skill-quality-standards

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

⭐5stars🔀1forks📁View on GitHub🕐Updated Jan 11, 2026

When & Why to Use This Skill

This Claude skill establishes a standardized framework for the creation and validation of agentic skills. It provides rigorous quality benchmarks, including token management, structural requirements, and verification checklists to ensure AI agents operate with high reliability and efficiency.

Use Cases

  • Standardizing Skill Creation: Providing a consistent template and metadata structure for developers building new Claude agent capabilities.
  • Quality Validation: Auditing existing skills against predefined standards to ensure they meet token limits and include necessary safety patterns.
  • Context Window Optimization: Managing token density (400-1000 tokens) to prevent context overflow and maintain high agent performance.
  • Documentation Governance: Ensuring all skills include critical sections like 'Anti-Patterns' and 'Verification Checklists' for better maintainability.
nameskill-quality-standards
descriptionWhen creating or validating skills. Reference for SKILL-CREATOR and SKILL-VALIDATOR.
version1.1.0
tokens~420
confidencehigh
- internal.claude/skills/REGISTRY.yaml
- https//www.anthropic.com/news/claude-2-1-prompting (context window optimization)
last_validated2025-12-10
next_review2025-12-24
tags[skills, quality, standards, meta]

When to Use

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

Patterns

Size Limits

Target:  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

HIGH:
  - 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

---
[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

Generic:    .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