Agents
AI skill
A skill is a file packaging a reusable capability: a name, a description that decides when it loads, and a body of instructions loaded only once it does.
The economics are the whole design. The description sits in context permanently and must be short and specific. The body loads on demand and can afford to be thorough.
The rule that follows: anything needed on most requests belongs in the system prompt, anything conditional belongs in a skill.
In practice
The useful test for whether something should be a skill rather than a prompt: could a different agent, in a different conversation, invoke it and get a consistent result? If it depends on context that only exists in the current thread, it is a prompt. Skills that pass that test are the ones that stay useful past the week they were written.
Common questions
What is the difference between a skill and a prompt?
Reusability across contexts. If a different agent in a different conversation could invoke it and get a consistent result, it is a skill. If it depends on context that only exists in the current thread, it is a prompt.
When is something worth making into a skill?
When you have written it twice. The overhead of packaging — a name, a description, defined inputs — only pays back if it is invoked more than once, and the description is what determines whether it ever is.
Measure it
- AI Skill File BuilderFrontmatter validation, trigger-quality checks, and the token cost of what loads into context.
- Skill File Frontmatter ValidatorThe checks that decide whether a skill ever gets loaded.
- System Prompt to Skills SplitterSplits a system prompt into what belongs there and what belongs in a skill.