Concepts
Understanding the two core AI asset types — skills and agents — helps you get the most out of Copilot when working with IDP projects.
Skills
A skill is a focused piece of AI guidance scoped to one developer workflow. Each skill covers a specific area of a framework project: for example, how to add validation rules, how to wire navigation between screens, or how to customize a grid.
Skills are loaded by the AI assistant on demand. When you ask a question that relates to a specific workflow, the assistant loads the relevant skill and uses it to answer accurately. Skills are not loaded all at once — only the skill that matches your current task is used, which keeps responses focused.
Skills are maintained alongside the framework source. They change when the framework changes, so the guidance they provide is always based on the current implementation.
Agents
An agent is a thin project-aware router. Its job is to understand the overall architecture of a framework project, decide which skill is the right one for a given question, and hand off to that skill.
When you ask a broad question — such as "how do I customize entity behavior in Innova.Data?" — an agent reads the project's architecture context and routes your question to the correct skill rather than answering from a generic base.
Agents know:
- which workflows exist in their project
- which skill owns each workflow
- which questions are out of scope for that project and where to look instead
How They Differ From General Copilot Usage
Without AI context, an AI assistant answers from its general training data. It may suggest patterns that do not match the IDP framework, use incorrect extension points, or miss IDP-specific conventions entirely.
With AI context added, the assistant answers from the framework's own guidance. It knows the correct base classes to extend, the safe hooks to override, the rule sets to use for validation, and the patterns the framework expects. The difference is most visible for tasks that require IDP-specific knowledge: writing validators, wiring navigation, customizing grid behavior, or bootstrapping an application.
The Layering Rule
The three layers have a clear division of responsibility:
- Agents route — they understand the project and point to the right skill
- Skills execute — they contain the actual workflow guidance
- Source code is the truth — the framework source is always the final authority; AI assets summarize and guide but do not replace reading the code when needed
The AI assets are designed to work with any AI assistant that supports context loading from a folder, such as GitHub Copilot (.github) or Claude (.claude). Choose the preferred AI folder that matches the assistant you use when adding the AI context through the Designer.