Claude Code Projects Adds Coordinator, /btw and /fork

Anthropic released a significant update to Claude Code Projects on November 14, 2026, transforming the feature from a static file repository into a coordinated system of parallel working sessions. The update introduces a coordinator that pushes status updates and answers questions in the main project chat while multiple worker threads execute tasks on separate branches. The rollout is live for a narrow subset of Pro and Max plan subscribers, with Team and Enterprise access still pending.
How the coordinator architecture works
Each Claude Code Project now operates as a coordinator plus an array of worker threads. The coordinator scopes incoming requests, delegates work across threads, reviews outputs, and assembles results. Each worker thread is a full cloud Claude Code session running on its own branch and copy of the repository, and can spawn subagents, loops, or workflows to divide its assignment further.
Anthropic's product documentation, published November 14, 2026, clarifies that when two threads modify overlapping code, the conflict surfaces as a standard merge conflict in a pull request. The system does not resolve these automatically; teams handle them using the same workflow they would apply to conflicts between human developers.
New commands: /btw and /fork
The update adds two slash commands for thread management. The /btw command spawns a side thread for tangential work, allowing users to branch a question or task without interrupting the main conversation flow. The /fork command creates alternate implementations of the same feature, running multiple approaches in parallel so teams can compare results before choosing a direction.
Projects also gain read-only thread links that honor local-file isolation, and a reminders feature that temporarily hides threads from the main view. Anthropic says progress can be monitored and steered from a mobile device, with work continuing after the user leaves their desktop.
Shared infrastructure and usage controls
Two shared components hold the architecture together. Projects maintain a shared memory that all threads write to and read from, reducing the need for complex prompt engineering. A library collects added files and thread-generated artifacts, preventing output from scattering across disconnected chats.
Users configure how often Claude checks in, how frequently it opens new threads, and how detailed its updates should be. Anthropic warns that projects can reach usage limits faster than single-session work because each spawned thread is a full Claude Code session burning plan capacity in parallel. The system provides project-specific usage tracking and lets users set model and effort levels for both the coordinator and worker threads independently.
Release date and rollout status
Officially launched on November 14, 2026, the feature is available only to a deliberate subset of Pro and Max subscribers who do not have existing projects. Users with legacy projects continue using the previous folder-based interface without changes. Threads run exclusively in the cloud; local execution on user machines is described as coming very soon but is not yet available.
Team and Enterprise plan access remains pending. Anthropic's product post explicitly states that the beta opened to a narrow user group, and the source article confirms no hands-on testing was performed outside Anthropic's own infrastructure.
What this means
The coordinator model shifts Claude Code Projects from a context store for single conversations into a system that delegates, executes, and assembles work autonomously. The feature is most valuable when handed large, multi-threaded goals, which is also when it consumes plan capacity fastest. Teams should start with bounded tasks like single endpoint migrations or targeted latency optimizations, monitor project usage after the first run, and review every generated pull request before merging. The merge-conflict handling signals that Anthropic treats parallel agents as parallel collaborators, not infallible automators, and expects human judgment at the review step.
on Emergent today





