
Tech • AI • Robotics • Game
Claude Code Projects is a new beta workflow that lets Claude coordinate long-running software tasks by splitting a goal into parallel threads, opening pull requests, tracking reviews, and running recurring checks across repositories.
Claude Code Projects centers on a single ongoing conversation where a user defines a goal, adds relevant repositories or folders, and lets Claude coordinate the work. Instead of manually creating sub-agents, sessions, or worktrees, the system breaks a large objective into smaller tasks and assigns them to separate threads.
Each thread functions as an independent session running on its own branch in the cloud, allowing work to continue even when the user is offline. That makes the feature suited to large goals spanning many sessions or multiple repositories, such as removing a deprecated package or updating an endpoint across API, web, and mobile codebases.
When a task needs access to local files or developer tools, a thread can be run on the user’s computer instead of entirely in the cloud. This hybrid model is designed to handle both remote coordination and tasks that depend on the local development environment.
One example use case focuses on investigating a run of website performance issues and establishing a daily safeguard against future regressions. After receiving metrics and user feedback, Claude explores the codebase, proposes performance improvements, and then launches a separate thread for each agreed task.
When a thread completes a concrete change such as a bug fix, it can open a pull request automatically or on request. The same thread continues monitoring that PR, pushing fixes if CI fails and addressing review comments, while the main conversation surfaces the next action needed.
Users can open any thread to watch its progress, give additional direction, or stop it. A dedicated threads view shows work that is waiting for user input, and an overview tab lists all pull requests and their current status before merge decisions are made.
Projects also include artifacts, which can be used to compile outputs such as a summary of code changes and before-and-after performance metrics. These artifacts are stored in a shared library alongside uploaded files such as mockups or design specifications, and every thread can access that material in later work.
The system can connect to a monitoring service to inspect real-user metrics directly. It can then be instructed to run a daily routine that checks for performance regressions and, if one is detected, automatically opens a new investigative thread for review.
Because every thread is a full Claude Code session and multiple threads can run at once, projects can consume plan limits faster than a single session. To manage that usage, project settings allow changes to the model and effort level, with low effort recommended for the coordinator and smaller models suggested for tightly scoped thread work.
Users can define how Claude should operate, including how many threads to run simultaneously, how often to provide updates, and whether to present a plan before starting. Threads also share memory, allowing project-wide preferences to persist and be edited in settings for future work.
Claude Code Projects is currently in beta and is rolling out gradually to Pro and Max plans. The feature is positioned for software work that is too broad for a single session and needs persistent, coordinated execution over time.
Claude Code Projects extends coding assistance from a single chat into a coordinated workflow for multi-step engineering work. Its main significance is combining task decomposition, parallel execution, PR handling, and recurring monitoring inside one persistent project structure.
Ask a question