Agent Team Work Zone: An Automated, Persistent Workspace for Long-Lived Claude Code Agent Teams | AI Paper Digest

TL;DR Highlight

Saving agent team state in files lets a Claude Code team be restored with one command.

Who Should Read

A developer running Claude Code Agent Teams who loses team context when the terminal closes or compaction blurs details

Core Mechanics

Session end kills the teammates too
- /resume restores only the team lead, not the in-process teammates.
- Working context held only in a teammate's context window disappears as well.
The core rule is files over context
- Roles, decisions, tasks, and commitments live in per-agent workstation directories.
- It rebuilds rather than preserves: a fresh teammate reads the files and resumes.
A hook enforces automatic checkpoints
- A TeammateIdle hook checks the last checkpoint time and blocks idle at 15 minutes or more until the teammate saves.
- Unlike compaction, a checkpoint only reads context and writes a file, so it works mid-task.
One /reactivate-team command revives the team
- It reads the registry (TEAMMATE_INFO.json) and spawns each teammate with an instruction to read its own workstation.
- Success is judged only by the teammate sending a message in the current session.

Evidence

No benchmarks or user studies
- The paper states it reports no benchmark or user-study results.
Checkpoint threshold of 15 minutes
- Under 15 minutes since the last checkpoint, idle proceeds; at or above it, the hook blocks idle and demands a save.
Recommended team size of 3 to 5
- Each teammate runs its own session, so token cost grows with team size and coordination cost grows faster than linearly.
Release history
- Two consecutive releases shipped without migration scripts and broke the upgrade chain, so release gates were added.
- v0.1.0 is the final release for CC 2.1.177 and earlier, and v0.3.2 is current as of July 2026.

How to Apply

Have Claude Code write checkpoint files first
- Paste the codeExample prompt into one long-running task today to create workstation notes and a checkpoint.
- Close the session and confirm a new session can continue from those files alone.
Give each teammate its own workstation directory
- Split role, current task, decisions, and open commitments into per-teammate files.
- Make a rule that teammates never edit each other's files and only send messages.
Write agent reports as prompt packets
- In handoffs, include what was done, what was tried, changed files, next needs, file locations, and why the task exists.
- This cuts the work of rewriting long prompts.
Install ATWZ in a small project
- Follow the public repo's setup and run /spawn-team and /reactivate-team.
- Check first that your Claude Code version is 2.1.178 or later.

Try It

copy & run

# Paste into Claude Code as is (simplified version of the paper's method: file-based checkpoints)

너는 장기 작업을 맡은 에이전트다. 컨텍스트는 사라질 수 있으니 중요한 상태는 반드시 파일에 남긴다.

1. 프로젝트에 workstation/{{에이전트이름}}/ 디렉터리를 만든다.
2. 아래 파일을 만든다: README.md(역할), checkpoint.md, TODO.md, commitments.md
3. checkpoint.md에는 현재 목표, 진행 중인 작업, 내린 결정, 열린 질문, 약속, 핵심 파일 경로를 쓴다.
4. 작업이 한 단계 끝날 때마다 checkpoint.md를 갱신한다. 최근 몇 턴은 자세히, 오래된 내용은 요약한다.

작업: {{지금 맡길 작업}}

Next step: close the session, then in a new session say "Read only workstation/{{에이전트이름}}/ and continue the work."

Terminology

Agent TeamsAn experimental Claude Code feature where one lead hands work to several Claude Code instances. Each teammate has its own context window.

CompactionCondensing a long conversation into a summary to fit the context window. It is fast but can blur roles and commitments.

WorkstationA persistent directory owned by one agent. Think of it as a human employee's personal notes drawer.

TeammateIdle hookA script that runs when a teammate is about to go idle. If the checkpoint is stale, it blocks idle and asks for a save.

ReactivationRebuilding a team by spawning fresh teammates from the saved registry and checkpoints. It does not revive dead processes.

Related Papers

Related Resources

https://github.com/SR-A-W/agent-team-work-zone

Original Abstract (Expand)

Large Language Model (LLM) agents have significantly improved coding and programming workflows. Claude Code, in particular, is one of the most powerful LLM coding agents and is capable of conducting complex coding tasks. However, several drawbacks can undermine long-term agentic workflows. (1) Irrecoverable agent teams: The Agent Teams feature is powerful, but the working state accumulated by each teammate is lost and cannot be resumed once the process stops, for example, when a terminal is closed. (2) Compaction erodes working detail: Compaction condenses the conversation into a summary, causing an agent's working details to become vague. (3) Agentic "technical debt": Over time, a user's decisions and the agents' operations become trapped in compacted old chats, making the project increasingly difficult to maintain and review. (4) Heavy prompt writing: Assigning or handing off tasks requires users to repeatedly write long prompts to achieve the expected agentic performance. We propose ATWZ (Agent Team Work Zone), a filesystem-based operations layer built around Claude Code's native Agent Teams that addresses these problems. Its central design principle is to treat each agent and teammate as a human employee and preserve their important working state in files stored in a dedicated directory called a "workstation," together with the skills, hooks, and scripts that use and maintain these files. With ATWZ, an agent team can periodically back up its working state, allowing an agent's knowledge to be recovered after compaction. After a process ends, the team can be restored with a single command. These features also substantially mitigate the agentic "technical debt" described above. Moreover, within ATWZ, agent "employees" can send documents to one another, greatly reducing the effort required to write prompts.