[DOCS]: Best practice for custom plug-in agent success

[esaic]

Description

It is not clear how the plugin is called during the agentic process.
What is the prescribed flow and metrics to decide whether an agent tasks has been successfully completed?
I would like to know how to best design plugins that minimize the loop calling. In my application many of the calls are successfully completed, but the agentic flow keeps making calls to the agent skill.
If there's documentation covering this, I'd like to know. Thanks!