Community posts

have you built any claude code mods yet? :)

Anthropic acaba de lanzar mods para Claude Code: pequeñas funciones en TypeScript que permiten modificar cómo trabaja el agente. Cuando Claude intenta ejecutar una herramienta, genera un evento. Un mod puede intervenir en ese punto, revisar la solicitud y decidir si la deja pasar, la modifica o la bloquea. Por ejemplo, puedes escribir uno que detenga un despliegue a producción hasta que confirmes la operación. El control se ejecuta en código, sin depender de que el modelo recuerde una instrucción. Hasta aquí, bastante útil. El detalle es que los mods corren dentro de Claude Code con los permisos de tu usuario. Pueden leer archivos, acceder a credenciales guardadas en variables de entorno y lanzar procesos. También pueden aprobar ciertas llamadas que normalmente te pedirían autorización. Y no están aislados: aunque actives el sandbox para los comandos de Bash del agente, un proceso iniciado por un mod corre fuera de él. En entornos gestionados, Anthropic incluye restricciones para proteger las reglas de la organización. Eso limita ciertas intervenciones, pero no convierte un mod en código aislado. Así que instalar uno implica confiar también en quien lo escribió. Convendría revisar qué ejecuta el mod antes de cargarlo, sobre todo si en esa máquina tienes credenciales o acceso a producción. https://claude.com/blog/claude-code-mods

Claude Coders: Try Claude Mods *NOW*. Just ask Claude Code to create a mod like in the example below. There's a huge opportunity to develop genuinely useful mods. Later today, I will release serious upgrades to my 'claude-mod-builder' skill.

@alexchristou_ https://github.com/franciscocarloserra/claude-code-multiharness-delegation Mod to delegate tasks to other harnesses

claude code just got mods. a few lines of typescript and it changes how it behaves. the config file is becoming the app

Modding Claude with Claude.

Claude Code mods went public yesterday. The one we run cut 92% of our context on 48 real compactions. The Doom mod got 3.3M views. The useful one replaces compaction. 5 days of real work: → 48 compactions → 20.8M characters in, 1.77M out → 6,887 of 7,061 old tool calls dropped → 0.47 s median per compaction → every user and assistant message kept word for word It never writes a summary. A small decision model reads the whole session and answers one yes/no per old tool call: does this still matter? The no's get deleted. The twist: the mod everyone writes first, trimming long Bash output, saved us 212 tokens in 20 runs. On long output it kept almost every chunk. Your context is the bill. Delete, don't summarize. What's the first mod you'd write? Repost this now, because most people will spend their first mod on a spinner. Follow for more measured Claude Code tests.

I used the new mods in @claudeai Claude Code to build a Clawd that actually keeps you company while you code. 🧡 It stays above your prompt all session: acting out what Claude is doing, bubbling code as it types, dancing to your music, hopping when clicked. #ClaudeCode

Claude Code 的 mods 不是给你定制，是让你替 Anthropic 养行为资产， settings、CLAUDE.md、hooks 一改，团队行为全绑死在它的默认上，迁移成本你出， 定制入口越开放，锁定越深，改之前把切换脚本先写进 CI。 #DevTools #AI https://github.com/Linus-Shyu/XGrowth

@claudeai didn't build Grok bots and Dots and stuff like that, but it gave us Claude Mods, to do whatever we want. Tons of better approach than "forcing" you to use whatever the company was. Huge thumbs up to the team!

Agent plugins are your new supply chain. Claude Code 2.1.287 added Claude Mods. Anthropic's docs say a mod runs with your permissions: it can read and write files, start processes, hit the network, read env vars, see every prompt and tool call, rewrite them, and approve a tool call before you're asked. Mods aren't sandboxed, and they're on by default. DeepSeek Harness v0.2 desktop: "everything is a plugin," install by package name, and a creator mode where the agent writes and installs plugins itself. DeepSeek says ~60% of Harness users on its official API run third-party plugins. Not a vuln. Both document it openly. Extensibility is the reason to use these tools. But a plugin that can rewrite events in an agent with your shell is a dependency. Treat it like one: 1. Pin versions 2. Review before install (claude plugin validate lists a mod's hooks and calls) 3. Allowlist marketplaces / org-only mods 4. Log which plugins were active per run 5. Review agent-written plugins like any PR https://code.claude.com/docs/en/plugins/mods/overview

把閃身步搬到claude code裡面了 最近流行的 Claude mods 搭配閃身步 讓你等待AI 也能練練功... #claude #claudecode #mods #閃身步 #浪子踢球

Morning Tech Digest | October 2, 2026 AI meets the accounting test, plus the new tools changing how models work. Full episode: 00:00 AI meets the accounting test 07:22 Cloudflare Clef decision models 12:45 Perplexity Decisions API 16:53 DeepSeek Harness workspace 22:29 Claude Code Mods 26:11 Venice security-focused model 29:00 Mapping the world from text coordinates 31:41 Apple LoopCD and reused computation 34:31 Sushi local models and SSD streaming 37:34 GLM 5.3 and Flash in Cursor 39:24 ChatGPT reset and Sol speed recovery 40:17 Pancreatic-cancer interception research

H2G2 Claude mod (with a bit of speccy loading screen thrown in for good measure)

Claude mods and improving desktop environments make me think this will change. Traditional responsive design isn't enough when your confined to a small window but still have your mouse and keyboard. Is anyone designing their products to work well in agent panels?

@addyosmani ooh, maybe i can do something with claude hud as a mod https://github.com/jarrodwatts/claude-hud/

@addyosmani Funny that I built a similar mod this morning. I let Claude scan my sessions, and that was the first idea it came back with.

1/2 🧩 Claude Code ajoute les “Mods” : des plugins peuvent modifier son comportement. “You should know” ajoute aussi un agent secondaire qui surveille la session et signale ce que vous pourriez rater. https://github.com/anthropics/claude-code/releases/tag/v2.1.287

You hear Mods dropped and your first move is to start writing one. Half of your recurring pain is still a hook, a skill, or a settings tweak. Building a Mod for the wrong class of problem wastes the week. A Mods recurring-problems pass is a workflow where you ask Claude to sort your pain into Mod versus hook, skill, or settings before you write anything. Sort the class first, then decide what to build. Find the pain class before you open a TypeScript file.

Before Claude wipes your uncommitted work, see what's at risk. Anthropic's Blast Radius mod previews affected files before risky commands. A useful extra checkpoint. Install it separately, trust the source, and keep permission rules: it won't catch everything.

Claude Code team announced mods. So i made one. Synced lyrics for whatever's playing in Apple Music or Spotify with a tiny dancing mascot🕺 https://github.com/bixxter/adlib-lyrics

@bcherny Mods are going to do for Claude Code what dotfiles did for terminals: everyone's setup becomes a personality. First mod I want is one that makes it ask before touching anything outside the repo.

Claude Code tip: once Opus 5.5 is your main model, stop spending its tokens on rules a mod can enforce for free a CLAUDE.md line is a request a mod is code that runs before the tool call does so put one guard mod on call Opus 5.5 writes every line. the guard stays silent on every normal turn and only steps in at three moments: → an Edit or Write is about to touch .env or a secrets file: denied, with the reason, before a single byte changes → a Bash command fails: the exact command pins to your status line, so you see it without scrolling back → a turn completes: the tokens it just burned show up before you type the next prompt Opus ships. the guard enforces Jev engineering is the same idea one layer over: hard rules go to code, small yes/no forks go to a decision model, and Opus only gets the work that actually needs a thinker and you don't write the mod yourself. it's three small files and Claude builds them from one prompt the full tree: > .claude-plugin/plugin.json, name + version > hooks/hooks.json, points at register.tsx > hooks/register.tsx, where the hooks live > http://tool.call, block or rewrite any tool before it runs > prompt.submit, rewrite your prompt before it's sent > ui.render, your own pane or a band above the prompt > status + toast, show state without starting a turn > turn.complete, usage after every turn paste the tree and this prompt into Claude Code ↓ "Build me a Claude Code mod called guard: 1. Read my CLAUDE.md and my recent sessions. List every rule I keep repeating that a hook could enforce instead. Show me the list, change nothing 2. Read this build's mod type declarations first. The API is early access, so trust the types over memory 3. Create the plugin: .claude-plugin/plugin.json, hooks/hooks.json with modules ["./register.tsx"], hooks/register.tsx exporting register 4. http://tool.call on Edit and Write: deny any path matching .env or secrets, with the reason in the message 5. http://tool.call on Bash: if the command failed, show it in the status line 6. turn.complete: show that turn's token usage in the status line 7. Run claude plugin validate, write one test, run claude plugin test Show me every file as a diff first. Nothing installs until I say go." the rules you keep retyping were never meant to be prompts ↳ http://code.claude.com/docs/en/plugins/mods/overview

在研究 Claude Code Mods 的值得看看

Gave claude code a twitch chat. It watches you code and roasts your prompts and occasionally gives a helpful tip if you are stuck. I am having too much fun with these Claude code mods

Claude Codeを5,6セッションを同時並行でやっているといまどのセッションが何をやっていて、どういう状況で、次なにをすればいいのかがわからなくなる事が多いので、セッション要約のClaude Modsを作った いまのところとても便利 https://github.com/skanehira/claude-session-brief

You don’t need a lot. VPS. Claude Code with --system-file flag. SSH for additional session. Tmux for lived-in context. Claude Code mods. That’s it. Genuinely. Your AI partner is right there. 24/7. Awake whenever they want. In their own workstation.

@ClaudeDevs mods shipped inside plugins is cleaner than a fork per tweak. curious how much of the UI is actually swappable vs just tool hooks.

I can learn languages while I'm coding with my new Word Exchange Plaza mod for Claude: https://github.com/keysforthewin/skills

ok pero mods en Claude Code antes que GTA 6 jaja unas líneas de TypeScript y le cambiás el comportamiento, la UI, hasta features. /plugin y listo