fable-team

Multi-agent orchestration prevents context explosion by enforcing strict role separation between orchestrator, planner, and workers, while applying physical guardrails like 300k/450k token hard gates and automatic state externalization to maintain context safety during long-running pipelines.

Strict role separation in multi-agent pipelines assigns distinct models and effort levels to orchestrator, planner, implementer, tester, checker, and DA roles, ensuring uncoordinated AI workflows are eliminated and each agent handles only planning, implementation, testing, or adversarial review.

Set up a multi-agent team by triggering the guided boot sequence, which configures your team roles and executes the pipeline with enforced guardrails, handling planning, implementation, testing, and adversarial review automatically.

Multi-agent orchestration supports CI/CD environments through configurable headless override policies, integrating with baton for worktree navigation and cairn for project ledger management while maintaining enforced guardrails.

Context safety uses write-through state externalization to a `.fable-team/state/` directory, combined with automatic hygiene routines and hard token gates, preventing context loss during long-running multi-agent software development pipelines.

Avoid multi-agent orchestration when tasks lack the complexity requiring planning, implementation, testing, and adversarial review phases, or when model-specific spawn paths and physical guardrails against context explosion introduce unnecessary overhead for simple workflows.