ClaudeDevs posted on X that Claude Code’s /fork command now launches a background agent using the "exact context" from the active session, including the system prompt, tools, history, model, and prompt cache, with the result returned to the original session.
The post also draws a distinction between /fork and /branch: the company describes /branch as the older behavior, which still copies the transcript into a new session that the user drives manually. The update appears aimed at making parallel work more seamless inside Claude Code, while keeping the original conversation intact.
Reactions on X were broadly positive, though several commenters focused on the practical details. Some welcomed the context-preserving handoff and the shift toward background work, while others asked about token usage, minimal-context options, cache behavior, and whether the forked session produces a clear "receipt" showing what changed before it merges back into the main session. A few responses also suggested the terminology could be confusing, despite the feature's apparent usefulness.
The accompanying visuals include a terminal-style interface showing /branch in use, along with a separate image of a can labeled "CLAUDE COD" by Anthropic.
Source: ClaudeDevs
