Scoring and feedback

Using agents to write feedback for a learner and score them against a competency model.

Agents analyze the conversation and produce outputs. They watch the conversation, grade the learner, and provide feedback based on a rubric. Their outputs populate feedback templates and, when scoring is enabled, write to user-attribute scores. An agent runs when the learner hits "End" during their run through.

Scoring a learner's skills needs approval on your account, and new competency models are built by SimInsights on request. Email [e-mail] to ask for either.

Agent settings

In the Agents section of the AI tab, Create agent adds an agent, the dropdown switches between agents, and Delete removes the selected one. Each agent's Configuration has:

• Name: the agent's name.

• When it runs: On session end. Agents run automatically when the learner ends the session.

• Language model: GPT-5, GPT-4o, GPT-4o mini or Gemini 2.5 Flash. See Large Language Models & Data Privacy.

• Use history: Lets the agent use the chat history.

Feedback for the learner

Two steps: the agent produces the feedback and the scenario flow decides where it appears.

1

Give the agent an output

Create an agent and add an output with Destination set to Display only.

Description is how the AI knows what to write, for example 2 bullet points naming what to work on. You can put that guidance here or in the agent's base prompt.

Reference name is how you point at this output later.

2

Reference it in the feedback panel

You can reference the AI's analysis/provided feedback so that the learner can receive feedback at the end of their session. In this example it is referenced as <Agent.SalesEvaluationAgent.coaching_summary>. The copy icon next to the reference name copies the exact text.

Scoring a skill

A competency model is a tree of skills a learner can be measured against. See Competency models.

1

Attach a competency model

Choose the model you want the Spark scored against.

2

Add an output the agent scores on

The agent judges the conversation against that skill and writes a value. To do this, add a User Attribute output to the agent.

The result lands on the learner's run in HyperSkill Platform, next to scores from 3D simulations. See How scores work.

Seeing the results

A Spark's analytics work like any simulation's: how many learners completed it, average completion time, completion rate and average mastery, then a table of individual runs. Device reads Web (Spark).

Opening a run shows the learner's skill mastery and how it moved, the session narrative of what they said and what the character said back, and the raw event log. See Simulation analytics and Run detail.