creating-local-skills

Creating Local Skills

Overview

This built-in guide defines how WorkClaw generates local skills in a stable, maintainable way. Use it as the single source of truth for skill scaffold structure and quality checks.

Core Principles

• Keep SKILL.md concise; include only task-relevant instructions.
• Description must be specific and discovery-friendly, including "Use when ..." trigger conditions.
• Use progressive disclosure: keep overview in SKILL.md and link extra files only when needed.
• Prefer one recommended workflow instead of many equivalent options.
• Ensure output can be verified with explicit quality checks.
• Treat trigger quality as a first-class concern: generated skills should reduce both false positives and false negatives.
• Keep default frontmatter minimal, but allow optional advanced fields such as allowed_tools, context, agent, and mcp-servers when the skill needs tighter runtime constraints.

Template Source

• Primary scaffold template: templates/LOCAL_SKILL_TEMPLATE.md
• If creation flow changes, update this guide and the template together.

Creation Quality Checklist

• Frontmatter includes only name and description.
• Description is in third person and contains concrete trigger context.
• Body includes Overview, When to Use, When Not to Use, Workflow, Prompt Examples, and Quality Checklist.
• Generated paths use forward slashes in examples and references.
• Default scaffold stays under a lightweight token budget.
• Prompt examples cover both should-trigger and should-not-trigger cases.
• Guide and template leave room for optional advanced frontmatter when needed.

Iteration Loop

1. Observe real user-created skills and identify weak sections.
2. Update template with minimal targeted changes.
3. Validate generated output readability, trigger quality, and prompt examples.
4. Look for 误触发 and 漏触发, then tighten the description or examples instead of adding verbose instructions.
5. Keep this guide aligned with the latest template behavior.