Create, edit, audit, tidy, validate, or restructure AgentSkills and SKILL.md files.
The skill-creator is a specialized workflow for managing AgentSkills and their SKILL.md documentation files. It helps create, edit, audit, tidy, validate, and restructure skills to maintain consistency and quality across skill repositories. Skills are compact triggerable workflows where metadata remains always visible, while the body loads only after triggering, and references, scripts, and assets load on-demand for efficiency.
This skill enforces hard rules to keep SKILL.md files lean and focused, recognizing that the base model already has extensive capabilities. It manages YAML frontmatter validation, ensures proper structure with name and description fields, and organizes supporting materials into dedicated directories (scripts for deterministic helpers, references for documentation, assets for templates and media). The skill removes generic advice that base models already know while preserving critical information like brittle command syntax, authentication caveats, and safety rules.
The skill-creator is designed for AI skill developers, documentation maintainers, and teams building skill libraries. It streamlines the editing workflow by reading existing skills, removing redundant content, relaxing prose to allow fragments, and validating both frontmatter YAML and any touched script tests. It includes validation tools like quick_validate.py to ensure skills meet structural requirements before deployment.
A skill must contain SKILL.md with YAML frontmatter. Optionally, it can include scripts/ for deterministic helpers, references/ for documentation loaded on-demand, assets/ for templates and media, and agents/ for UI metadata.
At minimum, name and description are required. The description must be quoted. Local OpenClaw skills may also use metadata, homepage, allowed-tools, user-invocable, and license fields.
Use the provided validation script: python skills/skill-creator/scripts/quick_validate.py skills/<name>. Alternatively, run the inline Python validation that checks frontmatter YAML syntax across all skills.
Remove generic advice the base model already knows, verbose tables that could be bullet lists, and redundant setup documentation. Keep only trigger-critical facts, brittle command syntax, authentication caveats, safety rules, and validation logic.
Use a short noun-phrase that describes the generic trigger, not the full workflow. Focus on what the skill does (e.g., 'Inspect, split, merge, OCR, redact, or convert PDFs') rather than how it does it. Keep it concise for efficient triggering.
Quick Setup:
.claude/skills/Repository
steipete/clawdis