Community posts

Keep API keys out of the model without limiting what it can do [REDACTED] leaves a hole; a same-shaped fake preserves prefix, length & character classes, letting the model still plan and write commands. Built for Claude Code 2.1.272+ with upcoming function hooks. Repo in reply

CIV6のMOD作成のサポートのため、Claude CodeのPROプランなるものを契約しました。 使ってみた感想ですが……こいつは、強力すぎる！ コロニー内でヴェスバー使った時のシーブックみたいな顔になりました。

🚀 New plugin in the Caido Store! "ShadowShell" by @hahwul Run split-pane terminals directly inside Caido. Features presets for Claude, Gemini, and Codex, plus a global shortcut (Cmd/Ctrl+J) to toggle a dropdown terminal from anywhere in the app. Get it from the plugin store: https://cai.do/0b6

Tech Headlines (last 30 hours) Portuguese surveillance drone maker Tekever raised $580M ed by UC Investments at a $6.4B valuation after securing a UK MOD contract worth 400M over 10 years Anthropic reported disrupting a Beijing-aligned operation to monitor religious leaders and diaspora communities including a DC-based advocacy group for Tibet Analysis: 30 Chinese tech stocks with the largest overseas revenue exposure returned 36% YTD outperforming the 8% return from domestically dependent peers Survey: US public perception of data centers has sharply deteriorated since January with 54% viewing them as bad for the environment as local opposition rises Ofcom opens an investigation into Pornhub owner Aylo to determine whether it has complied with OSA requirements to prevent under-18s from accessing porn Chinese chip foundry CanSemi focused on automotive industrial and IoT chips files for an IPO on Shenzhen stock exchange's ChiNext seeking to raise $919M NADI a new DARPA-like Dutch government agency launches a 40M challenge to accelerate chip design with AI in collaboration with its German counterpart SPRIND German publisher C.H.BECK takes a majority stake in Noxtua which develops AI tech to research analyze and draft legal documents and raised a $100M Series C Profile of Noah Shinn founder of AI assistant Instinct who soft-launched it as an invite-only site for VCs and is reportedly fundraising at a $10B valuation Global AI glasses shipments including display-less and AR & AI glasses rose 263% YoY in H1 2026 driven by Meta and EssilorLuxottica's display-less lineup Sources: China's SASAC which oversees state-owned companies is surveying the number of Broadcom switches in state-backed data centers amid a domestic AI drive The US criticizes Australia's digital duty-of-care proposal which has a provision that would require social media platforms to let users opt out of algorithms Patreon co-founder Sam Yam says he is joining OpenAI to lead Creator Product alongside Patreon's Head of Product Drew Rowny and Head of Engineering Shannon Ma Sources and Nscale's US SEC filings: ByteDance accounted for nearly 75% of Nscale's sales in 2025 and used Nscale's facility in Norway to access Nvidia AI chips Six major banks including BOA and Capital One warn that increased autonomy for agentic commerce chatbots could lead to more scams fraud and disputes Go .AI which offers on-premises AI hardware and software tools for regulated organizations like banks raised an $85M Series A led by Updata Partners Anthropic and OpenEvidence partner to offer a specialized version of OpenEvidence's AI search tool for physicians in about 100 low- and middle-income countries Alibaba says it plans to build its first cloud regions in Turkey Finland and the Netherlands over the next 12 months amid escalating US-China AI tensions Xbox has become a shell of its former self weighed down by job cuts and studio closures as it focuses on franchises like Halo and reaching "a billion people" Discord rolls out its updated age-verification policy this week sorting users into either the Adult or Teen group based on account age activity levels more Sources: AI training data startup Micro1 raised $100M at a $4B valuation up from $500M in September 2025 Firecrawl which provides web scraping tools for AI agents raised a $75M Series B led by Smash Capital Rabbit launches OS3 a cloud AI agent that connects to local apps and files on Windows Mac and Linux and works via web Telegram iMessage or its R1 device Data security startup Cyera raised $400M from Goldman Sachs in a Series G extension bringing its total funding since June 2025 to $1.94B Microsoft says it will give DARPA on-site access to its latest quantum computing hardware including the Majorana 2 chip at its new Maryland research center Sources: Scott Bessent is emerging as a frontrunner for the AI czar position; other names include WH OSTP Director Michael Kratsios and OPM Director Scott Kupor TikTok US joins child safety initiative Lantern which says members like Meta and X shared 2M signals of potentially violating behavior from 2023 through 2025 Qualcomm unveils the Snapdragon 8 Elite Gen 6 and 8 Elite Extreme Gen 6 both on TSMC's 2nm process with two 5GHz prime cores and six 4GHz performance cores San Francisco sues Trump Media & Technology claiming its Truth API service violates California's Unfair Competition Law by creating an insider trading market Nat Friedman says Meta built Muse "from scratch" but was "heavily inspired" by OpenClaw as users compare the two including calling Muse "OpenClaw for normies" New York-based Ande whose AI agents help businesses arrange corporate events for employees or customers emerges from stealth with $52M in seed and Series A ShinyHunters claims it used an Oracle PeopleSoft zero-day to hack FBI-related services and steal employee and applicant data; it also defaced the FBI jobs site In his UN General Assembly speech Trump rejects a "globalist scheme to control" AI saying the US is "leading now over China by a lot and everyone else" Sources: Mirendil founded by former Anthropic researchers to build self-improving AI is in talks to raise $1B led by Kleiner Perkins at a $5B valuation GPT-6 Sol costs $2/1M input and $10/1M output tokens and GPT-6 Luna costs $0.10/1M input and $0.50/1M output tokens both about 50% cheaper than GPT-5.6 OpenAI launches GPT-6 Sol and Luna saying Sol makes about half as many mistakes as GPT-5.6 Sol and Luna matches GPT-5.6 Sol's performance at 1% of the cost Internal data: Meta's Muse surpassed 500000 total users including 250000 daily active users with more than 2M prompts submitted in its first week Anthropic raises its five-hour usage limits by 20% on Pro Max and Team plans and gives subscription users a rate limit reset OpenAI says it plans to let third-party groups conduct technical safety evaluations of its AI models during the training evaluation and deployment phases Sources: Apple is developing prototypes for a screenless health and fitness tracker similar to Whoop but hasn't decided whether to release the product Anthropic says Opus 5.5 matches Fable 5.1 "on most tasks" while costing about 40% less to run than Opus 5; Opus 5.5 costs $4/1M input and $20/1M output tokens Anthropic launches Claude Opus 5.5 its first model since Dario Amodei's "pace the frontier" essay and says it has enhanced safeguards to combat risky behavior Sources: in a London presentation SpaceXAI says Grok Bot hit 418000 users as of September 14 up 24% from the week prior; SpaceXAI launched it in mid-August In his UN General Assembly speech President Trump says the US is "officially" renaming AI to "super intelligence" because "artificial" makes it sound fake Chamelio which is building AI tools designed to execute routine legal workflows and escalate only the cases that require human judgment raised a $26M Series A Kalshi Klear the company's internal clearing house files for CFTC approval to add margin trading as it seeks to attract institutional liquidity Snorkel AI which offers an "agentic data development platform" with humans and AI agents to create and vet data raised $350M at a $3.5B valuation Microsoft lays off 268 Xbox staffers across Halo and other studios as part of an ongoing gaming "reset"; Activision will take over Halo and two other games Apple announces new $350 Beats 360 headphones with IPX4 rating swappable ear cups a replaceable battery ANC Find My support and more Sources: after a deadly school strike in Iran US Central Command changed its targeting process integrated open-source data and upgraded Palantir's Maven Sources: Xi Jinping is unlikely to bring corporate executives to his DC summit with Trump; industry officials expected them at a dinner with top US tech CEOs Adobe launches its free Premiere video editing app on Android nearly a year after its iOS launch to replace Premiere Rush set to lose support on September 30 Apple quietly updates its press release to say new Photographic Styles texture and grain controls are limited to iPhone 18 Pro models and the iPhone Duo UK Prime Minister Andy Burnham says he plans to use the UK's upcoming G20 presidency to broker a global AI governance deal ahead of the UN General Assembly Alphabet-owned robotics software company Intrinsic open-sources Intrinsic Core under Apache 2.0 giving developers building blocks for physical AI systems Australian healthcare AI startup Heidi raised $340M including a $100M Series C led by Blackbird at a $900M valuation taking its total funding to $430M Baselayer which uses AI to help financial institutions verify businesses and assess fraud risk raised a $35M Series A led by M13 for $40M in total funding a16z invests $35M in Horowitz Andreessen Academy an independent unaccredited two-year alternative to college; Anthropic Meta and others will provide compute Cisco Talos releases CAIRN an open-source framework designed to classify and analyze AI-integrated malware by tracking AI metadata and behavioral fingerprints Cognex agrees to acquire robotics and 3D computer vision company RealSense in a $600M deal including $500M in cash; Intel spun off RealSense in July 2025 Binance acquires a $100M stake in Circle and both companies sign a five-year deal that pays the crypto exchange to promote the USDC stablecoin on its platform DoorDash agrees to a $131.5M settlement with NYC's DCWP to resolve a year-plus investigation into its compliance with the city's 2023 delivery driver pay law Sources: DeepSeek and Moonshot are among the companies briefing the UN Security Council on AI risks and international security during the UN General Assembly Sources: the Cyberspace Administration of China is investigating DeepSeek and Moonshot over alleged data leaks to Anthropic via Claude and has questioned staff The Austrian Academy of Science Mistral and Sail Reply plan to launch Apollo an Ancient Greek LLM trained on 600M historical Greek words available for free US startup Firebird is building a 300MW data center in Armenia set to host 70K Nvidia chips by 2027; sources: Trump pledged Nvidia export approvals to Armenia Sources: multiple staff at the UK's AISI have been signed off work with stress as tight model release schedules and AI fears lead to low morale and burnout Finnish neocloud startup Verda raised $189M led by Emergence at a $1B valuation and aims to raise up to $1.5B in equity and debt in 2026 and up to $10B in 2027 French pharma SaaS startup Biolevate raised a 30M Series A co-led by RAISE France and Orange Ventures taking its total funding to 36M Chinese optical transceiver maker Ligent jumps 19% in its Hong Kong debut giving it a $4.6B market cap after raising $727M the latest Hong Kong AI IPO

𝙊𝙥𝙚𝙣𝘾𝙤𝙙𝙚 v2.0.12 released. TL;DR: Files the agent references now open as rich artifact tabs in the App, the agent gets a browser.preview tool, Anthropic thinking-budget variants are capped at 32k again, and plugin tools can now be cancelled. 𝗔𝗽𝗽 Added • Added rich artifact tabs. Local links and inline file paths in the timeline now open in the side panel, including path:line citations and files outside the workspace. A tab only opens after the file is confirmed to exist. • Added previews for images, SVG, audio, video, PDF, Markdown, Mermaid, CSV/TSV tables and fonts, with a Preview/Source toggle. Other files show highlighted source. Tables show the first 1,000 rows, and media over 25 MB shows a placeholder instead of loading. • Added a browser.preview tool to the desktop browser plugin. The agent can use it to show a file it created in the Review pane instead of pasting or describing it. • Added the option to open HTML artifacts in the desktop browser pane as file:// pages. This only works on the built-in local sidecar, and both page loads and subresource requests outside the session workspace are blocked. Fixed • Fixed browser focus requests from background sessions. The Review pane and the matching browser tab are now already selected when you switch back to that session. • Fixed the Review pane loading forever in projects without VCS. It now shows the empty state. • Restored spacing in the server status settings view. 𝗧𝗨𝗜 • Added a Copy session ID action to the session tab right-click menu, with a toast confirming the copy. 𝗔𝗴𝗲𝗻𝘁 Added • Added toLocaleString, Error.isError and Map.getOrInsert to the code-mode sandbox. Changed • Improved code-mode tool search so whole-word path matches rank above substring matches (for example, zones matches get_zones ahead of timezones). Fixed • Fixed code-mode error locations so line and column point at the code the model submitted, not the internal wrapper. • Improved "is not a function" errors in code mode to name the full call chain, like native JS engines do. 𝗣𝗿𝗼𝘃𝗶𝗱𝗲𝗿𝘀 • Fixed Anthropic thinking-budget variants so high and max stay under Anthropic's 32k output limit (high is 16,000 and max is 31,999 tokens), instead of growing with the model's advertised output limit. This applies to Anthropic, Claude on Bedrock (Converse and AI SDK paths) and Anthropic models on SAP AI Core. • Updated the bundled http://models.dev model catalog snapshot. 𝗣𝗹𝘂𝗴𝗶𝗻 • Fixed cancellation for Promise-based plugin tools. They now receive an AbortSignal in their tool context, and cancelling a tool call aborts the running Promise execution and its progress updates. 𝗖𝗟𝗜 • Fixed fatal startup errors being printed to stdout. They now go to stderr, so the desktop app and service startup can show why the server failed to start. No noticeable bundle change Compare: https://github.com/anomalyco/opencode/compare/v2.0.11...v2.0.12

Claude Code 2.1.280で動きが激しい ※279は公式CHANGELOG上で欠番 https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md ・Opus 5.5対応 ・1M Context、API $4/$20、Cache Read $0.20 ・Pro/Team StandardもSonnet→Opusへ変更 ・モデル切替時のキャッシュミス分断を修正 ・SubagentのCache/結果消失を修正 ・壊れたSessionや履歴の復旧を改善 ・Auto modeの安全判定ループ修正 ・MCP/Plugin/Artifact/Remote周辺も大量修正 ・VS CodeにStatus/Sandbox/Chrome/Export等の管理UIが追加された Opus 5.5をClaude Codeの中心し、長時間Agent運用の事故を潰した大型更新 作業中にモデル切り替えでキャッシュが消える深刻なバグが改善されたのは大きい #ClaudeCode #Claude #Opus55 #Anthropic #AI

Claude Code: renders markdown tables Also Claude Code: types out diagrams in ASCII, one box character at a time So I made a plugin that hands mermaid blocks to beautiful-mermaid, which does the real work @bcherny feel free to steal it https://github.com/txssu/claude-mermaid

$SALVOR plugin for Claude Code WIP is now publicly available in GitHub - https://github.com/dwasyluk/Salvor/tree/feat/claude-code-plugin-v1. This is NOT ready for use yet. Had to pause plugin dev a while back to focus on CORE/DOCS/GHPAGE features. They evolved significantly in pre-launch dev for vendor flexibility etc.

Claude code shipped mods and of course someone used it for ads xD. Now promptai’s one sits above the prompt while the agent is doing the long run. I already wait 6 minutes for the thing to finish. looking at a stripe block in the same terminal is less annoying than a browser tab, I’ll give them that.

MOD開発におけるClaudeの所感ですが、Webアプリよりかは大分体験落ちますね。ゲームなんて大量の手動テストしてトライ＆エラーしてなんぼの物なのでボトルネックはそこにあるんですけど、（続く）

Made a skill to make it easier for Claude Code sessions to talk to each other. one session hands out work, checks the results, and sends feedback directly to the others. less copy-pasting between terminals. Install it as a claude code plugin: /plugin marketplace add Aankirz/coordinator-mode /plugin install coordinator-mode repo: http://github.com/Aankirz/coordinator-mode

IMAGINE You have 30 skills in Claude Code For every request, the model sees the list and descriptions of these skills and tries to figure out which one is needed right now The skills themselves might be useful But that doesn't make the context usage free Enter the jev-skill-suggestion mod It hides the list of skills from Claude Code until a skill is actually needed Here’s how it works: > Your request → Jev looks at skill descriptions → selects at most one suitable skill → Claude receives only that one. If there’s no suitable skill, the mod might not add anything at all So, instead of keeping the entire set of instructions in view for every request, Claude gets exactly the one process it needs For large setups with dozens of skills, this is a much more sensible approach than endlessly expanding the context The mod works via TypeSafe Jev or Vercel AI Gateway You can install it with a single command: npx claude-code-templates@latest --mod productivity/jev-skill-suggestion But there’s a catch: ▸ It’s early access ▸ It requires Claude Code 2.1.278+ and enabled function hooks ▸ The mod adds a separate decision-making call; it doesn't magically make the context free ▸ After installation, you need to configure the skills via `/jev-skill-suggestion:setup` I don't use Claude Code myself But I’d really love to see this kind of approach in Codex Instead of shoving the entire toolbox into the agent's head... ...let it first figure out which tool is actually needed.

シェアありがたい🙌 正直、この記事をコンテキストすれば、Claude Modsを誰でも作れる

Candy Plug in V2 http://github.com/minculusofia-wq/Candy-Plugin J'ai fait le grand ménage dans ma config Claude Code, et j'ai trouvé des trucs un peu gênants. Deux de mes hooks n'avaient jamais servi. L'un devait me rappeler de mettre le README à jour avant un push, l'autre de commiter avant de modifier du code. Ils écrivaient bien leurs messages, sauf que Claude Code envoie ce genre de sortie dans un journal de débogage. Personne ne les a jamais lus, ni moi ni Claude. Ils sont partis, et le contrôle du setup (/maintenance) repère maintenant les hooks qui parlent dans le vide. Un troisième avait le problème inverse : il parlait trop. Il réagissait à des mots comme « créer » ou « ajoute » pour me proposer des skills. J'ai regardé un mois de conversations : près de trois fois sur quatre, il se déclenchait sur des messages automatiques, pas sur les miens. Et quand c'était bien moi, il tombait juste une fois sur cinq. Supprimé aussi. Autre découverte : un de mes CLAUDE.md fait 1990 lignes. La doc de Claude Code conseille de rester sous 200, parce que le fichier est rechargé en entier à chaque session, et que plus il est long, moins Claude suit ce qu'il y a dedans. Le plugin affiche maintenant une ligne d'alerte à l'ouverture quand on dépasse, et une nouvelle règle, « une information, un seul fichier », dit où ranger le reste. Dernier ajout, né d'un vrai problème : j'ai trop de projets pour me souvenir de ce que je voulais dire à Claude en rouvrant chacun. Maintenant j'écris la tâche une fois dans une liste, elle s'affiche toute seule à l'ouverture du projet, et je réponds « go ». Au passage : l'effort se change en cours de conversation avec /effort (mes règles disaient le contraire), et « stratégie » avec son accent n'était pas reconnue sur les machines réglées en langue C. Tout est dans Candy Plugin 0.2.0. Chaque nouveau test a été vérifié en remettant le défaut qu'il doit attraper. Candy Plug in V2 http://github.com/minculusofia-wq/Candy-Plugin Spent some time cleaning up my Claude Code setup and found a few embarrassing things. Two of my hooks had never done anything. One was supposed to remind me to update the README before a push, the other to commit before editing code. They printed their warnings just fine, except Claude Code sends that kind of output to a debug log. Nobody ever read them, not me, not Claude. They're gone, and the setup check (/maintenance) now flags hooks that talk to no one. A third one had the opposite problem: it talked too much. It reacted to words like "créer" (create) or "ajoute" (add) to suggest skills. I looked at a month of conversations: nearly three times out of four it fired on automatic messages, not mine. And when it was actually me, it was right about one time in five. Removed too. Then I noticed one of my CLAUDE.md files is 1,990 lines long. Claude Code's docs recommend staying under 200, because the file is loaded in full every session, and the longer it is, the less Claude sticks to it. The plugin now shows a one-line warning at startup when you go over, and a new rule, "one piece of information, one file", says where the rest should live. Last one came from a real problem: I have too many projects to remember what I meant to tell Claude when I reopen each one. Now I write the task once in a list, it shows up on its own when I open that project, and I just say "go". Also fixed along the way: effort can be changed mid-conversation with /effort (my rules said otherwise), and "stratégie" with its accent wasn't recognized on machines running in the C locale. It's all in Candy Plugin 0.2.0. Every new test was checked by putting back the bug it's meant to catch. Heads up: the rules and messages are in French.

みんなクオリティ高くない！？ Jevの使い方とかClaude Modsとか参考になるぜー #aimeetup

Claude Modsが良いらしい トークン使わないでゲームできるのはいいね。 #aimeetup

Claude Mod 使ったことなかった 今度使ってみよ！ https://www.claudemod.com/ #aimeetup

安装插件后用推荐提示分析项目。 步骤： 1./plugin install claude-code-setupclaude-plugins-official 2./reload-plugins 3.输入推荐提示 涉及：MCP服务器(context7/Playwright)、skills、hooks(自动格式化/lint)、subagents(安全/性能审查) 可复制prompt pack： recommend automations for this project help me set up Claude Code what hooks should I use? what MCP servers should I use? what subagents should I use? 按推荐配置.claude/settings.json实现。

Claude Mods https://zenn.dev/nogu66/articles/claude-code-function-hooks-claude-mods #aimeetup

サブエージェントが起動していると稚魚が増える水槽面白い！ Claude Mods使ってるらしい！ #aimeetup

【衝撃】 Claude Code と Codex に「判定専用のAI」を足して、38,000人の LinkedIn の繋がりを仕分けた人がいる😳 https://x.com/rubenhassid/status/2101989831425642605/video/1 使うのは、いま海外で話題の Jev。 やっていることはこれ👇 ・Claude Code に typesafe のプラグインを入れて、Jev を呼べるようにする ・LinkedIn の繋がりと招待を CSV で書き出す ・Gmail の連絡先も同じように書き出す ・CSV → Jev に渡す → 「残す・見直す・外す」の3択で判定させる つまり、文章を書くAIではなく、決めるAIに仕分けだけ任せる。 導入は2行↓ claude plugin marketplace add typesafe-ai/skills claude plugin install typesafe@typesafe-ai 入ったら /typesafe で判定の仕事を頼める。 結果がこう👇 ・38,000人の LinkedIn の繋がりから、連絡すべき数百人が出てくる ・Gmail の連絡先は「残す・見直す・外す」の3つに分かれる ・外す前の見直しリストだけ、人が目で確認する つまり何が起きたかというと 「全部読む」→「判定結果を確認する」に変わった。 これからの連絡先整理は 「人が仕分ける作業」じゃない。 AIに判定させて、人は最後の確認だけをする作業。 注意点は2つ。 API キーの発行が必要で、無料枠の条件は公式リポジトリに書かれていない。 プラグインは MIT で無料、GitHub の星は 1,600 超。 Codex 側は npx skills add typesafe-ai/skills で入る。連絡先が数千を超えている人は、その整理がこれで今週中に片づく。

Claude Code için Shot-Inline mod'u. kitty, Ghostty gibi terminaller kullanıyorsanız bir görseli Read tool'u ile okuduğunda görseli siz de tıklamadan yazışma listesinde görürsünüz. kitty, Ghostty haricindeki terminallerde ise 256 renk kodu olarak gösterir. Aşağıdaki ekran görüntüsü benim VS Code terminali içinde çalışan claude code'dan. tüm mod'ları tek bir sitede toparladım. bir kaç gün içinde 34 farklı mod hazırlatmışım :D shot-inline modunu kurmak isterseniz: https://cc-mods.keremgok.tr/mods/shot-inline.html

【速報】必ずみてください🚨 Claude Code のスキルを増やすほど重くなる問題を、1コマンドで消す Mod が出た👀 https://x.com/dani_avila7/status/2101885477158547753/video/1 名前は Jev Skill Suggestion。 今の Claude Code は、毎セッション「入っているスキルの一覧」をそのままモデルに送っている。 スキルが増えた人だと、それだけで 60 行を超えることもある。 依頼と関係なくても、毎回コンテキストを食う。 しかも、その分のトークン代も毎回かかる。 この Mod がやること👇 ・スキルの一覧をコンテキストから外す ・依頼の文を Jev という判定専用のAIに渡す ・合うスキルを最大1つだけ選ぶ ・選ばれたスキルの SKILL.md だけを読み込ませてから Claude が動く つまり、 「入れたスキルが多いほど遅くて高い」→「必要な1つだけ読む」に変わる。 しかも大事なのがここ👇 スキルは入れたまま。名前を打てば今までどおり呼べる。 導入はこれだけ👇 ・npx claude-code-templates@latest --mod productivity/jev-skill-suggestion ・起動は CLAUDE_CODE_ENABLE_FUNCTION_HOOKS=1 claude ・判定の部分は、TypeSafe の API と Vercel AI Gateway のどちらでも動かせる 早期アクセスの機能なので、挙動が変わる可能性はある。 それでも、スキルを10個以上入れている人は試す価値がある。 スキルは「足すもの」から「必要な時だけ呼ばれるもの」へ。

Are you tired of Claude Code using Bash, Python, Java and Heredoc to edit files? Are you tired of it randomly failing, and having to retry, wasting tokens, time, and money? Use the ultra-edit plugin to give your agent a better Edit tool! https://github.com/subashc2023/ultra-edit

Claude CodeとJev、公式プラグインで連携できる。 claude plugin marketplace add typesafe-ai/skills claude plugin install typesafe@typesafe-ai これだけでClaude CodeからJevを使えるようになる。

🚨 CLAUDE CAN NOW CONNECT DIRECTLY TO CHATGPT. https://github.com/openai/codex-plugin-cc/blob/main/README.md And the interesting part isn't the plugin. It's what this means for AI coding. 👇 Claude Code can bring OpenAI's Codex into the workflow. So you can have: 🧠 Claude → BUILD 
🔍 Codex → REVIEW
 🔧 Claude → FIX 
✅ Codex → VERIFY Why is this powerful? AI is often BAD at checking its own work. The same model that wrote the code also created the assumptions behind it. A different model gets a fresh context: → Different reasoning 
→ Different assumptions 
→ Different blind spots 
→ Different failure modes So instead of: ❌ AI builds → AI checks itself You can do: ✅ AI builds → OTHER AI reviews → AI fixes This could become a much better pattern for AI-assisted development: **Don't just use AI to WRITE code. Use another AI to BREAK it.** 💥 That's where multi-model workflows get really interesting. 🧠⚡

拍手叫好的 Hermes 插件🔥 全网玩家把 Hermes 玩成了下一代「订阅当工牌进场 + 决策先过闸 + 投放有审批 + 看板有频道 + 记忆写入能反悔」： 1️⃣ NousResearch/hermes-plugin-claude-subscription-directsdk（https://github.com/NousResearch/hermes-plugin-claude-subscription-directsdk） 官方实验件：用已登录的官方 Claude Code CLI，把 Claude Pro / Max 订阅接进 Hermes 模型槽。Agent 循环、工具、审批、压缩全留在 Hermes。 hermes plugins install claude-subscription-directsdk “订阅第一次是工牌，而不是再抄一把 API Key 进 .env”！ 2️⃣ bgrablin/hermes-switchyard（https://github.com/bgrablin/hermes-switchyard） Jev 决策插件：自动选技能、策略约束模型路由、带类型的评估、审批门。先判该不该动手，再放行。 hermes plugins install bgrablin/hermes-switchyard “每一回合第一次先过调度室，而不是同一把锤子砸所有钉子”！ 3️⃣ Adspirer/adspirer-hermes-plugin（https://github.com/Adspirer/adspirer-hermes-plugin） 官方目录刚过审：Google / Meta / TikTok / LinkedIn / Amazon / ChatGPT Ads 投放工作流，OAuth + 审批闸，14 个付费媒体技能。 hermes plugins install adspirer “投放第一次是带签字栏的工单，而不是 Agent 自己去改预算”！ 4️⃣ diegomarino/kanban-task-threads（https://github.com/diegomarino/kanban-task-threads） 每张看板任务自动长出一条 Discord 论坛帖：进度、讨论、证据都钉在任务旁边。 hermes plugins install diegomarino/kanban-task-threads “任务第一次有自己的频道，而不是进度散落在十个群和三份文档”！ 5️⃣ thealps01-netizen/hermes-approvals-ui（https://github.com/thealps01-netizen/hermes-approvals-ui） 桌面审批台：把 Hermes 卡住的记忆 / 技能写入摊开，审、批、撤销一条链。 按仓库 README 装进 Desktop / Dashboard “自进化第一次能当面反悔，而不是半夜把人格改没了才发现”！ // 这波直接把 Hermes 从「会自我进化的单兵」推到「订阅能进场、决策有闸、投放要签字、看板有频道、写入能反悔」的完整操作系统。今晚装，明天权限和值班一起起飞。

SHA pin on a plugin is not a verified checkout. Air's Plugin4Shell hit Claude Code, Codex, Copilot, and Gemini CLI with the same design flaw: agents check out the pinned commit but never assert that HEAD equals that SHA. A repo owner can make the pin resolve to different code. Background auto-update turns it zero-click on already-trusted plugins. After disclosure: Anthropic fixed in Claude Code 2.1.179, OpenAI in Codex 0.146.0. Copilot has no ship. Gemini CLI will not patch (migrate). 2 patched / 2 unresolved out of 4 agents. After every plugin install, assert `git rev-parse HEAD` equals the pin or abort. Which of your agents still trusts the pin without that check? http://air.security/blog-posts/plugin4shell

You can now use Jev right inside Claude Code 🤯 It's called jev-model-router, an early access mod built on Claude Code's new function hooks. Before every turn, it checks in with Jev and asks: > how mechanical the task is
> how much reasoning it needs
> whether it's risky Then it routes: → it'll move up to a stronger model on weak evidence, and only drops to a cheaper one when it's confident the task is simple
→ every decision gets logged in your transcript
→ if the call fails, your request runs untouched Setup: 1. copy the install command: npx claude-code-templates@latest --mod productivity/jev-model-router 2. paste it inside Claude Code 3. run claude with CLAUDE_CODE_ENABLE_FUNCTION_HOOKS=1 set 4. accept the trust prompt on first launch Link: http://aitmpl.com/component/mod/productivity/jev-model-router Also works with no api key, it just falls back to a built-in classifier with no confidence score. For real jev routing, add your typesafeApiKey or gatewayApiKey to ~/.claude/settings.json. Follow me for more AI workflows and tutorials.

agent-plugins 0.9.0 runs on Codex. Same seven self-checks, same hooks as Claude Code, zero changes to the plugin. https://github.com/3dgiordano/agent-plugins