What problem does it solve?
Delegating work to sub-agents often wastes tokens on the wrong model, pollutes the main context window, and returns unverified output. This Skill analyzes each task, picks the right model and specialized agent, and enforces structured reasoning plus self-verification before results are accepted.
Core Features & Use Cases
- Automatic Model Selection: Classifies task complexity and output size to choose between Opus, Sonnet, or Haiku using an explicit decision tree.
- Specialized Agent Matching: Detects domain matches (developer, researcher, architect, tech-writer, and more) and injects the corresponding agent instructions into the sub-agent prompt.
- Zero-shot CoT + Self-Critique: Wraps every dispatched task with a mandatory step-by-step reasoning prefix and a five-question verification loop that blocks incomplete submissions.
- Use Case: Run
/launch-sub-agent Design a caching strategy for our API that handles 10k requests/second and the Skill selects Opus with the software-architect agent, dispatches an isolated sub-agent, and returns a self-verified design.
Quick Start
Ask the agent to launch a sub-agent for your task, for example: launch a sub-agent to implement pagination for the /users endpoint following the patterns in /products.
