CLAUDE CODE NEW FORKED SUBAGENTS

@agentic.james

19.9K followers · usually 1.3K views a post

Views

8.9K

Likes

323

Comments

17

Shares

66

Selling · SaaS › AI Tool

Claude Codeseen in 193 videos →https://www.skool.com/agent-architects ↗

Why it worked

The video provides a concise explanation of a new, potentially valuable feature for users of Claude Code, demonstrating its practical application and benefits. The use of on-screen text and clear spoken explanations makes the information easily digestible.

Summary

This video explains the new 'forked subagents' feature in Claude Code, which allows subagents to retain the full conversation history of the session that deployed them. This feature helps save tokens by avoiding the need to re-feed context and allows for more accurate evaluations from multiple perspectives.

Structure

1Introduction to forked subagents
2Explanation of how they differ from normal subagents
3How to enable and use the feature
4Use cases and benefits

Product placement

The video discusses and demonstrates the 'forked subagents' feature within Claude Code, showing a snippet of the settings.json file and the '/fork' command.

Call to action

Save for later and follow for more Claude Code tips.

Original caption

Claude Code forked subagents changed how I deploy subagents. A normal subagent boots fresh with just a summary, a forked one carries the entire conversation history of the session that spawned it, so you can evaluate one decision from four different perspectives without re-feeding any context. It can actually save tokens too since nothing gets reread.

More from @agentic.james

Their other posts in the index, biggest breakout first.

On-screen text

CLAUDE CODE NEW FORKED SUBAGENTS settings.json M "env": { "CLAUDE_CODE_FORK_SUBAGENT": "1" } /fork

Transcript

Claude Code just introduced forked subagents. Here is everything you need to know to use them. Save for later and follow for more Claude Code tips. Forked subagents essentially allow your subagents to have the entire context of the main session that deployed them. Normal subagents are deployed with a fully fresh context and are sent a summary of the previous session as well as instructions. When a fork session gets deployed with the full conversation history of the session that deployed it, as well as those instructions. Depending on what you're doing, this can actually save tokens with deploying subagents because your subagents don't have to reread all their system prompts and all of their basic tools as it's already the previous sessions context. To use the new fork subagent feature, add this to your global settings.json file. You can invoke this by doing slash fork. I think the obvious best use case when you wanna evaluate something from A, B, C, and D perspective, you need all the past conversation to do so accurately. So maybe you're making four different UI or you're evaluating a PR from four different perspectives, or anything like that.