skill-quality-standards

---

name: skill-quality-standards description: When creating or validating skills. Reference for SKILL-CREATOR and SKILL-VALIDATOR. version: 1.1.0 tokens: ~420 confidence: high sources:

• internal: .claude/skills/REGISTRY.yaml
• https://www.anthropic.com/news/claude-2-1-prompting (context window optimization) last_validated: 2025-12-10 next_review: 2025-12-24 tags: [skills, quality, standards, meta]

---

When to Use

When creating or validating skills. Reference for SKILL-CREATOR and SKILL-VALIDATOR.

Patterns

Size Limits

CopyTarget:  400-1000 tokens (optimal)
Maximum: 1500 tokens (hard limit)
Per task: max 3 skills loaded

Estimation:
  1 word ≈ 1.3 tokens
  1 code line ≈ 10 tokens
  Full skill ≈ 100-150 lines

Confidence Levels

CopyHIGH:
  - 2+ authoritative sources (Tier 1-2)
  - Patterns tested in production
  - Official docs cited

MEDIUM:
  - 1 authoritative source
  - Community-validated patterns
  - Reputable tech blog

LOW:
  - Blog posts only
  - Experimental/unverified
  - No official source

Required Sections

Copy---
[YAML frontmatter with metadata]
sources:
  - [URL or internal path]
---

## When to Use
[1-2 sentences - clear trigger]

## Patterns
[2-3 patterns with code examples]
[Each with source citation]

## Anti-Patterns
[What NOT to do + why]

## Verification Checklist
[Actionable checks]

Skill Types

CopyGeneric:    .claude/skills/generic/
            Tech-agnostic, reusable anywhere

Domain:     .claude/skills/domain/
            Industry-specific (fintech, healthcare)

Project:    .claude/skills/project/
            Repo-specific patterns

Anti-Patterns

• Skills over 1500 tokens (split them)
• Patterns without source links in frontmatter
• Vague "When to Use" triggers
• Theory without code examples
• Missing anti-patterns section

Verification Checklist

• Under 1500 tokens
• Every pattern has source in frontmatter
• "When to Use" is specific
• 2+ patterns with code
• Anti-patterns included
• Checklist is actionable