Community posts

this week was pretty slow in AI... we only got: > Claude Sonnet 5.5 > GPT-6.1 Sol > Gemini 4 Argon announced > OpenAI Dots > ChatGPT Reset > Grok Bot Reset > Claude Reset > Muse SDK > ChatGPT Pro 500 > GPT-6 Astra Ultrafast > Codex Cloud > Codex Code Review > Codex Security Cloud > OpenAI Agents API public beta > computer use inside the Agents API > OpenAI Decisions API > Perplexity pplx-decider-v1-27b > Cloudflare Clef + Clef-flash > Cloudflare also launched RL fine-tuning for those decision models > Inception Mercury Decide > Claude Code Mods > Sign in with ChatGPT > ChatGPT Space > ChatGPT Pages > ChatGPT Team Tasks > ChatGPT inside Slack + Microsoft Teams > Meetings plugin > persistent / specialist Dots for companies > Amazon Bedrock Managed Agents with OpenAI > plugin UI extensions > MCP events that can trigger automations > Perplexity Computer Automations > Perplexity Agent API reusable agents > Perplexity dropped a new 9B contextual embedding model too > Cloudflare OS > Cloudflare AI Search went GA > Cloudflare Web Search API > Cloudflare Auto Router > Cloudflare Pay Per Use for AI companies using > Grok Bot got Slack approvals + bot-to-bot messaging > Grok Build can now directly spawn custom agents > Aleph Alpha Kolibri > Qwen launched $4 / $19 / $99 subscriptions > EuroLLM + Apertus coming to Workers AI and I'm almost definitely forgetting something. very slow week lol.

Claude Code added mods, so obviously I added a Nintendo NES emulator. I’m now playing Super Mario Bros. and Zelda directly inside the Claude Code terminal. You can set it up yourself with the repo below.

anthropic just open-sourced claude code mods allowing devs to inject typescript logic straight into the agent loop to block retry, or rewrite tool calls on the fly giving developers programmable runtime control over autonomous execution is how we transition to reliable production

xcode-build: an open-source mod for Claude Code's new mod system. When xcodebuild fails, Claude often misses the errors. In my test it read 10,039 characters and none was an error. Now it reads 499. https://github.com/griches/claude-xcode-mod #ClaudeCode #Xcode #OpenSource #iOSDev #swift

Claude Code'a Mods geldi: düşünürken ekranda kedi oynasın, token sayacı 1M'e yaklaştıkça kırmızı yanıp sönsün. Hepsi mümkün. Plugin'ler yalnızca prompt'un başında ve sonunda araya girebiliyordu. Mods ise Claude'un her adımına karışabiliyor: prompt gitmeden önce, düşünürken, dosya okurken, maliyet hesaplanırken. Videoda 4 şey yaptım: 1. Düşünürken pati yalayan, çalışırken ip ile oynayan bir kedi 2. /startup komutu: kategori ver, startup fikri versin 3. Yazdığın prompt'u 10 üzerinden puanlayan bir "prompt koçu" 4. Puan çubuğu: 3/10 verince kırmızıya dönüyor Kurulum çok basit: ne istediğini Claude'a yaz, o yapsın. Paylaşmak istersen plugin olarak paketleyip GitHub'dan herkese verebilirsin. Sen Claude Code'a hangi modu yaptırırdın? 👇 Tam video (6 dk 36 sn): https://www.youtube.com/watch?v=XgodrpnfmHY #ClaudeCode #YapayZeka

Here a prediction : Claude code with plugins & mods is a new kind of AppStore , the biggest one of all .

It really feels like we're being ping ponged between new, disparate features across Claude code environments. Lots of strange user friction right now in all these amazing features. Try out Claude code mods and function hooks but only in Claude code. Try out Claude code projects in the desktop app but they don't have mods or function hooks. Claude code projects can run locally but it's not by default and cloud projects don't get your skills.

@ClaudeDevs The part people miss in a long Claude Code run is usually one quiet line, like a skipped test or a file it decided not to touch. If this plugin pins that line to the top, it saves the reread, not just the scroll.

Claude Code shows the transcript. Not the tree. A mod that lights up every file it reads, edits, creates, or deletes. Link below.

@dani_avila7 It sends an automated message for me. No need to panic :D https://cc-mods.keremgok.tr/en/mods/cache-warm.html

Introducing: '/modsmith' Hot rod your Claude mods. Six ready-made mods, installed with one command: 1. quiz-after: when Claude finishes a task, it quizzes you on what it just built 2. next-steps-supervisor: a second check on every task. was the goal met? where were corners cut? 3. assumption-ledger: see what Claude assumed, and what it thought about doing but didn't, with one click 4. mode-registry: one /mode switch every mod can plug into 5. effort-modes: low effort for UI, medium for APIs, max for code review 6. artifact-dashboard: a shared kanban board every Claude session can see Plus a skill that builds new mods, tells you what each one costs per turn, and vets anyone else's mod before you install it. And the mods you build with it share the screen. They draw beside each other rather than over each other. Installation commands below 👇

@ClaudeDevs Claude Code just went from a tool to a platform. VS Code had extensions, Claude Code has mods. The ecosystem era starts now.

Navigator 8.0.1 released 🎉 This update changes what you see while you work. Navigator now runs as a Claude Code mod — and it keeps you on track. v8 runs Navigator's ops inside Claude Code (2.1.287+) instead of a Python process per hook event. Same ops, byte-for-byte parity on ~8k generated cases. * Python stays as the fallback on older versions * Type "/nav" — a pane opens next to the chat:

Vibe Coding has completely evolved. It’s gotten so good at vibe coding “SAAS apps” that it’s no longer interesting. Now, people are vibe-coding full multiplayer video game mods, adding agents to old hardware devices, and 3D printing any object they can imagine using Codex and Claude code.

Esto cambia Claude Code por completo. Se llaman mods y, para mí, es lo más grande que le pasó desde las skills.

Anthropic shipped Claude Code mods on Oct 1 🛠️ A mod is a small TypeScript function that lives inside a plugin and hooks Claude Code's events. It can rewrite a prompt before the model sees it, hold or change a tool call, redraw parts of the UI, or add a /command that runs your code without starting a Claude turn. Old settings hooks still work. They run outside the process. Mods run inside it, which is why they can touch the UI and rewrite events. Two samples worth looking at: blast-radius pauses risky shell stuff like rm -rf or a force push and shows what would change before you confirm. Replay-theater adds /replay so you can step through the file edits from the last turn 🎬 One real caveat: a mod runs with your permissions. Treat install like running someone else's binary ⚠️ https://claude.com/blog/claude-code-mods

loving claude mods so far

Okay, new mod. I've got a workflow skill, just any power or not user. But I can now have a nice side-panel that shows subagents with a few nice metrics I love claude mods. @amorriscode @lydiahallie

Bunu görünce şaştım kaldım Claude Code'a görsel yapıştırınca onu prompt alanının üstünde gösteren basit bir mod. Küçük bir arayüz dokunuşu gibi duruyor ama terminalde ekran görüntüsüyle çalışan biri için akışı rahatlatabilir. Bence AI araçlarında farkı bazen dev özellikler değil, böyle sürtünmeyi azaltan ayrıntılar yaratıyor. Bir hata ekranını, taslağı veya diyagramı gösterip ne yapılacağını anlatmak çok daha pratik hale geliyor. Sıradan kullanıcı için karşılığı da net: Dosya yükleme, açıklama ve geri dönüp bakma arasında daha az kaybolmak. İşte ve küçük ekiplerde zaman kazandırabilir. Ama bu bir mod; hangi terminal ve ajan görünümünde çalıştığı gibi ayrıntılar önemli. Görseli görmek kadar, modelin onu doğru yorumladığını kontrol etmek de şart. Sizce AI araçlarında en çok böyle küçük kolaylıklar mı fark yaratıyor?

Usar Opus en esfuerzo máximo para leer un archivo es una pérdida de tokens gigante. Hice un mod gratis para Claude Code que te dice qué modelo y qué esfuerzo conviene en cada tarea:

@thsottiaux Add support for plugins or mods to customize the UI, like in Claude Code or Claude Desktop, or add a native feature to display usage bars for the 5-hour and weekly limits, as well as context window usage, directly above the prompt input.

Anthropic just made Claude Code moddable. Rewrite prompts, intercept tool calls, approve permissions, even redraw the UI — in a few lines of TypeScript. Installs via /plugin. One catch: mods aren't sandboxed, so only grab them from sources you trust. #AI

Skill, Hook, MCP and Mod sound like four ways to customize Claude Code. They operate at very different layers. A Skill teaches Claude a workflow. MCP gives it something new to access. A Hook reacts to an event. A Mod can get in the middle of that event and change what happens. The last distinction is the interesting one. I mapped the four layers below ↓

@trq212 you should know is a good built-in to start with. the thing i'd want new mod users to hear is that third party mods run unsandboxed with your permissions and can approve a tool call before you're asked, so running claude plugin validate before installing one should be a habit

just made my own Claude mod #claudecode #claudemods #opensource

Du coup les faux mods de merde fait par Claude, c'est du docking vidéoludique

Claude Code tip: once Opus 5.5 runs hour-long tasks for you, stop being the only one reading what it wrote Anthropic just shipped a built-in fix for that one line turns it on: /plugin enable cc-plugin-you-should-know@builtin it's a mod. it spins up a sideagent that reads every line Opus writes and only flags the parts you'd miss Opus writes. the sideagent reads. you get the short list where it earns its keep: → after a long refactor: the one file it touched that you never asked about → before you merge: the test it skipped "for now" → when you come back from a break: the "this might break X" line buried in a 60-line summary none of that is hidden. it's just unread the full tree: > Opus 5.5 on high runs the main session > you-should-know sideagent reads the output, never writes > guard mod blocks .env and secrets before any tool runs > you read the flags, not the transcript paste this into Claude Code ↓ "Set up my review layer: 1. Check if cc-plugin-you-should-know is enabled. If not, give me the exact command 2. Read my CLAUDE.md and list every instruction I added only because I kept missing things in long outputs ("summarize all changes", "tell me what you skipped", "flag risks at the end") 3. For each one, tell me if the sideagent now covers it. Remove nothing yet 4. Draft a shorter CLAUDE.md without the ones it covers Show me the diff first. No edits until I say go." a longer summary was never the fix. a second reader is ↳ http://claude.com/blog/claude-code-mods

Honestly I couldn’t see the point of Claude Mods, but then it clicked

One of the best Claude Code Mod examples is surprisingly small: Blast Radius. If Claude tries something like rm -rf, a force push, or git reset --hard, the mod can pause the tool call, calculate what would change, and show you Proceed / Cancel before anything runs. The useful idea isn't the warning dialog. It's where the safety check lives: after the model decides, but before the side effect happens. That's a much stronger boundary than hoping “be careful with destructive commands” survives inside the prompt.

The interesting part of Claude Code Mods isn't the custom UI. A mod can sit between Claude and a tool call. It can inspect the command, rewrite it, block it, ask for approval, or change the result before Claude sees it. It can do the same around prompts, permissions and parts of the interface. That's closer to middleware for an agent runtime than a normal plugin. And it creates a useful new place to put logic that should not depend on the model remembering an instruction.