The Orchestrator Skill

The WTmag skill is what turns your agent into an orchestrator. It's a single Markdown file — skill/SKILL.md in the WTmag repo — that teaches the agent how to understand your intent and dispatch workers. Without it, your agent is just an agent. With it, it becomes the thing that spawns other agents.

What the skill does

The skill tells the agent:

• How to spawn workers with wtmag create
• What the worker is supposed to do for each task type (issues go to PR, PRs are read-only review, custom prompts do what you said)
• How to resolve the default agent and placement from config
• When to use --agent and --model overrides (only when you explicitly ask)
• How to report back after dispatching a worker (worker ID + how to attach)
• Not to attach to workers itself — that's your job

The orchestrator's job is logistics: fetch context, create worktrees, dispatch workers, tell you how to jump in. The judgment calls — when to attach, what to review, when to clean up — are yours.

How to load it

When you start your orchestrator session, pass the skill to your agent:

tmux new -s project-dev
pi --skill ./skill

The skill file lives in the WTmag repo at skill/SKILL.md. You can also install it globally so any agent in any project can use it:

mkdir -p ~/.config/opencode/skills/wtmag
curl -fsSL https://raw.githubusercontent.com/obiMadu/wtmag/main/skill/SKILL.md \
  -o ~/.config/opencode/skills/wtmag/SKILL.md

What a conversation looks like

The skill lets you talk naturally. You don't have to construct commands — the orchestrator translates intent into wtmag create calls.

You can:

• Work through a rough problem and define it into a task
• Reference an existing ticket by number
• Ask it to pull open issues from whatever sources it's connected to
• Write the task up from scratch
• Ask for a specific agent or model for a particular run

The orchestrator fetches the full context and dispatches workers accordingly — one worker per task, each in its own worktree and tmux session.

What the orchestrator won't do

• It won't attach to workers on your behalf. It tells you the worker ID and how to jump in, then steps back.
• It won't recommend or install a different agent on its own. It uses the configured default unless you explicitly ask for something else.
• It won't use --model overrides unless you explicitly ask for a specific model or thinking level.
• It won't change placement unless you explicitly ask for -s, -w, or --launch.

The skill is designed to keep you in control. The orchestrator handles the logistics, you handle the work.