groundcrew CLI
groundcrew hands Linear and Jira tasks to local Claude Code, Codex and Cursor agents, each in its own worktree and sandbox
- 68 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Launches Claude Code, Codex and Cursor agents that write code and commits in your repositories without you
- --runner none gives agents your credentials with no sandbox
- Uses a Linear key and repository access
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/groundcrewDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Launches Claude Code, Codex and Cursor agents that write code and commits in your repositories without you
- --runner none gives agents your credentials with no sandbox
- Uses a Linear key and repository access
This is third-party code. Review the repository files before installing.
What it does
groundcrew watches tasks assigned to you, creates a separate git worktree for each, launches the agent CLI in a dedicated terminal pane and leaves the result on its own PR-ready branch. Agents run on your machine with your tools and credentials. The real claude, codex, cursor-agent or pi launches, so you can watch any session and take over. Every agent is sandboxed by default: Safehouse on macOS or Docker Sandboxes, and none turns protection off explicitly. The default task source is Linear, where you assign a task to yourself and add an agent-claude, agent-codex, agent-pi or agent-any label. agent-any picks the agent with the lowest session-weighted load. Jira and local files plug in through task sources. Presets exist for claude, codex, cursor, cursor-grok and pi, and you add your own CLI in config. Install: npm install -g @clipboard-health/groundcrew@latest, then crew init --global --project-dir ~/dev --repo OWNER/REPO --agent claude, check with crew doctor and start crew run --watch.
Who it is for. Developers and team leads with a backlog in Linear or Jira who want local agents to work through it in parallel and overnight.
Good fit when
- Linear tasks are labeled agent-* and need to become branches with code while you sleep
- Several agents must run in parallel without stepping on each other, each in its own worktree
- You want control: open any running agent session and step in when it drifts
- Agents must be isolated by a Safehouse or Docker Sandboxes sandbox on the host
- Load has to be spread between Claude Code, Codex and Cursor by session limits
Not a fit when
- You need remote cloud agents: groundcrew launches local CLIs on your machine
- You have no tmux, cmux or zellij and no sandbox runner: without them you must run with --runner none
- There is one small task: running the agent by hand on a branch is enough
- Your Node is older than 24.22.1: the README requires that version
Example request
Set up groundcrew for the acme/web repository with the claude agent, run crew doctor and show how to queue a taskLimitations
- Node 24.22.1, git, a terminal multiplexer (tmux, cmux or zellij), an agent CLI and a sandbox runner are required, and crew doctor checks them all.
- --runner none runs agents without isolation on the host with your credentials.
- Linear needs the GROUNDCREW_LINEAR_API_KEY key, and Jira and files have their own sources.
- Pi needs authorization with the selected model's provider before unattended use.
- tmux kill-server ends every tmux session, so save work and stop tasks before it.
How to disable. Stop a task with crew stop TASK and the worktree stays. crew cleanup TASK and crew cleanup --all remove worktrees. The package comes off with npm uninstall -g @clipboard-health/groundcrew, and config lives in ~/.config/groundcrew/.
FAQ
How is groundcrew different from cloud coding agents?
Agents run on your machine with your tools and credentials instead of in someone else's cloud. You can watch each session in a terminal and take over, and the result sits on its own branch for a pull request.
How do I queue a task?
In Linear assign the task to yourself and add an agent-claude, agent-codex, agent-pi or agent-any label. crew run ignores tasks without an agent-* label, but crew start TASK starts one by hand.
What does the sandbox do?
Safehouse on macOS or Docker Sandboxes isolates each agent on the host. --runner none turns it off, after which the agent runs with your rights and no isolation.
Does it work without Linear?
Yes. Jira and local files connect through task sources, described in docs/task-sources.md.
Which agents does groundcrew support?
Presets are built in for claude, codex, cursor (plus a cursor-grok variant) and pi. You add your own CLI in config, and agent profiles set model tiers.
Discussion
No questions or reviews yet.
Related
CLI toolsGeckobox runs Claude Code, Codex and other CLI agents on your computers and in sandboxes and tracks the jobs in a local dashboard
A self-hosted control plane for running several agents: task dispatch, run review, spend tracking, connected via MCP, CLI or REST
A Go server that runs Claude Code or Codex on your machine or Kubernetes and delivers the work as a pull request after review
Orchestrator for Claude Code, Codex and Gemini CLI: YAML workflows that loop until your own quality gates pass
The @itsaplan/runner CLI takes tasks from the It's a Plan tracker and hands them to Claude Code, Codex or opencode locally.
A CLI that loops Claude Code, Codex, OpenCode and other agents over a task list or PRD until everything is done
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| groundcrew (this card) | no | 68 | |
| pi-dispatch | no | 197 | pi-dispatch launches Pi agents, while groundcrew takes tasks from Linear and Jira and routes them to Claude Code, Codex, Cursor and Pi |
| Crewship | no | 5 | crewship is another tool with a similar job, while groundcrew keeps agents local with one worktree per task |
README in short
The README explains why local agents, lists the prerequisites and gives a six-step quick start. It then describes picking tasks by Linear labels, agent-any routing and the crew commands: stop, cleanup, stage and others. Details are moved to docs.