agent-teams

You coordinate parallel work by spawning synchronized agent teams across backend and frontend layers using canonical spawn prompts. Teammates share a task list and communicate via direct messaging to accelerate full-stack feature development and avoid bottlenecks.

Yes, you can run parallel code reviews by spawning an agent team with specific ownership roles like reviewer. Teammates operate concurrently with shared task visibility, enabling simultaneous investigation and feedback across the codebase.

The best way to debug competing hypotheses is spawning an agent team where teammates investigate different theories in parallel. This approach uses per-team ownership roles and shared task lists to coordinate findings without creating investigative bottlenecks.

Yes, experimental enablement requires configuring settings.json to activate parallel team execution. The Skill provides safety rails and display modes, including in-process or split-pane options, directly through this configuration file.

Full-stack agent teams manage ownership through explicit per-team roles like backend-implementer, frontend-implementer, schema-owner, and reviewer. Guided spawn prompts delegate tasks while a shared task list tracks progress and direct messaging handles collaboration.

You should not use parallel teammate sessions for simple, single-layer tasks that lack cross-domain dependencies. The overhead of spawning and synchronizing teams is best suited for complex full-stack features and parallel investigations requiring coordinated effort.