HeadsUpAI

Anthropic updates Claude Code to launch autonomous background agents via fork

Anthropic

Anthropic updated the /fork command in Claude Code, its terminal-based agentic coding tool (AI that autonomously writes and iterates on code), to launch autonomous background agents. These sub-agents inherit the exact context of the active session. The background agent executes tasks independently and returns results to the main session.
Inherited Context
System prompt, tools, history, model, and prompt cache
Manual Fork Behavior
Copies transcript to a new user-driven session

This update streamlines parallel workflows by utilizing prompt caching (storing frequently used data to reduce latency) to make delegation nearly instantaneous. By allowing agents to fork with full state, Anthropic addresses context gaps that occur during task delegation. This follows a similar pattern seen when OpenAI introduced OpenAI parallel subagents.

Developers can now use /fork to offload sub-tasks, like debugging or running tests, while continuing work in the primary session. Manual forking is now available via the /branch command. These background tasks can be monitored through the agent view dashboard.

View the full update on x.com
ClaudeDevs
ClaudeDevs
@ClaudeDevs
X

We've updated /fork in Claude Code /fork now runs a background agent with your exact context (system prompt, tools, history, model) and prompt cache. The result gets returned to your session. /branch (the old /fork) still copies the transcript to a new session you drive. https://t.co/eMyFrk8PfR

215retweets3.6klikes
View on X

Still wondering? A few quick answers below.

The updated /fork command in Claude Code allows developers to launch an autonomous background agent that inherits the full context of the current session. This agent works independently on a delegated task and returns the results directly to the primary session, enabling parallel workflows without requiring the user to manually manage multiple terminal windows.

In the updated Claude Code, /fork now launches an autonomous background agent that inherits the session state and returns results automatically. The previous behavior, which copies the conversation transcript to a new session for the user to drive manually, has been renamed to the /branch command to distinguish it from autonomous background tasks.

A forked agent inherits the exact state of the primary session, including the system prompt, available tools, conversation history, and the specific model being used. Crucially, it also inherits the prompt cache, which allows the background agent to begin working immediately without the latency or cost of reprocessing the entire history.

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

Share this update