What problem does it solve?
When delegating work to subagents, tasks often fail due to polluted context, unclear boundaries, missing acceptance criteria, or unverified output. This Skill defines a disciplined orchestration protocol so subagents receive only the context they need and their results are reviewed before integration.
Core Features & Use Cases
- Dispatch Standards: Each subagent receives its task from tasks.md, relevant constraints from the plan package (requirements.md, plan.md, or prd/ files with decisions.md), verification commands, and DESIGN.md excerpts when UI work is involved.
- Fresh Context Isolation: Subagents start without main-session history, handle a single responsibility, and return only results, evidence, or blockers without wrapping output in outer HelloAGENTS formatting or writing turn-state.
- Two-Stage Review: The main agent performs a requirements-compliance review followed by a code-quality review with verification commands before proceeding to the next task.
- Use Case: When parallelizing a feature build across multiple subagents, the main agent assigns non-overlapping files, tracks progress as a controller, and handles BLOCKED reports by adding context, upgrading the model, or splitting the task.
Quick Start
Use the hello-subagent skill to split this feature into independent tasks and dispatch them to subagents with clear acceptance criteria.
