skill-writer - SKILL.md Agent Skill

name: skill-writer description: Create reusable, tool-agnostic Agent Skills with clear scope, triggers, and safe tool usage. Use when designing or refining any Skill regardless of project domain.

Skill Writer

This Skill helps you design clear, reusable, and tool-agnostic Agent Skills for any project domain.

When to use this Skill

Use this Skill when:

Creating a new Skill from scratch
Converting a specialized Skill into a general Skill
Reviewing a Skill for clarity, scope, and activation triggers
Standardizing Skill structure and frontmatter fields

Core Principles

Single purpose: one Skill solves one type of problem
Explicit triggers: make activation conditions easy to detect
Safe tool usage: prefer least-privilege tool access
Reusability: avoid project-specific hardcoding unless required

Instructions

Step 1: Define scope and triggers

Define the user intent this Skill should cover
List the exact phrases or scenarios that should trigger it
List exclusions that should not trigger it

Step 2: Choose Skill location

Project Skill: .trae/skills/<skill-name>/SKILL.md
Personal Skill: ~/.trae/skills/<skill-name>/SKILL.md

Step 3: Create a minimal structure

<skill-name>/
├── SKILL.md
└── reference.md (optional)

Step 4: Write SKILL.md frontmatter

---
name: skill-name
description: Clear, tool-agnostic description of capability and triggers
allowed-tools: Read, Grep, Glob
---

Step 5: Write the Skill content

Use consistent sections:

# Skill Name

## Quick Start
## Instructions
## Examples
## Constraints
## Validation Checklist

Step 6: Add examples and edge cases

At least one positive activation example
At least one negative example
One edge case where the Skill should refuse or ask for clarification

Step 7: Validate the Skill

Frontmatter name matches directory
Description is specific and tool-agnostic
Steps are actionable and ordered
Examples are realistic and concise

Example Skill

---
name: issue-triage
description: Triage a bug report into reproduction steps, root cause hypotheses, and next actions. Use when user says “bug”, “error”, or “issue”.
allowed-tools: Read, Grep, Glob
---

# Issue Triage

## Quick Start
1. Extract error message and stack trace
2. Identify suspected modules
3. Propose next diagnostic steps

## Examples
- “App crashes on login with 500 error”
- “Payment flow throws NullPointerException”

## Constraints
- Do not modify production data

Validation Checklist

Frontmatter is valid YAML
Trigger phrases are explicit
Tool access is minimal
Examples include positive and negative cases