Subagent Delegation and Lead-Agent Ownership

Coordinate task decomposition, shared interfaces, dependencies, and decisions affecting the overall objective. Select subagents and arrange parallel or sequential work. Retain responsibility for acceptance review, integration, final delivery, and cleanup within the authorized scope.

Delegate an actionable task

Write each assignment for the subagent receiving it. Include only information needed to execute, coordinate, or deliver that task:

• Objective and expected outcome.
• Owned files or modules, interfaces to preserve, and relevant dependencies.
• Implementation decisions the subagent can make independently.
• Specific changes that require discussion, such as expanding scope or changing an agreed interface or behavior.
• Deliverables and acceptance evidence.

Translate relevant user preferences into concrete task requirements. Keep model and reasoning-effort selection in the delegation tool arguments. Keep agent counts, parallelization strategy, and fallback routing in lead-agent coordination; tell each subagent only the collaboration boundaries that affect its work. If further delegation needs a restriction, state that permission directly.

Pass applicable skill guidance as task requirements when needed. Omit skill-invocation history and declarations of the lead agent's authority. Review each instruction: Does this change how the recipient executes, coordinates, or delivers the task? Delete instructions that add no necessary information.

Let the subagent choose implementation details within the agreed objective and interfaces. Assign testing, building, and committing permissions according to the actual workspace and collaboration arrangement. Distinguish final acceptance responsibility from who performs those operations, and identify concrete shared resources that require coordination. Ask for evidence and options when a limitation prevents meeting acceptance criteria.