trae-skill-writer（特雷技能作家）

Trae Skill Writer

Analyze project code AND business context, then delegate to skill-creator for SKILL.md creation.

Phase 1: Understand Project (REQUIRED)

Before creating ANY skill, you MUST understand the project first.

1.1 Check Project Size

If project is too large (>50 top-level items, monorepo):

• STOP - Don't analyze entire project
• ASK - Use AskUserQuestion for target folders
• SCOPE - Only analyze user-specified folders

"This project is large. Which folders should I focus on?"
Options: "src/features/", "lib/core/", "Other..."

1.2 Scan Project Structure

Quick scan to understand what exists (NOT deep reading):

1. List top-level directories
2. Identify tech stack (package.json, go.mod, etc.)
3. Find existing skills/rules/automation
4. List major domains/modules for skill breakdown

Note: Deep code reading happens in Phase 2 for each specific skill.

1.3 Understand Business

Ask user and read docs to capture business context:

Ask: "What business problem does this solve?"

Phase 2: Create Skills (SEQUENTIAL)

Create skills ONE at a time, not in parallel.

2.1 Plan Skill Breakdown

First, identify all skills needed:

Example: ecommerce-app
├── app-style      (style system)
├── app-component  (UI components)
├── app-api        (API patterns)
└── app-monitor    (monitoring)

2.2 For EACH Skill (Sequential Loop)

┌─────────────────────────────────────────────┐
│  For each skill:                            │
│                                             │
│  0. Initialize SKILL.md from template       │
│     - Run scripts/init_skill.py             │
│     - Edit generated scaffold               │
│                                             │
│  1. Deep-dive into THIS skill's code        │
│     - Read 2-5 key files for this domain    │
│     - Extract patterns specific to it       │
│                                             │
│  2. Design spec for THIS skill              │
│     - Name, triggers, exclusions            │
│     - Key files, workflow                   │
│     - Business context                      │
│                                             │
│  3. Delegate to skill-creator               │
│     - Use template below                    │
│     - Wait for completion                   │
│                                             │
│  4. Move to next skill                      │
└─────────────────────────────────────────────┘

Initialization command:

python {skill_dir}/scripts/init_skill.py \
  --skill-dir {skill_dir} \
  --name {new_skill_name} \
  --summary "One-line skill purpose" \
  --output-root {project_root}/skills

2.3 Delegation Template

Use skill `skill-creator` to create the skill with this spec:

**Skill Name:** {prefix}-{domain}
**Purpose:** {what it does}
**Triggers:** {phrases that should activate it}
**Exclusions:** {when NOT to use}
**Language:** English (or specify)
**Key Files:** {list of files - relative paths only}
**Workflow:** {numbered steps}
**Location:** .trae/skills/{name}/

Project context:
- Tech stack: {detected tech}
- Patterns: {from actual code}

Business context:
- Domain concepts: {terminology}
- Business rules: {constraints}

Phase 3: Quality & Lessons Learned

⚠️ Common Mistakes (CRITICAL)

These mistakes break skills. Always check:

Path Rule: Always use paths relative to project root. Just src/file.ts, never /absolute/path/src/file.ts.

Quality Checklist

Before delegating each skill:

• Paths - All relative, no absolute paths
• Naming - {prefix}-{domain} format
• Language - Single language throughout
• Code study - Based on actual codebase
• Business - Includes domain terminology
• Granularity - Single responsibility

Best Practices

Naming: {prefix}-{domain}

Single Responsibility

Each skill = ONE domain:

• Good: app-style (only style)
• Bad: app-style-and-component (too broad)

Cross-Reference

After creating all skills:

• Check for overlap
• Add "Related Skills" sections
• Verify no duplicate content

Agents

References

• Trae Skills Documentation
• Best Practices