skill-creator_skill

Overview

This skill guides creators to build concise, maintainable skills that extend an agent with domain knowledge, workflows, and tool integrations. It focuses on designing reusable resources, managing context efficiently, and matching instruction specificity to task fragility. Use it to plan, initialize, and iterate on new or existing skills for reliable agent behavior.

How this skill works

The skill inspects desired user scenarios and maps them to reusable resources: small manifest metadata, executable scripts, reference documentation, and output assets. It prescribes a progressive-loading design so the agent only loads detailed files when needed, and it recommends degrees of freedom (high/medium/low) to control how prescriptive instructions should be. It also provides a step-by-step creation process from discovery to packaging and iteration.

When to use it

• You want to create a new skill that adds domain-specific workflows or integrations.
• You need to refactor or update an existing skill for clarity, reliability, or token efficiency.
• You want to convert repeated ad-hoc instructions into reusable scripts or templates.
• You need guidance on organizing large reference docs so the agent loads only relevant parts.
• You plan to add assets (templates, images, sample code) used directly in agent outputs.

Best practices

• Keep manifest metadata short and explicit so the agent triggers correctly.
• Prefer concise examples over long prose to save context tokens.
• Choose the degree of freedom (high/medium/low) based on task fragility.
• Move large or detailed material into separate reference files and load them conditionally.
• Include executable scripts for deterministic operations and assets for output templates.
• Avoid extraneous documentation files; include only what the agent needs to act.

Example use cases

• Create a PDF toolbox skill with a rotate script and a small reference for advanced options.
• Build a finance-query skill that ships schema references for only the queried domain.
• Package a frontend-boilerplate asset folder so the agent can produce a starter webapp.
• Refactor an existing image-editing skill to include deterministic scripts for fragile steps.
• Design a support-agent skill that loads policy references only when compliance questions arise.