Claude Code can now ask you interactive questions when it needs more information or when there are multiple paths forward.

Claude Code SDK is crazy. For the last few months we’ve been slowly collecting all of our company’s data into a single place and now I’m hacking on an agent that can operate over that data.

Premium men’s jewelry that elevates every fit. Black Friday deals live now!

When building Claude Code our philosophy is to ship fast and learn, while at the same time keeping it intuitive and simple to use. Please keep the feedback coming — we love hearing how we can make Claude Code even better.

To recreate an output style that overrides a system prompt, use the following CLI flags: --system-prompt-file --system-prompt --append-system-prompt We've also seen good success using a dedicated subagent with your particular output style.

Around November 5th, your user-level output styles will be converted automatically to a plugin. You can recreate your own output styles by using hooks that append to the system prompt. See an example here: See an example here:

Always curious and full of energy, Beagles turn every walk into an adventure 🐾🎉.

Next week, we're replacing output styles in Claude Code. We found in practice that less than 1% of users actively used output styles, and they can be recreated through other extension points via plugins.

Claude Code Weekly Round Up This week we made subagents resumable, added a Plan subagent, introduced prompt based stop hooks, fixed several high priority bugs & much more across the CLI, web and SDK. We're also replacing output styles with with more powerful alternatives:

I hate San Francisco halloween, what do you mean you're a diverging loss function

We've been working on this pretty much since we first released Claude Code. It's been a long road, but we are now confident that our native install is better for 99% of users. If you're already using Claude Code, try `claude install`.

gm

if you go as Clawd for halloween, please tag me

For example, for customer support write scripts to access the Zendesk API and put your internal docs in markdown files. Write a Claude MD with instructions on how to triage, when to escalate, etc. Then ask Claude Code: "triage the 10 oldest open tickets and draft responses"

The easiest way to prototype a general agent is to use Claude Code. Write some scripts to access your APIs, write a custom Claude MD, then prompt Claude Code with some queries. Does it work? Use the Claude Agent SDK to productionize it.

okay a TSM comeback sponsored by Claude Code would go so hard

You can see how I’ve implemented all of this in my email agent demo using the Claude Agent SDK. Just merged this in today! Let me know if you have any questions or feedback, I'll do deep dive on this specific implementation later. https://t.co/s1gqbQC7Js

This adds a whole new dimension to your agent, now it's not just about prompts & tools, but also API design. There is still a lot to figure out in using codegen well for agents, but if you do I guarantee your agent will be lightyears ahead.

Adding Subagents What if the user wants to say something like: “If I get an email about my accounting or taxes”, this is a bit squishy, there’s probably not a simple rule in code that would detect this. For this, you can add subagents as a function call, for example:

This feels nice, but is very rigid. For example what if I wanted to say: "Forward emails from my boss to my personal email, but only on weekends" Code generation lets us handle this easily.

I want my agent to handle requests like: “Every day at 11am, email me the most urgent tasks in my inbox” “When I get an email from my boss, send me a notification” The most common way people do this is that they invent a JSON schema or DSL to represent every action like so

Every Agent is a Coding Agent The best way to make your agent more adaptable & powerful is using code generation. 🧵 Here’s how I’m using it for my open source email agent built on the Claude Agent SDK.

Boris & Cat are absolute legends, incredibly lucky to learn from them everyday.