Skills vs models vs APIs
The core building blocks of the ecosystem — when to use each one and how they fit together.
Three words come up constantly across the agent harness ecosystem: skills, models, and APIs. They're related but distinct, and using the wrong one for a job creates frustrating problems. Here's the practical distinction.
Skills — what they are
A skill is a pre-packaged capability. It's designed to do one specific job well: search the web, extract data from a PDF, monitor a URL for changes, generate a social media post from a brief.
Skills handle the interface. They know how to accept input in a standard format, do the work, and return output in a standard format. You don't need to understand how they work internally — you just need to know what they take in and what they return.
When to use a skill: When you need to do a defined, repeatable task and you don't want to build the logic yourself.
Models — what they are
A model is the AI reasoning layer. It's what actually thinks — generating text, analysing images, classifying content, answering questions. Skills often use models internally, but models are also available directly when you need raw reasoning power without the pre-packaged wrapper.
Different models are genuinely different. A model optimised for code is not the same as one optimised for creative writing. A fast, cheap model is appropriate for some tasks; a slower, more capable one is worth the cost for others.
When to use a model directly: When you're building something custom, evaluating output quality, or a pre-built skill doesn't exist for your specific need.
APIs — what they are
An API is a connection point to live external data or a third-party service. Think: a weather data API, a financial data feed, a database of company information, a messaging platform integration.
APIs don't do reasoning — they provide data. Skills and models do the thinking; APIs supply what they think about.
When to use an API: When your workflow needs real-time or external data that doesn't live inside the AI system itself.
How they fit together
Most real workflows use all three. A research workflow might:
1. Use a search API to pull live information on a topic
2. Pass the results to a reasoning model to identify what's relevant
3. Run them through a summarisation skill to produce a clean output
ClawCurrent's directory tags every listing clearly so you always know which category you're working with. The trust score applies equally to all three.
Skills do jobs. Models reason. APIs connect to the world. Most workflows need all three — the skill handles the job, the model does the thinking, the API brings in the data.