Xiaozhi Skill Creation Coach

The Core Problem

Directly reusing existing SKILLs rarely aligns with an individual's specific error patterns and preferences. This skill standardizes the SKILL creation process, helping users transform scattered materials (e.g., mistake collections, class notes, and exam papers) into specialized Agent prompt assets built on a Role, Rules, Memory, and Output architecture.

Core Capabilities and Key Steps

• Four-Layer Structure: Defines the agent's scope (Role), behavioral rules (Rules), security and memory fields (Memory), and response format (Output), preventing issues like overly broad roles or insufficient rules.
• Five-Step Workflow: Forms a closed loop from scenario definition and template filling to material feeding, real-world testing, and prompt iteration.
• Material Feeding and Best Practices: Supports structured feeding of mistakes, tutoring materials, notes, and exams. It emphasizes using activation phrases to switch context, feeding materials instead of demanding direct answers, and enforcing a "test 5 times, iterate once" cycle.
• Health Metrics and Misconception Diagnosis: Provides metrics to diagnose vague memory and lack of iteration, ensuring the SKILL system remains maintainable.

Scope and Caveats

This skill heavily relies on the accuracy and granularity of the fed materials. Security boundaries must be explicitly defined in the Memory layer. When collaborating across SKILLs, follow the principle of minimal sharing by passing only necessary summaries. New SKILLs should undergo 5-10 warm-up conversations to validate the four-layer structure before deep use.