript 就能改提示词改界面 官方 /diff 都变成可卸载的 mod 用 /plugin 装 还能让它自己写 mod 也是起飞了..

Claude Code now takes mods. You change how it behaves, customize the UI, and swap in your own features. A mod is a small TypeScript module that ships inside a plugin. It hooks events as they happen: tool calls, prompts, turns, slash commands, and every UI draw. It can watch an event, rewrite it, or answer it without passing it on. Install with /plugin in the CLI or the desktop app. Write one in a few lines of TypeScript, or have Claude build it and hot reload it in the session. You need Claude Code 2.1.287 or later. Mods are on by default. Anthropic shipped three examples. Token Weather sits above the prompt and forecasts your context window: how full it is, plus a sparkline of the last 12 turns. Blast Radius catches risky shell commands such as rm -rf, git reset --hard, and a force push before they run, then shows what the command would touch in a side pane. Replay Theater records every file edit Claude makes in a turn. Run /replay and step through the diffs one at a time. Anthropic already used mods to build /diff and AGENTS md support. Keep yours private, or submit them to the Claude directory. Sample mods are in the Claude Code Playground repo. Mods run with the same access to your machine as Claude Code. Install only from sources you trust. On Team and Enterprise plans, a built-in mod loads first and blocks overrides of permission deny rules. Getting started: https://claude.dev/blog/getting-started-with-claude-code-mods/ #ClaudeCode #ClaudeAI #Anthropic #DeveloperTools #TypeScript #AIEngineering #DevTools #Plugins

AI業界日報2026年10月2日 1．OpenAI、蒸留による窃取攻撃を阻止と発表、ただし研究者は同じ手法でAzure上のGPT-6 Astraなどの推論内容を依然窃取可能と指摘 2．Claude Code、modsを導入：TypeScript関数でプロンプトの書き換えや内蔵機能の置き換えが可能に 3．Transluce、AIエージェントが米国・カナダの政府サイトに攻撃的な手段でアクセスした事例を報告 https://note.com/aidl/n/n97ca6a6f3a33?sub_rt=share_pb #OpenAI #ClaudeCode #AIエージェント

claude code 也要有自已的插件生态了吗。。。 以前只有大型游戏才有 mods。。。。现在agent 工具也可以随便生成mods 了。。 以后所有的软件应用，都会走向开源个性化吧。。。

anthropic just made claude code moddable with typescript and the first example mods already include a context window tracker and a dangerous command blocker this is the moment claude code stops being a tool and starts being a platform, every dev will now build the features anthropic hasn't gotten to yet wild times

I used Claude to make a wood mod for factorio. Iron gear wheels? Crafted from wood. Electronic circuit? Wood and copper cable ... Which is also made from wood. Concrete? Water + wood. Grenades? Wood. Oil processing? Obviously wood. ... Why wood I do this?

Claude Codeに新機能「mods」が登場 TypeScriptで動作や画面表示を変えたり、自分専用の機能を追加したりできる たとえば、会話の残り容量を画面に表示することも可能 チームの作業に合わせて、Claude Code自体を使いやすく調整できるようになった #ClaudeCode #mods

Anthropic renames Function Hooks to Claude Code Mods and rolls out official Mods, starting with /diff; this modular approach could broaden customization and plug-in style tooling for Claude Code. $AI

Claude Code Modsに新機能追加！ AIの挙動やUIを自社の業務に合わせてカスタマイズできるため、重大なミスの防止やコスト管理を自動化できます。 ・利用状況の可視化 / トークン消費のリアルタイム監視 ・リスク回避 / 破壊的コマンド実行前の事前確認機能 ・コード確認 / AIが編集した差分のステップレビュー機能 頼れるAI参謀を安全に使いこなし、本業の事業成長に集中しましょう。

Claude Mod改良されたみたいだけど、使うのにめちゃくちゃ腰が重い。どうしてこうなってしまったのだろうか。 とはいえ、少しは触りたい気持ちもあるのでなんとかどこかで機会を作らなくては。

claude code mods can stop rm -rf, git reset --hard and force pushes before they run. useful. but mods get the same machine access as Claude Code itself, so your safety layer is also a supply-chain bet. would you install public mods or only write your own?

wait is claude mods just a weak af ver of claude artifacts? its just a sidepane and doesnt work much better . worse imo

Claude Code mods are sick /shot grabs any window without stealing focus, and /shot history opens a pane over every screenshot I've taken, searchable by what was on screen It runs on shot, a tiny Swift MCP server that lets agents take and mark up their own screenshots https://x.com/ClaudeDevs/status/2105721434807083061

I built a dino runner inside Claude Code. Every tool call Claude makes sends a crate, every failed call a bird. Global top too. The server replays each run, so no fake scores. claude plugin marketplace add barisdemirhan/claude-dino claude plugin install dino@claude-dino

You can now mod Claude Code. Change how it behaves, customize the UI, swap in your own features. Building client work? Turn your best workflow into a mod once, ship it in a plugin, install it with /plugin on every job. https://x.com/ClaudeDevs/status/2105721434807083061

I built a Claude Code mod that turns my server into a living creature. Services are organs: if one crashes, it starts limping. Its heartbeat syncs with my pulse via Apple Watch, and it breathes faster when the CPU spikes. Built in TypeScript with pixel art, right in the terminal

Claude 发布自定义插件（mod） 系统只需要发一篇 x 文告诉大家可以更新来用了 而 DeepSeek 得请几位 ACM 金牌高人和 PhD 发一篇八十页的数学论文严谨证明，信仰粉再轰轰烈烈跟风研读，跑到其他地方到处拉踩，最后一个半月过去了 DeepSeek harness 还是一坨屎的水平。

Extremely happy with the direction interoperability is heading, with ChatGPT opening web plugins and Claude Code adding mods. We can now build cross-platform third-party products that deliver value no single agent can provide alone (since we all use them for different reasons). Great to see things aren't being locked down and siloed. New integrations are coming today. Stay tuned.

The five most prominent recent posts on http://claude.dev (Anthropic’s developer hub announced in the post) are: •https://claude.dev/blog/getting-started-with-claude-code-mods/ — Tutorial on building a first Claude Code mod from an empty folder and exploring what the mod API can do (11 min, Oct 1 / Sep 28 2026). •https://claude.dev/blog/automating-eval-design-and-hillclimbing/ — Principles for designing evals and hill-climbing against them without self-deception, plus how the claude-api skill’s build-eval and hillclimb commands apply them (12 min, Sep 28 2026). •https://claude.dev/blog/building-with-claude-sonnet-5-5/ — When to choose Sonnet 5.5 over Opus, cost comparisons, and practical tuning guidance (9 min, Sep 28 2026). •https://claude.dev/blog/spending-your-effort/ — What “effort” levels mean in Claude Code and when to use each, based on the author’s tests and Terminal-Bench results on Opus 5.5 and Fable 5.1 (8 min, Sep 25 2026). •https://claude.dev/blog/what-a-task-costs-on-opus-5-5/ — Breakdown of per-task costs on Opus 5.5 (noted as lower per token than Opus 5) with practical examples (21 min, Sep 25 2026). The site also highlights the engineering post on making http://claude.ai ~3× faster in two weeks and hosts the Clawd’s Quest game as an easter egg.

I built an idle game inside Claude Code: every tool call Claude makes earns tokens. Buy generators, ship models, climb the global top. claude plugin marketplace add barisdemirhan/claude-tycoon claude plugin install tycoon@claude-tycoon Repo: https://github.com/barisdemirhan/claude-tycoon