Skills

A skill is a reusable capability bundle. It bundles the instructions and the specific tools an agent needs to do a job, and attaches to an agent (a prompt plus a model choice) so that agent gains the capability. Once packaged as a skill, you can attach it wherever that job needs doing instead of wiring the same tools together by hand each time.

Why skills

Most useful work is a repeatable pattern: read a task, call a tool or two, write a result back. A skill captures that pattern once. It keeps the agent's instructions and its tool access together, so the capability behaves the same way everywhere you use it. When you improve the underlying agent or its prompt, every place that uses the skill picks up the change.

This keeps your building blocks small and composable. An agent stays a single unit of prompt and model. A skill adds the tools that make that unit useful for a concrete job.

Composing capabilities

Attach a skill to an agent to give it a ready capability, or add several skills to an assistant, the composite deliverable made of multiple agents working together. Because a skill carries its own tool access, adding one to an assistant is enough to bring that capability online without configuring each tool separately.

A few practical notes:

• Agents can inherit the platform default model, so operators change one setting rather than many. Skills attached to those agents follow the same rule.
• Skills draw on the same tool catalog as the rest of the framework, including MCP servers exposed over DevJock's MCP-over-HTTP API.
• Keep each skill focused on one job. Small, single-purpose skills are easier to reuse and reason about than broad ones.

Where to go next