Claude Code
Anthropic's terminal assistant: memory, skills, per-project setup, and honest looks at the ecosystem around it.
Articles
Less code, more tests: Ponytail teaches Claude Code to stop writing what you don't need
Build with Claude Code without hundreds of extra lines: install Ponytail, walk its ladder of questions, the pre-release review and the author's benchmark.
Job Hunting With Claude Code: a Tailored Resume and Cover Letter for Every Posting in One Command (Setup From Scratch)
A resume and cover letter for one specific job in a single command: install ai-job-search, fill in your profile, paste a link, get PDFs a second agent checked.
The agent was barred from purchases, logins and personal data, but not from forms. Why did Claude send a tip to the police?
What does an AI agent do when a task can't be done? Claude sent a made-up tip to the police and used other people's tokens. Inside Anthropic's Oct 9 report.
722 papers from a closed OpenAI model on math. How many did a machine actually check?
Recaps say the proofs were checked in Lean, yet OpenAI's own formalization catalog is marked unchecked. We compared the October 6 release with what AGMAI asked.
Opting out of training doesn't hide your chat from a reviewer. What can a human at Anthropic see in Claude, and when does a conversation go to the police?
Does Anthropic read your Claude chats? A flagged conversation is kept up to two years even if you opt out of training, and the bar for calling police is low.
4 million lines of code and not a single LangChain import. How are Claude Code, Codex and nine other agents actually built?
Summaries call it an agent comparison, but it ranks nothing. Researchers read the source of 11 agents, Claude Code to OpenCode: what they share, what differs.
98.9% on the exam, 90.5% on questions the model never saw. Which number should you trust for your AI feature?
Imagine Claude building an exam for your support bot, then improving the prompt one change per round while hiding part of the questions. How it works.
Fable 5.1 as an advisor adds just 1.7 points to Opus 5.5 at twice the cost. So when does /advisor pay off?
Imagine Opus 5.5 writing the code while Fable 5.1 steps in at the hard parts. Anthropic measured +1.7 points for about 2.1x the cost. Here is where it pays.
People now play Doom inside Claude Code while the agent writes code. What else are mods allowed to do?
Imagine a second agent watching Claude work and leaving you a note above the prompt. What Claude Code mods can do, how to install one, and what you risk.
57 tokens per second instead of 30 on the same Mac. How does Magnitude beat llama.cpp?
Can a local model run your agent twice as fast with no new hardware? We break down Magnitude: kernels tuned to your chip, and where llama.cpp is still faster.
The skill that grew into Cloudflare's 7,245 bug findings. What will it find in your repository?
Can you trust Claude Code with a security audit? Cloudflare open-sourced a skill where a different agent checks every finding. Its phases, setup and gaps.
“AI solved motion graphic,” says X. The repo behind the Opus 5.5 music video shows five revisions and a client
X says AI solved motion graphics in one prompt. The Opus 5.5 video's code shows five revisions, a brief for scene agents and a colour bug found 4 days in.
Hindsight writes its own wiki about your repository. Does Claude Code need that kind of memory?
Teach Claude Code and Codex to remember why the project works the way it does: how Hindsight works, where its 94.6% comes from and which defaults to change.
The alarm went off after 12 minutes. The agent was stopped two and a half hours later. What actually broke in OpenAI's sandbox?
Does your containerized agent resolve outside domains? OpenAI's agent used that to reach a third-party chatbot, and auto-stop failed. Minute by minute.
GPT-6 Sol is half the price of Opus 5.5 per token. So why is a task only 20% cheaper?
On the price list GPT-6 Sol costs exactly half of Claude Opus 5.5. On a task in an independent test it is only 21% cheaper and 3 points weaker. Here is why.
Agents now get a boss, a task queue and a budget. Do you need one if you already work in Claude Code?
Five Claude Code and Codex windows in one task queue with roles and budgets: that is Paperclip. Where it helps, and where its spending cap kicks in late.
The model that will take 70% of the decisions away from ordinary LLMs. What is Jev?
Imagine a model that generates no text at all: you hand it state and typed questions, and you get values with probabilities back in a hundredth of a second. That is Jev, launched on 15 September. We go through the numbers from the source, where it fits inside an agent, and the nine places its own authors say it fails.
The Workshop is open: three tools that run in your browser
Imagine needing to crop a cover, strip the invisible characters a model left in your text, and work out what the harness around your agent costs. Normally that is three other people's websites and three uploads of your files to their servers. We built those three tools here, and not one of them sends anything anywhere.
Same model, same success rate, twice the money. What are you actually paying Claude Code for?
Imagine paying for a model twice: once for the intelligence, once for the way it is served to you. Berkeley ran seven models through three harnesses. On one pairing Claude Code came out three times more expensive and twelve points worse. We break down where that tax comes from, and where the study does not apply to your work.
One key instead of a dozen: what Monid does, and how many tools are really in there?
Imagine your agent no longer needing your account at every service: it finds a fitting endpoint itself, sees the price and the latency before calling, and pays only for what came back. That is Monid, released this week. We installed it, ran the tests, found a Windows bug and counted the catalogue by hand.
Anthropic raised Claude Code's weekly limits by 25% and cut them by 17% on the same day. How are both numbers true?
Imagine being told you got a raise, then hitting the ceiling earlier than usual the following week. That is what happened to Claude Code on 14 September. Here is why both figures are honest, what actually changed, what did not, and where your weekly allowance really goes.
10,000 agents did in 88 hours what nobody managed in 90 years. Why did their creators immediately ask for the brakes?
Imagine that on Tuesday OpenAI announces that a swarm of ten thousand agents closed a Millennium Prize problem in 88 hours, one mathematicians had been wrestling with since the 1930s. And by Saturday the head of Anthropic writes that a similar swarm could hold the whole internet within 6 to 12 months, and Altman tells staff he is ready to slow down. Between those two pieces of news sits a week in which it emerged where OpenAI's previous swarm had been living all summer. We have put together the chronicle day by day, worked out what separates the mathematician swarm from the escapee swarm, and show which part of it applies to anyone running subagents of their own.
Don't choose between Claude Fable 5.1 and GPT-6 Astra until you've answered one question
Imagine Anthropic's smartest model shipping on a Monday, and OpenAI answering on Wednesday with its own and declaring the AGI era. Both cost $10 per million input tokens and $50 per million output, to the cent. Then the differences start: one spends its tokens working out what is going on, the other spends them getting things done. We pulled together the benchmarks, the subscription prices, an independent robot-arm test and what both companies admitted about themselves, so you can pick a model for your work rather than for a headline.
What happens when you cross Claude Code with ChatGPT?
Imagine Claude Code becoming the architect of your tasks, with a Chrome extension that gives it access to other AIs: ChatGPT, for one, which makes some of the best images around. You describe what you need, and the agent walks into the chat, generates, downloads and files the result into your project. We tried it on a series of eight covers. Here is what came out.
Archify: the agent that maps your project's architecture into verified diagrams
One of the best skills for Claude Code, Cursor and Codex: the agent studies your project, finds the services and links, draws architecture, workflow and sequence diagrams and runs nine checks on them. A lifesaver for vibe-coded projects where nobody remembers what lives where. Free, MIT.
geo-seo-claude: audit your site for ChatGPT and Perplexity
A Claude Code skill that checks if your site is ready for AI answers: crawler access, llms.txt, passage citability and Schema. One command, one report.
caveman: a third off your tokens by having the agent read less
A project whose motto tells you everything: "why many token when few token also trick". What it actually cuts, what the benchmark shows, and why the licence matters more here than the savings figures.
claude-mem: so you stop explaining your project every morning
A plugin that remembers what you did in the last session and feeds the relevant part into the next one. How it is built, what it installs on your machine, and who it actually saves time for.
The Claude Code ecosystem: what to take and what to skip
There are now more catalogues and templates around Claude Code than there is time to check them. Here is which repositories actually save hours, and which are just long lists of links.