Skill Creator

Overview

Maintains and grows the skill library by identifying opportunities for new skills and ensuring they follow proven patterns. Uses the docx skill as the gold standard for skill architecture.

IMPORTANT: ALWAYS PERFORM THESE STEPS

• VALIDATE NEED - ONLY CONTINUE TO NEXT STEP IF NEEDED. SEE SECTION Don't create skills for

• CHECK FOR EXISTING SKILLS - MOVE TO EDIT FLOW IF EXISTING SKILL EXISTS

• SCAFFOLD

• GATHER INTELLIGENCE

• CUSTOMIZE THE SKILL

Scaffold Script

Run the scaffold script to create the new skill:

For project skill

sh ~/.claude/skills/skill-doctor/scripts/scaffold.sh
--path .claude/skills
django-orm-optimization

This creates:

~/.claude/skills/django-orm-optimization/ ├── SKILL.md # Pre-filled template ├── scripts/ # Optional: for helper scripts └── templates/ # Optional: for config templates

For personal skill

sh ~/.claude/skills/skill-doctor/scripts/scaffold.sh
--path ~/.claude/skills
django-orm-optimization

When to Use This Skill

Automatic triggers:

• User explicitly asks to create/improve a skill

• Working on complex task that will likely recur

• Domain-specific patterns emerge (ORM optimization, K8s debugging, API design)

• User says "we should document this" or "let's make this reusable"

Judgment call triggers:

• Task has non-obvious gotchas

• Requires specific tool chains or workflows

• Involves institutional knowledge worth preserving

• Team will benefit from standardized approach

Don't create skills for:

• One-off tasks

• Trivial operations

• Well-documented standard practices

• Personal preferences without technical substance

Skill Quality Standards

Reference skill: Check .claude/skills/ or ask user for path to best existing skill.

Quality Principles

From the docx skill gold standard:

Be prescriptive, not descriptive

• BAD: "You might want to consider using X"

• GOOD: "Use X. Run command here "

Front-load decision logic

• Decision tree at top saves context window

• Reader knows immediately which path to follow

Show anti-patterns

• BAD vs GOOD examples

• Explain why the bad approach fails

Mandate reading when complex

• "MANDATORY - READ ENTIRE FILE: Read reference.md "

• Don't assume Claude will figure it out

Operational wisdom

• Batch sizes that work well

• Performance implications

• When things break (and how)

Concrete over abstract

• Actual bash commands

• Real code snippets

• Specific file paths

What to AVOID:

• Philosophical rambling about why things matter

• Excessive politeness ("please consider...")

• Hedging language ("maybe", "perhaps")

• Long prose without code

• Tutorial style - this is a runbook

Workflow for Creating a New Skill

Step 1: Validate Need

Ask the user:

I notice we're working on [DOMAIN]. This seems like good skill material because:

• [REASON 1: will recur / has gotchas / team benefit]
• [REASON 2]

Should we create a skill for this? Suggested name: [domain-specific-name] Location: .claude/skills/[name]/SKILL.md (project) or ~/.claude/skills/[name]/SKILL.md (personal)

Step 2: Check for Existing Skills

Before creating, verify no overlap:

Check project skills

ls -la .claude/skills/

Check personal skills

ls -la ~/.claude/skills/

Search descriptions

find .claude/skills ~/.claude/skills -name "SKILL.md" -exec grep -l "keyword" {} ; 2>/dev/null

Step 3: Scaffold the Skill

Run the scaffold script to create the new skill:

For project skill

sh ~/.claude/skills/skill-doctor/scripts/scaffold.sh
--path .claude/skills
django-orm-optimization

This creates:

~/.claude/skills/django-orm-optimization/ ├── SKILL.md # Pre-filled template ├── scripts/ # Optional: for helper scripts └── templates/ # Optional: for config templates

For personal skill

sh ~/.claude/skills/skill-doctor/scripts/scaffold.sh
--path ~/.claude/skills
django-orm-optimization

This creates:

.claude/skills/django-orm-optimization/ ├── SKILL.md # Pre-filled template ├── scripts/ # Optional: for helper scripts └── templates/ # Optional: for config templates

Step 4: Gather Intelligence

Extract from current work:

• What commands/tools are being used?

• What gotchas have we hit?

• What patterns emerged?

• What would we want to remember next time?

Step 5: Customize the skill

Then edit SKILL.md to fill in:

• Description (YAML frontmatter)

• Domain-specific workflows

• Actual commands/code

• Gotchas discovered

• Examples from current work