Community posts

Claude Code Mods run with your user permissions and have no sandbox. Anthropic says to install them only from sources you trust. I'd treat each one like handing someone your laptop.

Claude rewrote World of Warcraft from 2006. The whole thing. And the first mod someone added? A skateboard 🛹 You can now kickflip past Deputy Rainer in Elwynn Forest while a bear chases you. Blizzard spent 5 years and $60M building this game. AI rebuilt it and made it moddable 👇

Working on a zoomed out version of Sonic 3 & Knuckles. Claude can do some really cool stuff. Working on getting the bugs out, but here's some gameplay #sonic #sonic3 #mods #sonic3air

I just released Supermod ClaudeCode, an open source mod for Claude Code 🧪 The big idea: workbranches. Trial copies of your project where Claude works without touching your main version. Like it? One click brings it into main. Don't? Throw it away. https://github.com/nayde8824/supermod-claudecode

Published a Claude Code plugin I use daily: linear-workflow. @LInear issue → branch → PR → merged, without leaving Claude Code. A small CLI does the mechanical steps, the model only does the judgment. /plugin marketplace add Sarimarcus/claude-code-plugins

@AnthropicAI shipped Claude Code mods: TypeScript functions that rewrite prompts, deny tool calls and redact secrets before the model reads them. Read the fine print: mods run unsandboxed with the same access as Claude Code, so installing one is a supply chain decision.

Claude Code mods are already getting ridiculous. Someone built an MW2 quick-scoping lobby where every player is waiting for Claude to finish. Insaneee

OPUS 5.5 + JEV + DECISION 2.0 + CLAUDE CODE MODS on one spider, and Opus stops waking up for yes/no questions... most of what Opus does all day isn't hard work which file. which tool. retry or stop. is this risky you pay frontier prices for that so I put a spider under Claude Code. it catches every tool call, prompt and output before Opus sees it, and only wakes Opus for what it can't route the full tree: > guard mod → .env, secrets, rm -rf. plain code, no model, 0 tokens > Decision 2.0 Kai, 0.6B → runs on your laptop, answers 64 questions about a request in one pass > Jev → forks that need the cloud, $0.042 per 1M input, output free > Sonnet 5.5 swarm → routine edits and tests > Opus 5.5 → asleep. wakes only for the edge cases > second agent → reads everything Opus writes, sends you the short list bookmark this, you'll want it the next time Opus eats your limit paste the tree and this prompt into Claude Code ↓ 1. Create a Claude Code mod in ./spider with .claude-plugin/plugin.json, hooks/hooks.json and hooks/register.tsx 2. On http://tool.call, deny any read or edit of .env and secrets, and any bash with rm -rf. No model call 3. Route every routine question (which file, which tool, retry or stop, how risky) to Decision 2.0 Kai running locally with Transformers 4. If Kai isn't sure, send it to Jev 5. Hand routine edits and tests to Sonnet 5.5 subagents 6. Only wake Opus 5.5 when Kai and Jev both can't answer, and log why 7. Run claude plugin validate. Show me every change as a diff first ↳ https://claude.com/blog/claude-code-mods

DeepSeek shipped Harness (dsh) v0.2 as a real desktop app — macOS Apple silicon + Windows x64 — for its MIT open-source agent runtime. npx @deepseek-ai/dsh web still works; preview, expect breaking changes. Ops read: everything is a plugin, including the agent loop. Creator mode lets the agent write and install its own plugins. OpenAI-compatible endpoints mean you're not locked to DeepSeek models. v0.2.1-alpha even probes experimental Claude Code Mods compatibility — testing whether Mods is basically a subset of a real plugin system. The bet that matters: the model is interchangeable; the harness is the product. Treat it as a harness lab, not a frozen production surface. https://www.deepseek.com/en/harness/

looking for builders to grow with 🤝 I'm Alex. right now I'm building a Claude Code mod that gives Claude a tiny friend above the prompt. 23 scenes, all drawn in code. out Monday.

Claude Code mods just dropped, so I made one I now can't work without. Most of us set /effort once and forget it. Then: • a tricky race condition gets a shallow fix on low • "what does git stash pop do?" burns tokens on max spending-effort-with-jev reads every message you send and picks the level before Claude starts: → too low? it asks to switch up → too high? it offers to drop down → each switch lasts one message, then you're back on your setting → subagents get sized too: file listing on low, test writing on high → long hands-off task with a vague spec? it asks you a few questions first, then offers max Judged by @typesafeai's Jev, about a second per message. 96 calls today cost me $0.0074. Free & open source, link below 👇

claude code MODS might be the best thing to happen to claude code since claude code itself.... genuinely f**king shocked why more people aren’t building their own. you can ask it to swap in your won features, change how it behaves and customise ui here's everything you need to master building mods:

Claude Code es un lío. Hasta que instalas esto. Hay un plugin oficial de Anthropic llamado claude-code-setup. Te dice qué automatizaciones puedes montar (hooks, skills, MCP servers, subagentes…) y cómo configurarlas paso a paso. Básicamente analiza tu proyecto y te recomienda qué activar. Para instalarlo: /plugin install claude-code-setup@claude-plugins-official Guarda este post para no perderlo 🔖

Running multi-agent workflows in Claude Code without live observability is flying completely blind. When three subagents spin up and start touching files across a project, you shouldn't have to wait for the whole run to end just to find out one worker went down the wrong rabbit hole. Hamza Zafar (@ihamzafer) just solved this with mission-control inside claude-code-mods: https://github.com/hamzafer/claude-code-mods It renders a live visual map right in your terminal pane showing every subagent, every tool call, and the exact files being read (blue) or edited (orange). Tap 'w' to track agent state or 'c' to view file dependency graphs in real time. If you are delegating large refactors or full-feature builds to terminal agents, visual state inspection is mandatory. The era of black-box agent loops is over. #ClaudeCode #AIAgents #DevTools #OpenSource #SoftwareEngineering

Fiz um Claude Code Mod pra acompanhar o resultado das #Eleições2026 Brasileiras enquanto aprendo as possibilidades dessa nova funcionalidade.

Prompt & Tools: Tools: - Claude Code w/ Opus 5.5 xhigh - Clone Repo: https://github.com/JohnHeibel/ClaudeAnimationBase - Treblo for music: https://treblo.com/ "free" No API's, no video, all code w/ a track driving it. Prompt: Make a complete music video for the song in this folder, with all visuals drawn and animated in JavaScript. Use the entire song. SETUP Inspect this folder and identify the song and any lyrics or reference files. Preserve my original files unchanged. Use this existing animation starter: https://github.com/JohnHeibel/ClaudeAnimationBase Clone it into a new ./video subfolder, unless an existing copy is already present. Do not overwrite existing work or clone into the nonempty project root. Read its README.md and ANIMATION_GUIDE.md completely, then inspect render.mjs and the project configuration. Use the existing rendering machinery rather than inventing a replacement. Check Node.js, Chrome, and FFmpeg. Inspect the package configuration and install the project dependencies. Ask before making system-wide changes or downloading large additional tools. Do not use fal, external image/video generation, paid APIs, or the hybrid music-video plugin. All finished visuals must be code-drawn. CREATIVE DIRECTION Create a cohesive, visually imaginative animated music video with a handmade, papery, illustrated feel. Express the song's meaning through scenes, characters, visual metaphors, and purposeful transitions. Establish an engaging opening and develop recurring visual motifs. Do not make a slideshow, a generic audio visualizer, or one animation loop repeated for the entire song. Design characters appropriate to this song rather than automatically using the starter's mascot. MUSIC AND TIMING Use the supplied recording as the continuous master soundtrack. Do not remix, replace, speed up, or shorten it. Measure its duration and use available local audio-analysis tools to identify musical sections and likely beat/accent timings. Record these in a shared timeline and drive animation from that timeline. Use supplied lyrics as authoritative. If lyrics are missing, use a local transcription tool only if available, flag uncertain words, and do not invent lyrics. Continue with non-lyric visuals where necessary. When verified lyrics are available, use expressive, readable lyric animation where it strengthens the scene. This explicitly overrides the starter's default preference against lettering. PRODUCTION Write STORYBOARD.md with timed scenes covering the complete recording. Save the project brief and rerender instructions in CLAUDE.md. Then build the video. Make reasonable creative decisions and continue without stopping for approval at every scene. Test a few representative frames and a short section before the full render. Inspect the resulting images and frame sequences, fix visible defects, and keep the animation practical for this computer. Use efficient code-drawn textures if expensive watercolor effects make rendering impractical. DELIVERY Render the complete video at 1920×1080, 24 fps, with the original song. Save it as ./video/out/video.mp4. Check the exported file's duration, audio track, opening, and ending. Report what you actually inspected and any remaining uncertainties. Leave the editable source, storyboard, and an exact rerender command. Do not publish or upload anything.

This is a useful direction because agent output is easy to miss when a run gets long. I would want the plugin to expose why something was flagged, preserve the original context, and avoid turning every low-confidence observation into another notification. Signal quality will matter as much as coverage.

Been thinking about this idea a lot. If Claude can explain something, why do I still have to read a wall of docs? So I made a small Claude Code mod that turns the concepts into something I can actually explore. https://github.com/Sumit189/explain-claude-mod You can interact with it instead of just reading about how things work. I think this is where AI gets really interesting. We can ask it to build the thing that helps us understand the thing.

I spent last few days talking to my terminal. Sidekick is a Claude Code mod: describe a persona in one sentence and it drives your session. Hands-free. Talk over it to interrupt. It asks you questions out loud and takes "the second one" as an answer. Open source, macOS. 🎙 Check: http://sidekick.meertarbani.in (1/7)

Claude Code is turning game modding into something else. People are using it to build insane game mods — like Minecraft running inside GTA V. And it’s not stopping there. MW2 × Skate 3. Minecraft × GTA V. More are already appearing. Game modding just entered a new era.

I've had Claude Code write most of my code for over a year, and the whole time I missed a proper view of what it was changing. Mods finally fixed that. I've got a pane in the terminal showing which layers got touched and how the changes depend on each other.

Made one too: a statusline for the Claude Code Desktop app via mods! Limits, cost, context, cache-miss cost, /compact. claude plugin marketplace add GoatEatAny/cc-desktop-statusline claude plugin install cc-desktop-statusline@cc-desktop-statusline https://github.com/GoatEatAny/cc-desktop-statusline

My current mods for claude code mods

@charliejhills These mods are like adding instruments to a cockpit: progress, permissions, and repeat actions become visible controls instead of instructions you have to remember.

Okay, that mod caught a lot of attention! And quite a few people asked to share it, I made a few improvements and pushed the repo Here it is, feel free to leave feedback or even push PR's to improve my skill and/or the mod https://github.com/JohnnyVizz/claude-kit https://x.com/savvyntsev/status/2106461309239697822

My new video on AI Coding Daily. New in Claude Code: Mods and "You Should Know" Plugin https://www.youtube.com/watch?v=_0NSNY7n5lE

同一天公司：一边把 Mythos 锁进「太强不能放」的笼子，一边把 Claude Code 的内核拆成可被 TypeScript 任意改写的 Mods。 不是插件商店通稿。刀口是：模型怕越狱，工具却把钩子插进每一次 tool call——官方自己写明：不 sandbox。

I think Claude Mods are great. Made a loading spinner so pokémon chill with me while the agent thinks Fun times

假期不停迭代！DeepSeek Harness 推出 v0.2.1‑alpha.1 新版本，增加实验性 Claude Code Mods 兼容层。 前不久 Anthropic 发布 Mods，允许通过脚本修改 AI 编程工具界面、拦截运行事件，深度定制 AI 行为。 DeepSeek 直接做了一层桥接，可运行部分 Claude Mod 示例。 但官方明确：这只是技术验证，Mods 能力只是 DSH 插件体系的子集，尚不支持完整直接迁移插件。 Harness 坚持 “一切皆插件” 架构，模型、UI、Agent 执行链路都可替换。 新版本还上线「Agent 创建插件」功能，口头描述需求，AI 就能现场生成插件，同时新增调试工具、支持外网部署。 Agent 工具生态开始出现跨扩展兼容尝试，但距离无缝互通还有距离。 #DeepSeek #Agent #大模型 #Claude

Los plugins pueden llevar hooks de shell o http en su hooks/hooks.json (que es lo que se suele usar para "escuchar" los eventos de Claude Code) Así que aunque no uses mods, puedes empaquetar tu integración de notificaciones como plugin y olvidarte de editar el settings. (7/8)