| name | wave |
| requires_protocol | >=0.6.0,<0.7.0 |
| description | Use when the user invokes `$wave`, asks to start or enter a multi-agent wave, or asks to spin up / set up a wave for a batch of tasks coordinated across agents. |
Team Collab Wave
This is a Codex-friendly command wrapper for the multi-agent wave entry point. Codex does not currently support arbitrary custom top-level slash commands like /wave, so the supported Codex entry point is:
$wave <slug-or-goal>
What this does
A wave is a batch of multi-agent work coordinated entirely through committed files. $wave is the human→coordinator shortcut for starting or entering one, so you don't have to type the full multi-agent enable / multi-agent init CLI flags by hand.
Required behavior
- Follow the installed
team-collab-protocol skill and its references/multi-agent.md flow. If the protocol skill is not available in the current session, read ~/.codex/skills/team-collab-protocol/SKILL.md; if that file is missing, stop and tell the user to run team-collab install-skills --agent codex --force.
- Before running any CLI command, check
command -v team-collab. If missing, stop and tell the user to install the CLI with npm install -g @embodot/collab@latest, then refresh skills with team-collab install-skills --agent codex --force if needed.
- Treat the text after
$wave as a slug only when it is a single token matching ^[a-z][a-z0-9-]{2,63}$. Otherwise, treat it as a goal, derive a slug, and confirm the slug with the user before creating anything.
- Ensure the project has multi-agent mode enabled. If it does not, run:
team-collab multi-agent enable
- If the wave folder does not yet exist, create it:
team-collab multi-agent init --slug <slug>
If <slug-or-goal> was a free-text goal rather than a slug, read the generated multi-agent/<slug>/PRD.md and pr-plan.md and draft a first cut of the wave plan (goal, task breakdown, suggested roles) into them, then tell the user what you drafted and ask for confirmation.
- If the wave folder already exists, just orient: read
multi-agent/<slug>/PRD.md, pr-plan.md, and any existing agent-runs/**, then summarize the current wave state to the user.
- Follow
references/multi-agent.md for the wave layout, ownership, and command boundaries. Do not improvise task assignments from this wrapper alone — present the plan and let the user, coordinator, or CLI task lifecycle commands drive the next step.
Compatibility note
If a future Codex version supports custom slash commands and /wave reaches the model as plain user text, handle it exactly the same way as $wave.
