Articles · Topic
Claude Code
17 articles
-
Claude Code ultrathink now runs at xhigh effort, and subagent effort has its own setting ultrathink in the terminal and VS Code now runs its turn at xhigh effort, and CLAUDE_CODE_SUBAGENT_EFFORT sets what subagents run at. The week also brings a write guard for paths outside your working directories, idle compaction in VS Code and the Desktop Code tab, and the new Ultra plan. -
Tutorial How Claude Code Handles Memory About Your Project You correct Claude on the same thing for the second time and wonder whether it will remember it this time. Claude Code now has two layers of persistent memory: rules that you write, and notes that it keeps on its own. Both are plain Markdown files on your disk. That is the biggest advantage of the whole approach and also its weakness, because a note in a file only helps if Claude remembers to open it. How do you work with this so that you end up satisfied? - Claude Code Code Review gets review levels, REVIEW.md includes and a stop command Code Review now has Quick, Standard and Deep levels, REVIEW.md can include other files, and `@claude review stop` ends an always-on subscription. The week also brings model and effort controls in agent view, gateway deny-lists and sandbox changes.
-
Tutorial Fewer Subagents, or When Too Many Agents Hurt in Claude... It's tempting. You split an agent into specialists, each team takes its own piece, and an orchestrator ties it all together at the top. Finally, dozens of agents are working for you! It looks clean, it grows along with the organization, and everyone knows what they're responsible for. Based on its experience with enterprise deployments, Anthropic claims that such a setup often performs worse, runs slower, and costs more than a single agent with Skills. So when should you merge subagents, when should you keep them, and how do you tell that merging didn't break something? -
Claude Code Claude Duet: The Assistant That Reaches Out on Its Own. Once Every Few Hours It's tempting, I'll admit. In the morning your assistant tells you what's on fire, during the day it watches your email, your calendar and your unpaid invoices, and it only pings you when it needs a decision from you. Anthropic has now released exactly that kind of agent (via a waitlist) into a field of several souped-up competitors. What does it promise? And what does it actually do? -
Claude Code Add-ons that promise to make Claude Code cheaper: I measured them on 84 runs Authors of Claude Code add-ons promise savings from 20 to 98%. I ran the five most recommended ones (Caveman, Ponytail, RTK, Context Mode and Headroom) against plain Claude Code on a real project. Only one saved money consistently, and a shorter CLAUDE.md helped more than any add-on. - Claude Code Idle compaction keeps long conversations cheap after a break Claude Code now compacts a 200K+ token terminal conversation after about 54 idle minutes, before the one-hour prompt cache expires. The week also tightens MCP allow rules and aligns subagent defaults across the terminal, SDK, desktop and VS Code.
- Claude Code Haiku 5.5 becomes the default Haiku, and sub-agents get their own effort and compaction window Claude Haiku 5.5 is now the default Haiku model on the Anthropic API, with a 1M context window. The week also brings sub-agent effort and compaction controls, fail-closed hooks, and a new default MCP protocol version.
-
Claude Code Mods for Claude Code: Anthropic Lets You Under the Hood What has always annoyed me about Claude Code is how it prints colorful diffs while it works, and I don't want to read them. But there was no way to hide them. Until now: Anthropic has introduced Mods, which let you properly rebuild the interface. -
Claude Code Claude Code Projects Now Split Up the Work on Their Own Anthropic has just launched redesigned Projects in Claude Code. A project is now a single long-running conversation in which Claude acts as a coordinator. It breaks the assignment into individual threads, each run as a separate session on its own branch, and they report their results back. -
Claude Code Anthropic's head of AI safety testing: We don't have a plan yet for aligning superintelligence At the beginning of September, Jacob Coxon announced he was leaving Anthropic. The name probably won't mean anything to you: he did pre-training research on models, earlier at OpenAI, most recently at Anthropic. He accompanied his resignation with a thread on X that took off immediately. In it, he claims that both OpenAI and Anthropic are heading toward self-improving superintelligence in a way he doesn't consider responsible. -
Claude Code An AI watermark proves touch, not authorship. What should we do about labeling AI content? On Monday, Anthropic confirmed that it embeds an invisible watermark in text generated by Claude. The hate toward the EU started immediately, blaming it for the move, but the EU requires nothing of the sort. Where are the limits of AI in text, and how do I use AI to write my own? -
Tutorial Claude Code rewrote /fork. For the second time this month. And this time, it was useful Anthropic has changed the /fork in Claude Code for the second time this month; it’s now a background agent with inherited context. This time, it’ll come in handy. How do you work with it? -
Tutorial Dynamic Workflows in Claude Code: An Announced Experiment Now Set to Be Put to the Test Anthropic has introduced a new feature called Dynamic Workflows in Claude Code, alongside the Opus 4.8 model. Claude writes its own orchestration script, which launches tens to hundreds of parallel subagents in a single session, and checks the results itself before delivering them to you. -
Claude Code Claude Code Agent view—it’s a whole new way of working Anthropic is striving to demonstrate that swarms of agents are right at home in the Claude Code. To that end, it has introduced Agent View, which allows users to monitor the status of all agents on a single screen. However, this significantly changes the way we work. So what’s new and different? -
Claude Code Claude Code also offered /goal for automatically reaching a destination Anthropic has added the `/goal` command to Claude Code, which keeps running a task until it’s finished. Typically, it keeps correcting an error until it’s fixed. It’s like a Ralph Wiggum Loop for adults... - Tutorial Claude Code Hooks — The Complete Guide Hooks let you run shell commands before and after every Claude Code action. Here's how to use them to enforce linting, automate git, and build custom workflows that run silently in the background.