What problem does it solve?
When you use the delegate() function to spawn a child agent for a sub-task, the child has no access to your available skills index, load_skill tool, or parent chat history, leading to context-blind execution, wasted tool calls, and incorrect or incomplete results. This skill provides the official contract to avoid those pitfalls.
Core Features & Use Cases
- Full Context Inheritance Guide: Clearly documents what context the child agent inherits from the parent (tools, datetime grounding, pushed skills) and what it does not (skill catalog, chat history, workspace context) to set accurate expectations.
- Delegate Configuration Best Practices: Covers how to push relevant skills and surgical reference files to the child, set MCP server allowlists to scope access, configure budget envelopes (max steps, output tokens, delegation depth), and write effective handoff prompts with time anchors and output specifications.
- Decision Framework for Delegation: Provides clear criteria for when to delegate a sub-task versus calling a tool inline, preventing unnecessary delegation of simple one-off tasks.
- Return Contract Specification: Defines the standard {ok, answer | reason, toolsUsed} return envelope for child agent results to ensure consistent observability and error handling.
- Use Case Example: For a task like finding stale GitHub issues across multiple repositories, use this skill to push the GitHub MCP server and relevant GitHub skill to the child, specify the date cutoff and output JSON schema in the handoff, and set a max step budget to avoid runaway execution.
Quick Start
Use the delegate-handoff skill to configure a delegate() call for a sub-task that requires searching multiple GitHub repositories for stale issues, ensuring the child agent has the necessary GitHub MCP access and context to return structured results.
