| name | yolobox-orchestrator |
| description | Use when the agent is outside yolobox and needs to start, inspect, or control yolobox sessions, choose the right yolobox subcommand or flags, read merged config, decide between scratch, readonly, docker, or network options, or debug how a yolobox launch should be configured. |
| license | MIT |
| compatibility | Requires the yolobox CLI on the host. Runtime behavior depends on the configured container runtime and local Docker, Podman, or Apple container access. |
Yolobox Orchestrator
Use this skill for host-side orchestration of yolobox sessions.
Do not use it for questions about the current environment from inside a running box. Inside the container, use yolobox instead.
- Start from the user's intent, then choose the smallest
yolobox command or flag set that accomplishes it.
- Check
yolobox config when defaults, merged config, or flag precedence matter.
- If
default_harness is set, bare yolobox launches that shortcut; use yolobox shell for an explicit shell.
- Prefer explicit isolation and safety flags:
--scratch for disposable or concurrent sessions that must not share /home/yolo.
--readonly-project when the agent only needs read access to the project tree.
--no-env-passthrough when host API/token environment variables should not enter the box automatically.
--open-bridge only when the agent needs to open HTTP(S) URLs in the host browser.
--docker only when the agent needs Docker access or sibling containers.
- When you need exact command patterns or edge-case reminders, read references/commands.md.
- If you launch a box for another agent, point it at
yolobox and YOLOBOX_CONTEXT_FILE for inside-the-box introspection.
- When discussing concurrency, distinguish isolated per-run manifests from shared persistent state: manifests are per-run, but
/home/yolo and /var/cache are shared unless --scratch is used.