core-skill-creation_skill

Overview

This skill guides creators through designing, building, and packaging effective skills that extend an AI agent with specialized knowledge and repeatable workflows. It provides a clear end-to-end process, resource planning advice, and lightweight templates to keep skills concise and token-efficient. The goal is to produce modular skill packages that are easy to maintain and safe to load into an agent's context.

How this skill works

The skill inspects your proposed use cases and helps you convert them into a compact skill manifest, reusable scripts, reference docs, and output assets. It recommends what content to keep in the core manifest versus what to split into external reference or asset files, and offers stepwise guidance for initializing, iterating, and packaging the skill. It also enforces progressive disclosure so only relevant resources are loaded at runtime.

When to use it

• You want to create a new skill to automate a repeatable domain workflow.
• You need to turn domain expertise or company logic into an agent-ready package.
• You must decide which code, documentation, or templates to bundle with a skill.
• You need a checklist and packaging steps before publishing a skill.
• You want to make skills token-efficient and easy for the agent to load.

Best practices

• Keep the core manifest concise—only include information the agent truly needs in the common context.
• Use progressive disclosure: metadata always loaded, core instructions loaded on trigger, large references loaded only when required.
• Classify content by purpose: scripts for deterministic tasks, references for domain data, assets for files used in outputs.
• Match degree of freedom to task fragility: narrow for brittle procedures, wide for flexible text tasks.
• Avoid extraneous documentation files that bloat the skill; include only resources that support execution.

Example use cases

• Build an image-processing skill with a bundled rotate/resize script and an assets folder of templates.
• Create a finance-skill that stores table schemas in reference files and exposes a compact workflow in the manifest.
• Package a frontend scaffolding skill with boilerplate assets and a short manifest describing usage and triggers.
• Convert a fragile ETL routine into a skill with low-freedom scripts and a short runbook in the manifest.
• Iterate on an existing skill: add reference docs for new edge cases and keep the main instructions stable.