Guide · Agent skill
How to use the Claude DevFleet Orchestration agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The claude-devfleet skill teaches Claude to act as a manager for several Claude Code coding agents working at the same time. It turns one large coding goal into a chain of missions, starts each agent in its own isolated git worktree, tracks progress, and reads structured reports on what changed. This guide explains what the skill does, how to install it, how to trigger it with plain requests, and what you need running before it works.
Claude DevFleet Orchestrationaffaan-m/ECC · MIT Download the skill Quick start
- 1Start Claude DevFleet. Run a Claude DevFleet server locally so it is reachable at http://localhost:18801/mcp.
- 2Connect it to Claude Code. Run claude mcp add devfleet --transport http http://localhost:18801/mcp to add the server over MCP.
- 3Download and unzip the skill. Download the claude-devfleet ZIP from this page and unzip it so SKILL.md is in a claude-devfleet folder.
- 4Place the skill folder. Move it to ~/.claude/skills/ for personal use or to .claude/skills/ in a project. In the Claude apps, upload the ZIP under Settings → Capabilities → Skills instead.
- 5Ask for a multi-agent task. Describe a large coding goal and ask for parallel agents using DevFleet.
- 6Approve the plan and dispatch. Review the mission list Claude shows you, then approve so the first mission is dispatched.
- 7Read the reports. Use get_mission_status to monitor progress and get_report to review files changed, errors and next steps.
What is an agent skill and what does this one do?
An agent skill is a SKILL.md file that gives Claude extra instructions for a specific kind of task. When the task matches the skill's description, Claude follows the steps written in it.
The claude-devfleet skill covers multi-agent coding work. Its workflow is: Claude plans a project with plan_project, shows you the mission plan and waits for your approval, dispatches the first mission, and then dispatches dependent missions as earlier ones finish. Each agent runs in a separate git worktree with full tool access, so agents do not clash with each other. When a mission completes, the source describes it being merged automatically and the next dependent mission starting. Finally, Claude reads the report with get_report, which lists files_changed, what_done, errors and next_steps.
What you need before installing the skill
The skill does not run on its own. It only describes how to call the tools of a Claude DevFleet server, so that server must be running locally and connected to Claude Code over MCP. The source uses this endpoint: http://localhost:18801/mcp
Add the server to Claude Code with the command from the source. Claude Code must support MCP for this to work.
- A Claude DevFleet server running locally at http://localhost:18801/mcp
- Claude Code with MCP support
- The MCP connection added with the claude mcp add command
claude mcp add devfleet --transport http http://localhost:18801/mcpInstall the claude-devfleet skill
Download the skill ZIP from this page and unzip it. The folder contains SKILL.md. Put the folder in one of these locations, depending on how you want to use it:
For a personal setup available across projects, place it in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder.
In the Claude apps you can instead upload the ZIP under Settings → Capabilities → Skills.
- Personal: ~/.claude/skills/
- Project only: .claude/skills/ inside the project
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
Trigger the skill with natural requests
You do not need a special command. Describe a large coding task that could be split across agents and mention that you want parallel agents or DevFleet. The skill is meant for work where several Claude Code agents need to run at once on separate parts of a goal.
The source's own example request is to build a REST API with authentication and tests. A request like that can lead Claude to plan the project, show the mission list, and wait for your go-ahead.
- Build a REST API with authentication and tests, split across parallel agents using DevFleet.
- Plan a project for adding a settings page, then show me the missions before dispatching anything.
- Check the status of the missions in my current DevFleet project and tell me what has finished.
Use the main tools step by step
The skill exposes a set of DevFleet tools. plan_project turns a description into a project and mission chain. create_project and create_mission let you build the plan by hand, with depends_on listing mission IDs and auto_dispatch starting a mission once its dependencies are met. dispatch_mission starts an agent on a mission, and cancel_mission stops a running one.
To monitor work, get_mission_status checks progress without blocking, while wait_for_mission blocks until a mission finishes. get_dashboard gives a system overview of running agents and statistics. list_projects and list_missions let you browse what exists. Use get_report to read the outcome of a finished mission.
- plan_project(prompt): split a description into a mission chain
- create_project(name, path?, description?): create a project manually
- create_mission(project_id, title, prompt, depends_on?, auto_dispatch?): add a mission
- dispatch_mission(mission_id, model?, max_turns?): start an agent
- get_mission_status(mission_id): check progress without blocking
- get_report(mission_id): read the structured report
- cancel_mission(mission_id): stop a running agent
Review the plan before agents start coding
The workflow in the source asks Claude to show you the mission plan and get approval before dispatching. Read the missions, check their order and dependencies, and make sure each one is scoped to a clear part of the goal.
After dispatch, check the report for each mission. The source flags unattended use as a risk because completed missions are merged automatically, so a bad change can reach your branch if you do not look at the reports.
- Confirm the mission order matches the dependencies you expect.
- Ask for a narrower mission if a prompt covers too much.
- Read get_report output, especially errors and next_steps, before moving on.
What people use it for
- Splitting a feature such as a REST API with authentication and tests into dependent missions.
- Running several independent parts of a goal in parallel, each in its own git worktree.
- Checking which missions are still running or have finished on a DevFleet dashboard.
- Reviewing structured reports of what each agent changed before deciding what to keep.
Frequently asked questions
Is the claude-devfleet skill free?
The skill is published under the MIT licence, as stated on this page. Claude DevFleet itself and any costs of running Claude Code are separate and are not covered by the source.
How do I install the claude-devfleet skill?
Download the ZIP from this page and unzip it. Place the folder in ~/.claude/skills/ for personal use or in a project's .claude/skills/ folder, or upload the ZIP under Settings → Capabilities → Skills in the Claude apps.
Does the skill work without Claude DevFleet?
No. The skill only describes how to call the tools of a running Claude DevFleet server. Without that server connected over MCP at http://localhost:18801/mcp, the tools will not exist.
What command connects DevFleet to Claude Code?
Use claude mcp add devfleet --transport http http://localhost:18801/mcp. This adds the server over HTTP as an MCP connection.
Does the skill run scripts or use the network?
The source states that the skill runs no scripts and uses the network. It calls the DevFleet server over MCP, so that server must be reachable.
When should I use a single agent instead?
For small single-file edits, one agent is simpler and cheaper. Use DevFleet when a large feature can be split into parts that run in parallel or in a dependency chain.
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 this skill if you already run Claude DevFleet and want to split a large feature across parallel agents, each in an isolated worktree. Pick something else if you do not have a DevFleet server, if the task is a small edit, or if you want unattended changes on an important repository, since completed missions are merged automatically and reports need review.
Download the skill