---

name: julien-skill-creator description: "Create or update Claude Code skills. Use when: user wants to create a skill, add a new skill, update skill, or mentions SKILL.md. Includes Skill Chaining documentation." license: Apache-2.0 metadata: author: "Julien (based on Anthropic skill-creator)" version: "2.0.0" category: "development" triggers:

• "SKILL.md"
• "skill.md"
• "create skill"
• "new skill"
• "add skill"
• "build skill"
• "skill template"
• "write skill"
• "créer skill"
• "nouvelle skill"
• "ajouter skill"
• "construire skill"
• "écrire skill"
• "créer une skill"
• "faire une skill"
• "développer une skill"
• "create a new skill"
• "build custom skill"
• "skill development"

---

Skill Creator

Enhanced guidance for creating effective skills with proper structure and documentation.

Observability

First: At the start of execution, display:

Copy🔧 Skill "julien-skill-creator" activated

Core Principles

Concise is Key

Context window is shared. Only add what Claude doesn't already know. Challenge each piece: "Does this justify its token cost?"

Progressive Disclosure (3 Levels)

1. Metadata (name + description) - Always in context (~100 words)
2. SKILL.md body - When skill triggers (< 500 lines)
3. Bundled resources - As needed by Claude

Skill Structure

Copyskill-name/
├── SKILL.md (required, < 500 lines)
│   ├── YAML frontmatter (name, description required)
│   └── Markdown instructions
└── Bundled Resources (optional)
    ├── scripts/      - Executable code
    ├── references/   - Documentation loaded on-demand
    └── assets/       - Output resources (templates, images)

Key rules:

• References must be one level deep from SKILL.md
• Files > 100 lines need table of contents at top
• Use forward slashes only (not Windows backslashes)
• No extraneous files (README.md, CHANGELOG.md, etc.)

Naming & Description

See references/naming-conventions.md for full guidelines.

Quick rules:

• Name: max 64 chars, lowercase + hyphens, gerund form preferred (processing-pdfs)
• Description: max 1024 chars, third person, include "what + when to use"

Skill Creation Process

Step 1: Understand with Examples

Ask clarifying questions:

• "What functionality should this skill support?"
• "Give examples of how it would be used"
• "What should trigger this skill?"

Step 2: Choose Destination

ALWAYS ask the user where to place the skill:

Questions to ask:

• "Cette skill est-elle spécifique à ce projet ou réutilisable ailleurs?"
• "Voulez-vous la partager via le marketplace?"

Naming convention by destination:

• Local: simple name (my-project-deploy)
• Marketplace: prefixed (julien-{category}-{name})

Step 3: Plan Reusable Contents

For each example, identify:

• Scripts needed (deterministic/repeated code)
• References needed (schemas, docs, policies)
• Assets needed (templates, images)

Step 4: Initialize Skill

For Local (project):

Copymkdir -p .claude/skills/<skill-name>
# Or copy template
cp -r ~/.claude/skills/julien-skill-creator/assets/template-skill/ .claude/skills/<skill-name>/

For Marketplace:

Copy# Use marketplace path
MARKETPLACE="$HOME/OneDrive/Coding/_Projets de code/2025.11 Claude Code MarketPlace"
mkdir -p "$MARKETPLACE/skills/<skill-name>"
# Or use init script
python "$MARKETPLACE/scripts/init_skill.py" <skill-name> --path "$MARKETPLACE/skills/"

Step 5: Edit the Skill

YAML Frontmatter

Required Fields:

Copy---
name: skill-name                      # Required: max 64 chars, lowercase + hyphens
description: What it does and when to use it  # Required: max 1024 chars, third person
---

Optional Fields:

See references/yaml-frontmatter-guide.md for complete documentation on all optional fields.

Quick Reference - When to Use Each Field:

Note: triggers is our marketplace extension, not official Anthropic. Official approach uses keywords in description.

Body Instructions

Write using imperative form (verb-first). Reference bundled resources clearly.

Document Skill Chaining (Critical)

Add Skill Chaining section at end. See references/skill-chaining-template.md for complete template.

Required subsections:

1. Skills Required Before
2. Input Expected
3. Output Produced
4. Compatible Skills After
5. Called By
6. Tools Used
7. Visual Workflow
8. Usage Example

Step 6: Add Activation Announcement (REQUIRED)

Every skill MUST announce its activation for observability.

Add this instruction at the beginning of the skill's execution section:

Copy## Observability

**First**: At the start of execution, display:

🔧 Skill "your-skill-name" activated

Copy

This:

• Confirms which skill is running
• Provides feedback to users
• Contourns Issue #4084 (hooks can't display in UI)

Step 7: Package Skill

Copyscripts/package_skill.py <path/to/skill-folder>

Validates then creates .skill file.

Step 8: Add Semantic Routing Triggers

Add triggers to YAML frontmatter for automatic skill suggestion:

Copy---
name: your-skill-name
description: What it does. Use when [specific triggers].
triggers:
  - "excel"
  - "créer un fichier excel"
  - "ajouter des formules"
---

Trigger Writing Methodology

Golden Rule: Write triggers as users naturally speak, not technical jargon.

3 Categories of Triggers (include all 3):

1. Keywords (1-2 words): Core concepts users mention

   • excel, pdf, docker, skill

2. Action Phrases (3-5 words): What users want to DO

   • créer un fichier excel
   • traduire ces sous-titres
   • déployer sur le serveur

3. Problem Phrases: How users describe their PROBLEM

   • mon site ne marche plus
   • je veux automatiser
   • comment faire pour...

Bilingual Coverage:

• Always include French AND English variants
• French first (primary user language), then English

Quantity Guidelines:

• Minimum: 5 triggers
• Optimal: 10-20 triggers
• Maximum: 50 triggers (diminishing returns)

Test Your Triggers:

Copypython scripts/benchmark-semantic-router.py "votre phrase test"

Step 9: Iterate with Self-Assessment

Never fix everything at once. Each iteration focuses on ONE improvement.

1. Score against quality rubric (see references/quality-rubric.md)
2. Identify top 3-5 gaps
3. Fix ONE dimension per iteration
4. Re-score until average ≥ 3.5/5

Reference Files

Degrees of Freedom

Match specificity to task fragility:

What NOT to Include

• README.md, INSTALLATION_GUIDE.md, CHANGELOG.md
• User-facing documentation
• Setup/testing procedures
• Any auxiliary context about creation process

Security & Portability

Security Checklist

• ✓ No hardcoded credentials (passwords, API keys, tokens)
• ✓ Sensitive data in .env file
• ✓ .env added to .gitignore
• ✓ Instructions use environment variables ($VAR or source .env)
• ✓ Use allowed-tools to restrict tool access for sensitive operations
• ✓ Set disable-model-invocation: true for admin-only/destructive skills
• ✓ Add version field for hot-reload tracking (Claude Code v2.1.0+)

Portability Checklist

• ✓ Use relative paths (.env, ./scripts/) instead of absolute paths
• ✓ Use ~/.claude/skills/ for user paths, not hardcoded home directories
• ✓ Avoid platform-specific paths (e.g., C:\Users\... in examples)
• ✓ Use forward slashes only (not Windows backslashes)

Anti-Patterns to Avoid

❌ Description >200 chars (breaks progressive disclosure) ❌ Content >300 lines (too much detail) ❌ Duplicates existing docs (should reference instead) ❌ Development notes in content (timestamps, changelog) ❌ Generic/vague description (doesn't explain when to use) ❌ Second-person language ("you should...") ❌ Information in both SKILL.md and references files ❌ Hardcoded credentials (passwords, API keys, database credentials, tokens) ❌ Sensitive data in code examples (use placeholders or .env variables) ❌ Absolute paths in examples (/home/user/..., C:\Users\...) ❌ Platform-specific paths (Windows-only or Unix-only examples)

Skill Chaining

Skills Required Before

• None (entry point skill for skill creation)

Input Expected

• User request to create/improve a skill
• Domain knowledge or examples from user
• Optional: existing skill to improve

Output Produced

• Format: Complete skill directory with SKILL.md + resources
• Side effects: Creates files in specified path
• Duration: 10-30 minutes depending on complexity

Compatible Skills After

Recommandés:

• julien-dev-tools-skill-reviewer: Review and score the created skill

Called By

• Direct user invocation: "Create a skill for X"
• Skill improvement workflows

Tools Used

• Read (usage: read existing skills, templates)
• Write (usage: create SKILL.md, references)
• Bash (usage: run init_skill.py, package_skill.py)
• Edit (usage: modify existing skills)

Visual Workflow

CopyUser: "Create skill for X"
    ↓
[THIS SKILL]
    ├─► Step 1: Gather examples
    ├─► Step 2: Plan resources
    ├─► Step 3: Initialize structure
    ├─► Step 4: Write SKILL.md + refs
    ├─► Step 5: Package
    ├─► Step 6: Add semantic triggers
    └─► Step 7: Iterate
    ↓
skill-name.skill created
    ↓
[Optional] skill-reviewer for quality check

Usage Example

Scenario: Create deployment skill for VPS

Command: "Create a skill for deploying to my Hostinger VPS"

Result:

• hostinger-deployment/SKILL.md created
• hostinger-deployment/references/ with server docs
• hostinger-deployment/scripts/ with deploy scripts
• Packaged as hostinger-deployment.skill