#claudecode — Public Fediverse posts

New rule in my global CLAUDE.md: no agents unless I ask for one, or Claude Code asks me first. And agents are not allowed to start agents.

Agents work out of sight. One searched my whole disk to find a package, another filled 8 GB of Docker memory with its builds. Each one also starts without the conversation, so it spends tokens rebuilding what the main session already knows.

There is also an incentive to keep in mind: Anthropic sells usage plans, so it gains when Claude Code spends more tokens and you move to a higher plan. Whether or not that shapes the defaults, I prefer to set them myself.

Now the main session does the work with its own tools, and asks me when an agent would really help.

#ClaudeCode #AIAgents #Productivity

On Friday I wrote that Opus 5.5 does my work and a Fable 5.2 agent checks it, because Fable is too token hungry for everything. A new benchmark backs that up.

Deepnote's Agent Initiative Benchmark tested 12 models as proactive agents in the Slack of a simulated company: help when needed, stay quiet when not.

Fable 5.2 scores 4% higher than Opus 5.5, but makes twice the tool calls in twice the time, at $1.87 a task against $1.28. OpenAI's GPT-6 Astra and GPT-6.1 Sol top the list, Sol at $0.11 a task.

The benchmark, dataset, and code are open source:
https://deepnote.com/research/benchmark

#AIAgents #ClaudeCode #LLM

🚀 Deploy #Claude Code on #Ubuntu #VPS
This article demonstrates how to deploy Claude Code on Ubuntu VPS.

What is Claude Code?
Claude Code refers to using Claude (Anthropic’s AI model family) as ...
Continued 👉 #anthropic #nodejs #selfhosted #letsencrypt #claudecode #selfhosting #reverseproxy

🚀 Deploy Claude Code on Ubuntu...

【Agent Skills】モデルが賢くなるとスキルはどうなる？
https://qiita.com/ryu-ki/items/202d150ad48dbccf7981?utm_campaign=popular_items&utm_medium=feed&utm_source=popular_items

#qiita #生成AI #Claude #AIエージェント #ClaudeCode #AgentSkills

От чата до тёмной фабрики: шесть уровней ИИ в разработке и почему мы остановились на L4

Больше года я внедряю ИИ в разработку зрелой SaaS-компании, вместе со сложившейся инженерной командой, которая поначалу относилась к ИИ скептически. Мы прошли большой путь: от копирования кода из ИИ-чата до того, что отдаём агентам разработку фич целиком. Мы не занимаемся вайб-кодингом и не делаем продукт за вечер одним промптом. То, чем мы занимаемся, я называю AI-инженерией : мы строим систему, которая стабильно выдаёт результат по нашим стандартам качества. Сейчас мы экспериментируем с полуавтономной агентной фабрикой в облаке. Этот путь я разложил на шесть уровней, от привычной ручной разработки до «тёмной фабрики». Ниже — как меняется жизненный цикл разработки на каждом уровне, как понять, где сейчас ваша команда, и что нужно, чтобы двигаться дальше.

https://habr.com/ru/articles/1106718/

#sdlc #codereview #ииагенты #claudecode #aiинженерия #specdrivendevelopment

Ars Technica: Claude Code is accidentally exposing secrets all over GitHub. “Claude Code is leaking sensitive information on public GitHub instances thanks to a partial oversight in the way its settings files are handled. That’s the finding of AI security researchers at Apiiro, who discovered thousands of secrets, including database passwords, API tokens, and root logins, amid Claude Code […]

Entropic Thoughts: Claude Code token cost explorer. “I tried some existing tools like ccusage and claude-agent-ledger, but they weren’t able to give me the sort of breakdown into specifics I needed to optimise my usage. Thus, I shamelessly had Claude Code whip up a tool that gives me a better view of its token usage.” This article is aggressively footnoted; a footnote was omitted from […]

I gave two design skills the same brief

https://derekhanson.blog/i-gave-two-design-skills-the-same-brief/

Two models, two jobs in my Claude Code setup.

I've tested Fable 5.2 since its release: very powerful for creative tasks and for reviewing work from a different angle, but probably too token hungry for all my workflows.

So Opus 5.5 does the work, and a Fable agent checks it, midway through a task or before my colleagues, the subject matter experts, review it.

#ClaudeCode #AIAgents #TechnicalWriting

Why #HI is better than #AI

I need to procedurally generate 265 videos which represent #SessionReplay for each of the #web pages that were #audited for my #website #surveillance #research paper.

#ClaudeCode wanted to spend 74 days rendering the videos in #Blender - I said no, instead I suggested using #playwright / #VLC to record the existing replays from the interactive 3d web app created to explore the #data...

Time to generate 265 videos... 3h Hanff 1 : ClaudeCode 0

Dew Drop - November 13, 2026 (#4792)

https://alvinashcraft.com/2026/11/13/dew-drop-november-13-2026-4792/

Claude Fable 5.2はOpusの単価2倍、でも1タスクは約半額!? Mythos5.2との違いと使いどころ
https://qiita.com/ryosuke_ohori/items/17995524b433cf04fd55?utm_campaign=popular_items&utm_medium=feed&utm_source=popular_items

#qiita #Python #LLM #Anthropic #Claude #ClaudeCode

Reviewing a training course after a product release? A Markdown diff is hard to read, so I asked Claude Code for a change page.

One command writes one HTML page:
- tracked changes word by word, or whole passages when rewritten
- before and after for screenshots and diagrams
- each change labeled by lesson, section or quiz question
- a Reviewed checkbox per file

I regenerate it after every edit, and the AI review agents check the same changes I do.

#TechnicalWriting #DocsAsCode #ClaudeCode

I love Claude Code, but I use it so much that I regularly exhaust my weekly quota before the week ends. https://hackernoon.com/agent-skills-compatibility-how-claude-code-vibe-and-copilot-compare #claudecode

"The policy tweaks show that big AI corporations will protect other big corporations when their users find ways to use LLMs to circumvent the artificial constraints that hardware manufacturers put on their devices. AI-powered copyright circumvention is OK with big tech when it’s a massive company scraping the entire internet to build their LLMs, but is not OK when people then use those tools to liberate their internet-of-things devices. It’s not the end of AI-powered reverse engineering, but people suddenly getting locked out of their accounts for asking AI to reverse engineer their gadgets suggests that the future of this practice lies in open weight, local, and uncensored models, making it less accessible to the masses and the people who could probably benefit from it most.

Tarun Gogineni, an OpenAI engineer, tweeted that people
are not supposed to have their OpenAI accounts restricted for reverse engineering: “these blocks are mostly accidental and a result of difficult decision boundaries; there is no prohibition against reverse engineering with codex. it’s getting fixed,” he tweeted.

But people who are doing AI-powered reverse-engineering suggest that they are moving away from big companies’ models to alternatives where they aren’t going to have their accounts disabled or have to figure out how to work around AI company guardrails."

https://www.404media.co/reverse-engineer-anything-ai-companies-crack-down-on-the-coolest-thing-people-were-doing-with-ai/

#AI #GenerativeAI #LLMs #ReverseEngineering #Programming #SoftwareDevelopment #VibeCoding #ClaudeCode #Anthropic #OpenAI #Codex #Copyright #IP

財源幻想はなぜ生き残るのか——三層防衛構造という設計バグ : システム設計視点の行動経済学 (61)
https://qiita.com/maskot1977/items/0f8816d5e85ba791176f?utm_campaign=popular_items&utm_medium=feed&utm_source=popular_items

#qiita #システム設計 #Gemini #行動経済学 #ChatGPT #ClaudeCode

I find that using #claudecode via a remote control server is the _best_ way to use #claude. I have a powerful workstation with claude where it has excellent bandwidth to the cloud. Then I control the #ai from my phone or laptop even in low-bandwidth settings.

blog.lhotka.net/2026/10/08/R...

The recording of my closing keynote at AI DevCon NYC is now on YouTube: "A Dream Factory for an Instrument Made of Light."

How I built Glinster in two months with Claude Code while reading very little of the code, what went wrong, what stayed with the player, and a live performance on Vision Pro at the end. The video has chapters.

https://youtu.be/5bDYF02lt1E

Jump straight to the performance at 31:10: https://youtu.be/5bDYF02lt1E?t=1870

#AIDevCon #ClaudeCode #VisionPro #MusicTech

#Claude Code podporuje režim učení, kdy se vás to při řešení problémů doptává nebo vás to nechá napsat kus kódu. Výborný nápad, pokud chcete AI používat a zároveň aspoň nějak zůstat v obraze, co děláte.

https://code.claude.com/docs/en/output-styles#learning

#claudecode

https://www.europesays.com/dk/203859/ Maersk Engineer Dmitry Buykin Says AI Agents Made Code Cheap — and Maintenance More Expensive — BigGo Finance #Addyosmani/agentSkills #AIEngineer #Anthropic #ByteDance #ClaudeCode #Cursor #DeepSeekHarness #DmitryBuykin #GitHub #Google #Mærsk #OpenMausBot

CLAUDE.md に「このコードはあなたのもの」と書くと、重複したコードをまとめてくれるのか。Claude Code・Codex・Gemini CLI で132回試したら、まとめたのは Claude だけだった
https://qiita.com/suwa_nobu/items/1dd9df248ee6f4c9dfd4?utm_campaign=popular_items&utm_medium=feed&utm_source=popular_items

#qiita #AI #生成AI #Claude #AIエージェント #ClaudeCode

#Development #Releases
Google releases Modern Web Guidance 1.0 · The setup that avoids unwanted code https://ilo.im/16hvnx

_____
#Programming #Coding #AiSkills #AiAgents #Prompts #AI #ClaudeCode #WebPlatform #WebDev #Frontend

NEW! Leanpub Track LAUNCH 🚀 AI Agents Masterclass: Understanding, Building, and Managing AI Agents by Thomas De Vos, Ian Johnson, Russ Miles and J. Owen

https://youtu.be/HB2fNhmjjnM

https://leanpub.com/set/leanpub/aiagentsmasterclass

#courses #leanpublishing #selfpublishing #AIAgents #ClaudeCode #HarnessEngineering #CodingAgent

日々アップデートを重ねる「yomiyasu」、日本語の推敲はどう変わる？v1.0.1・v2.5.0・v3.0.0に同じ文章を通して比べてみた
https://qiita.com/inoyu-qiita/items/b0a7aee8d9f97cc21e2c?utm_campaign=popular_items&utm_medium=feed&utm_source=popular_items

#qiita #AI #生成AI #AIエージェント #AI活用 #ClaudeCode

【改定前受験推奨】Claude Certified Architect – Professional（CCAR-P）本日受験してきました。【合格体験記】
https://qiita.com/takeshi_Qiita/items/a84c8eab09c878071daa?utm_campaign=popular_items&utm_medium=feed&utm_source=popular_items

#qiita #AI #Anthropic #Claude #ClaudeCode #CCAR_P

US venture capital concentration reached new heights in 2026, with established firms capturing 88.2% of total capital while emerging funds secured only 4.5%.

Despite the capital imbalance, Carta data indicates that micro-funds ranging from 1 to 10 million dollars frequently outperform larger vehicles based on TVPI metrics.

An inside look at how boutique funds like RedBlue Capital structure investments, utilize custom dat…

https://en.hacks.gr/poso-dyskolo-einai-simera-na-xekiniseis-ena-neo-ependytiko-tameio-stis-ipa/

#VentureCapital #Startups #Finance #ClaudeCode