X-MAN CLI

Run and remotely manage AI workers powered by Claude Code, Codex, and Gemini.

X-MAN keeps agent execution on your computer. You define workers with an agent, kit, skill, model, and permission level; attach local projects; then run tasks from the terminal or the X-MAN dashboard.

Requirements

• Node.js 20.10 or newer
• Git
• At least one supported agent CLI: claude, codex, or gemini
• Your own account/sign-in for the agent you use

Install

npm install --global @coderodilov/xman
xman --help
xman about

Quick start

# Check installed agents and their sign-in state.
xman agents

# Attach a local knowledge kit.
xman kit add ~/kits/flutter-agent-kit --link --as flutter

# Create a worker and attach a project.
xman worker new olivia --kit flutter --skill flutter-tester --agent codex
xman project attach ~/Projects/my-app
xman worker assign olivia --project my-app

# Run locally.
xman run olivia "Review the authentication flow"

X-MAN includes six professional agent kits, with ten skills in each: Backend, Frontend, Testing, Flutter, Research and Orchestration. Demo includes exactly backend-dev, frontend-dev and tester. The other 57 skills require an active Pro account, including all Research and Orchestration skills.

xman kit catalog list
xman kit catalog install backend --into ./my-api
xman kit catalog install frontend --into ./my-web
xman kit catalog install testing --into ./my-project
xman kit catalog export backend --output ./backend-kit

Install and export include only skills available to your account. Connect with xman login for Pro access. Unconnected bundled-kit use defaults to the three Demo skills; standalone user-owned local kits preserve the existing CLI behavior. Connected custom kits require Pro. Skill files are readable instructions; XMAN enforces subscription access at worker creation, execution and connected installation.

install uses the regular drift protection and receipt. export creates an editable copy and refuses to overwrite an existing destination. The old kit sample command remains a hidden compatibility alias. Existing sample worker references remain readable on Pro, but new workers should select the professional names.

See the kit catalogue for the skill list and content provenance.

Run xman without arguments for interactive mode. On TTY startup it briefly checks the canonical npm registry. If a newer release exists, choose 1 Update and continue or 2 Continue without update; explicit commands and non-interactive scripts never perform this check. Set XMAN_DISABLE_UPDATE_CHECK=1 when a controlled offline environment must skip it. Slash commands such as /help, /about, /worker, /project, and /dashboard are available in the session.

Dashboard connection

Connect the current computer to your X-MAN account:

xman login
xman dashboard

xman login opens a browser approval flow and starts the background daemon. The dashboard can then send tasks to agents running on this computer.

Manual mode is the default. Open the interactive X-MAN terminal and choose the persistent device permission with /permission:

xman
# then select /permission → Auto or Manual

You can also run xman permission auto or xman permission manual directly. Auto runs dashboard tasks without per-task approval; Manual keeps write-capable tasks waiting for approval on the device.

You can inspect and approve waiting local jobs with:

xman pending
xman approve <command-id>

Disconnect and revoke the device tokens with xman logout.

Demo limits

A Demo account supports one connected device, one office, one project, and one worker. Teams and room customization are unavailable on Demo. The server remains authoritative even when the CLI performs an early limit check.

Core concepts

| Concept | Meaning | |---|---| | Agent | A supported local LLM CLI: Claude Code, Codex, or Gemini | | Kit | A knowledge bundle containing skills, rules, and workflow guidance | | Skill | One specialist instruction set from a kit | | Worker | An agent + kit + skill + model + permission configuration | | Project | A local directory attached to X-MAN |

Permissions

| Level | Codex | Claude Code | Gemini | |---|---|---|---| | read-only | --sandbox read-only | --permission-mode plan | --approval-mode plan | | write | --sandbox workspace-write | --permission-mode acceptEdits | --approval-mode auto_edit | | dangerous | danger-full-access | bypassPermissions | yolo |

dangerous is never stored as a worker default. It must be requested explicitly with --dangerous. Parallel team execution remains read-only.

Before a local write task, X-MAN checks the Git working tree and creates a recovery branch. It stops if the working tree is dirty or the safety branch cannot be created.

Privacy and data flow

• Project files, absolute project paths, agent credentials, and provider sign-ins stay on the connected computer.
• In connected mode, account authentication, device metadata, worker/project configuration, task text, status, and bounded run metadata travel through the configured X-MAN API.
• Login requests result synchronization by default, and the device approval screen shows it before pairing. Logging in again or selecting Auto/Manual permission enables result synchronization automatically; no separate command is needed. The explicit advanced opt-out remains available with xman login --content off.
• Ctrl+C exits the interactive CLI, including text and selection prompts. It cancels an active direct worker task before leaving. Esc cancels a choice without closing the CLI.
• Auto permission is granted locally from /permission and can be changed back to Manual at any time.
• Sign-in and project attachment sync automatically. xman sync remains available as an optional manual refresh.

Useful commands

xman agents [install|login|probe]
xman kit [add|list|show|install|uninstall|status]
xman worker [new|list|show|edit|assign|suggest|rm]
xman project [list|new|attach|use|info|scan|rm]
xman run <worker> "<task>"
xman team run <team> "<task>"
xman inbox
xman runs
xman usage
xman office
xman handoff
xman sync
xman daemon [start|stop|status]
xman login | dashboard | logout
xman lang [uz|ru|en]
xman about

English is the default language. Uzbek and Russian are available through xman lang or the --lang flag.

License

MIT © Abdugaffor Odilov