Skip to content

Use Teams Mode

When Claude Code's experimental teams feature is available, Lineup can run agents as teammates instead of subagents. Each agent appears as a named tmux pane, giving you real-time visibility into what every agent is doing.

Prerequisites

  • Claude Code with teams support
  • The CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 environment variable set
  • The TeamCreate tool available in your session

Enable teams mode

Set the environment variable before launching Claude Code:

bash
export CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1
claude

Or set it inline:

bash
CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 claude

Then run the pipeline as usual:

/lineup:kick-off Add a caching layer to the API

The orchestrator detects TeamCreate availability and automatically creates a lineup team. No other configuration is needed.

What changes in teams mode

Agent visibility

In standard mode, subagents run in the background and you only see their output when they finish. In teams mode, each agent spawns as a named tmux pane -- you can watch the researcher scanning files, the architect writing the plan, and the developer implementing code, all in real time.

Prompt embedding

The teams runtime does not apply agent definition files (frontmatter + body) automatically when team_name is specified. To work around this, the orchestrator reads each agent's .md file and embeds the full instructions directly in the spawn prompt. This means agents behave identically in both modes -- the instructions are the same, only the delivery mechanism differs.

Tool restrictions

In standard subagent mode, the tool list from agent frontmatter is enforced by the platform. In teams mode, tool restrictions are advisory only -- a known platform limitation. Agents still follow their defined tool boundaries through their instructions, but the platform does not block disallowed tool calls.

No nesting

Teammates cannot spawn their own teammates. The platform blocks nested team creation, so the orchestrator does not attempt it.

Fallback behavior

If any of the prerequisites are missing -- the environment variable is unset, TeamCreate is unavailable, or team creation fails -- the pipeline falls back to standard subagent mode transparently. You do not need to change any configuration. The orchestrator logs which mode it selected during initialization.

Note: Claude Code limits each user to one active team at a time. If another session is already leading a team, the pipeline will notify you and continue in subagent mode. Both pipelines run normally -- only the team coordination UI is affected.

When to use teams mode

Teams mode is useful when you want to:

  • Monitor agent activity in real time during long-running pipelines
  • Debug pipeline behavior by watching what each agent reads, searches, and writes
  • Demo Lineup to others with visible, parallel agent activity

For most day-to-day use, standard subagent mode works identically. The pipeline output and quality are the same in both modes.