Guide · Agent skill
How to use the Coordinate Qwen Code Teammates agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The coordinate skill turns Qwen Code into a team leader. You give it a goal, it splits the goal into separate investigation tasks, and it hands each task to a named teammate who works read-only. This guide explains what the skill does, how to install it, how to run it with /coordinate, and when a single agent is the better choice.
Coordinate Qwen Code TeammatesQwenLM/qwen-code · Apache-2.0 Download the skill Quick start
- 1Enable the agent team setting. Turn on experimental.agentTeam in Qwen Code and restart the application so the team tools load.
- 2Install the skill. Download the ZIP from this page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder, or upload it under Settings → Capabilities → Skills in the Claude apps.
- 3Confirm you have a Git repository. Open a Git repository if you want the single worktree writer to make code changes. Otherwise the team stays read-only.
- 4Type /coordinate with your goal. Enter /coordinate followed by a clear goal. Natural-language requests do not trigger the skill.
- 5Review the teammates' evidence. Watch the Agent View tabs for conversations, messages, status, and approvals, and check that each result has the evidence the task required.
- 6Review the writer's changes. If a writer was spawned, review and verify its worktree result before the leader integrates it into the current branch.
What the coordinate skill makes the agent do
The skill makes the main agent act as a team leader. The leader decomposes the goal, keeps task ownership clear, reconciles disagreements, and delivers the final result.
Investigation teammates are read-only. The runtime enforces this with a positive allowlist: they cannot use shell, edit or write files, save memory, create schedules, invoke arbitrary deferred tools, or spawn agents. They can look at code and report evidence. When changes are needed, one writer teammate works in a Git worktree that the leader pins, and the leader decides whether and how to integrate the result.
- Read-only teammates investigate separate workstreams and message each other through the shared task list and send_message.
- The leader accepts or rejects each result based on its evidence and reassigns rejected work.
- Only one writer is spawned, and only when the goal requires code changes.
- Existing Agent View tabs show teammate conversations, messages, status, and approvals.
Install the coordinate skill
The skill ships as a single file, SKILL.md. Download the ZIP from this page and unzip it into one of two locations. Use ~/.claude/skills/ to make the skill available across your projects, or use a project's .claude/skills/ folder to limit it to that project.
In the Claude apps you can instead upload the skill under Settings → Capabilities → Skills.
- Personal install: unzip into ~/.claude/skills/
- Project install: unzip into the project's .claude/skills/ folder
- Claude apps: Settings → Capabilities → Skills, then upload
Enable agent teams before you run it
The team tools require the experimental.agentTeam setting. Enable it and restart Qwen Code. Without this, the team tools are unavailable, and the skill tells you that experimental.agentTeam must be enabled and Qwen Code restarted. It then offers ordinary subagents as a fallback. Subagents report only to the leader and cannot collaborate as a team.
For the writer step you also need a Git repository. If the folder is not a Git repository, or worktree creation fails, all teammates stay read-only and the leader makes the final change in the current checkout.
- Qwen Code with experimental.agentTeam enabled, then restarted
- A Git repository if you want a worktree writer
Run the skill with /coordinate
The skill is invoked explicitly. It runs only when you type /coordinate, followed by your goal. Natural-language requests do not trigger it, so describe the goal in the command itself.
The leader creates one team with one self-contained task per investigation workstream. It prefers at most three read-only teammates and never exceeds the team cap that team_create reports. Each task needs an objective, scope, completion condition, and required evidence.
- Use one teammate for a narrow task.
- Add a fourth teammate only when the workstreams are truly independent, and never past the configured cap.
- Do not use Arena mode with this skill. The skill says Arena is for competing solutions to the same task, not collaboration on different tasks.
/coordinate <goal>How implementation works after investigation
When the goal needs code changes, the leader first finishes the parallel investigation and sends shutdown requests to the investigation teammates. It enters one worktree and keeps the returned path. It then creates the implementation task and spawns exactly one named writer, with working_dir set to that path and read_only not set. The writer receives the accepted evidence and must keep all changes inside the worktree.
The leader reviews and verifies the worktree result, then alone decides how to integrate it into the current branch. The worktree is not removed until accepted changes are integrated or deliberately discarded. After synthesis, the leader shuts down active teammates and deletes the team.
- Investigation finishes before any writer starts.
- Only one writer works in the worktree.
- The leader runs the smallest relevant verification once, after implementation is finished.
Safety notes: scripts and network
The skill does not run scripts and does not use the network, according to its listing. Its safety comes from the read-only allowlist for investigation teammates and from the leader's review of the writer's worktree before anything is merged.
Because the writer can edit files inside its worktree, review its changes before accepting them. The skill itself also tells the leader not to claim that in-process teammates are independent PTY sessions or heterogeneous CLIs.
- Scripts: none included
- Network: none used by the skill
- Licence: Apache-2.0
What people use it for
- Splitting a codebase audit into separate read-only investigations, such as one teammate per module, then having the leader merge their findings.
- Researching a bug from several angles at once, with one teammate tracing the code path and another checking the related tests, before a single fix.
- Running an investigation first and then a single implementation in an isolated worktree, with the leader reviewing the result before merging.
Frequently asked questions
How do I install the coordinate agent skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder for one project. In the Claude apps, you can upload it under Settings → Capabilities → Skills.
What do I need to run the coordinate skill?
You need Qwen Code with the experimental.agentTeam setting enabled and restarted. A Git repository is needed for the worktree writer; without one, all work stays read-only.
Is the coordinate skill free to use?
The skill is licensed under Apache-2.0, as listed on this page. Check the repository's licence file for the full terms.
Why doesn't my natural request start the coordinate skill?
The skill is set to run only when you explicitly type /coordinate. Use the command followed by your goal, for example /coordinate <goal>.
What should I do if the team tools are unavailable?
Enable experimental.agentTeam in Qwen Code and restart it. Until then, the skill can only offer ordinary subagents, which report to the leader and cannot collaborate with each other.
How does coordinate compare with Arena mode?
They solve different problems. Coordinate splits one goal into different tasks handled by a team, while the skill says Arena is for competing solutions to the same task, so it is not the right mode for this skill.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use coordinate if you work in Qwen Code and want several read-only investigations run in parallel, with the leader reviewing every result before any change is accepted. Skip it for simple, narrow tasks, where one agent is faster, or if you cannot enable experimental.agentTeam.
Download the skill