ecu.test agent Chat¶

The ecu.test agent chat provides an interactive way to ask questions and work with existing Packages and trace analyses. A planner interprets each request, forwards it to the most suitable enabled subagent, and analyses. A planner interprets each request, forwards it to the most suitable enabled subagent, and summarizes the result. For a larger task, the planner can call several subagents in sequence until the task is complete.

Open the Chat panel with Window ‣ ecu.test agent. The Chat client starts automatically when required.

Starting a conversation¶

Enter a request in the input field and send it. The selected model, Chat mode, and enabled subagents apply to the current conversation.

Start a new conversation with the plus button or press Ctrl+R. This clears the current conversation. Selections made directly in Chat are also reset to their configured defaults.

Configure your chat¶

Use the cog-wheel button to configure defaults for new conversations:

Model

Model used by Chat. This selection is independent of the Generation model used for button-based generation.

Chat mode

Default access level for a conversation.

Active subagents

Subagents available to the planner. All subagents are enabled by default and can be disabled individually.

You can change the same selections directly in Chat for the current conversation. These temporary changes are discarded when you start a new conversation. Changes saved with the cog-wheel become the defaults for subsequent conversations.

Chat modes¶

Agent

Allows enabled subagents to read workspace content and modify supported content in open editors.

Ask

Read-only mode. Subagents can use all available read tools, but cannot modify Packages, trace analyses, or other content.

Use Ask when you only want an explanation or review. Use Agent when the request should change supported test content.

Builtin subagents¶

Test Step Agent¶

The Test Step Agent reads and modifies existing Packages. It can:

• inspect the current Package

• add, edit, or remove Blocks and supported test steps

• edit Package variables

• use available workspace mappings and Package context

• find and open an existing Package by workspace path

The phrase current Package refers to the Package in the focused editor. Use this phrase to make the target explicit, for example: Add the required checks to the current Package. You can also provide a workspace path or ask the subagent to find a Package in the workspace.

The subagent cannot create a new Package and cannot execute tests. Changes are applied directly to the editor but are not saved to disk. A review of the changes is recommended before saving.

Trace Analysis Agent (Beta)¶

The Trace Analysis Agent reads and modifies existing trace analyses. It can add, edit, and remove supported trace steps. The supported step types and limitations are listed in Button-based generation.

The active trace analysis in the focused editor is the default target. Changes are applied to the editor but are not saved to disk. A review of the changes is recommended before saving.

Review Agent¶

The Review Agent performs a read-only review of an existing Package. It checks whether the implementation matches:

• the Package description

• the rules in package_rules.md

Place package_rules.md in the ecu.test agent customization directory configured under Options ‣ Settings ‣ Directories. The Review Agent reads the current file when it performs the review.

A rule file can contain project-specific Package guidelines:

Package variable naming convention: lowerCamelCase
Input parameter prefix: P_
Return value prefix: R_

The Review Agent reports findings but does not change the Package. In Agent mode, ask the planner to hand relevant findings to the Test Step Agent if they should be fixed.

Knowledge Agent¶

The Knowledge Agent answers questions about ecu.test by searching the installed user documentation. Use it for explanations of product concepts, dialogs, settings, and workflows.

Reviewing and saving changes¶

Changes made by Chat appear directly in the corresponding editor. There is no separate diff view. Review the changed Blocks, test steps, variables, or trace steps in the editor.

Use the standard undo and redo commands to revert or reapply changes. Save the Package with the normal editor command only after reviewing the result.

Examples¶

Modifying the current Package¶

Focus the Package to be changed, select Agent mode, and enter a specific request:

In the current Package, add a Block that writes 50 % to the driver request.
Wait five seconds, read the vehicle speed, and check that the value is 50 km/h with a
tolerance of 2 km/h. Reuse existing mappings and variables where appropriate.

The planner delegates the task to the Test Step Agent. Review the generated Block and steps in the Package editor, then save the Package manually if the result is correct.

Modifying a trace analysis¶

Focus the trace analysis, select Agent mode, and enter:

In the current trace analysis, add a Block that verifies that Speed reaches
50 km/h at least once. Plot Speed in an overview plot.

The planner delegates the task to the Trace Analysis Agent. Review the generated trace steps and confirm that they only use the supported options.

Reviewing a Package¶

Configure package_rules.md, focus the Package, and enter:

Review the current Package against its description and our Package rules.
Summarize every finding and do not change the Package.

To address accepted findings in the same conversation, switch to Agent mode if needed and enter:

Hand the accepted variable-naming findings to the Test Step Agent and fix them
in the current Package.

Review the resulting editor changes and use undo if a change should not be kept.

Asking about ecu.test¶

Select Ask mode and enter:

Explain the difference between a Package parameter and a return value in ecu.test.
Include the relevant editor workflow.

The planner uses the Knowledge Agent to answer from the installed user documentation without changing workspace content.