using-specialists

Delegate complex code review and debugging tasks by routing substantial work to specialist agents with explicit tracking, scope, and review boundaries instead of performing investigation or implementation directly. You create a tracking bead to serve as the task prompt and preserve outputs across execution cycles.

A bead-first workflow for AI delegation creates a tracking bead upfront, uses it as the task prompt, and preserves outputs across explorer, executor, reviewer, and fix cycles. This prevents the orchestrator from wasting time on discovery and maintains context throughout the chain.

Use proactive delegation for architecture analysis and refactoring when you have substantial multi-file tasks. You decide to hand off work to specialist agents before doing the investigation yourself, ensuring complex operations run with explicit scope and review-fix loops.

Worktree-aware execution supports isolated environments by running edit-capable specialists in isolated worktrees. It allows reusing job workspaces when needed and keeps delegation chains alive through review and repair cycles without manual git operations.

Merge-gated publication rules coordinate specialist outputs by using chain and epic lifecycle rules to coordinate merges, verify readiness, and publish through specialist commands. This orchestrates the final merge without requiring manual git operations from the user.