ecu.test agent¶
The ecu.test agent helps you generate test steps with AI. It takes the current Block description and relevant workspace context and inserts suitable test steps for you. To do so, it connects to your active global mapping, global constants, configured library workspaces, and analyzes the current Package. It can also learn from your existing Packages, if you provide reference implementations.
To utilize the agent in your workspace, you need to connect to the agent service and configure it according to your needs.
Setup¶
Use to connect and configure the agent.
Connection¶
Url: Base URL of the agent service.
Api token: Authentication token for the service. You can show/hide the token.
Proxy settings: Optionally, configure a proxy. Supported modes: No proxy, System proxy, User-defined (with optional authentication). Browser/PAC proxies are currently not supported.
Info
Click Check to verify connectivity. On success, the dialog shows OK. If you are connected to our service, your remaining credits are displayed, too.
Reference implementation contexts¶
To improve the quality of the generated steps, you can allow the agent to learn from your existing Packages. Reoccurring patterns and solutions are picked up and applied to the current Block. This can significantly improve the quality of the generated steps.
Add one or more named contexts. Each context points to a folder that contains Packages to be used as “reference implementations”.
Click Generate to create references for that folder. The status shows Up to date, Outdated, or Not generated including the generation date.
You can select which context to use from the Generate with AI toolbar’s dropdown while working in a test case.
See Reference Implementations for details.
Filter settings¶
Here you can configure how many items of each category the agent should consider when generating steps. This is a kind of pre-filtering to limit the amount of data sent to the service.
Signal mappings: Number of signal mappings considered.
Jobs: Number of Tool and Port Jobs considered.
Package mappings: Number of Package mappings considered.
Packages from library workspaces: Number of library Packages considered.
Package variables: Number of Package variables considered.
Global constants: Number of global constants considered.
Reference implementation blocks: Number of closest matching reference Blocks included.
Info
The sum of all limits must not exceed 50. Higher values can improve quality but increase request size and latency.
Custom instructions¶
With these instructions, you can further steer the agent to your projects individual needs and improve generation quality.
Info
Formulate clear instructions (e.g., style rules, naming conventions, preferred/avoided items).
How to generate test steps¶
In the test case editor, select a Block whose contents you want to generate. Right-click and use the to generate test steps for this block.
You can also use the button in the toolbar (dropdown menu):
: Generates only for the selected Block(s).
: Generates only for Blocks that don’t have steps yet.
: Generates for all Blocks.
Info
In the same dropdown, pick a Reference implementation context to guide the agent. Only contexts which have been generated in the settings are selectable. See Reference Implementations for details. Review the generated steps before executing the test case.
Info
Provide a clear Block description (Action, Parameterization, Expectation), see How to improve the quality of the generated test steps.
Supported test steps¶
The agent can generate and parameterize these test steps:
Read and check mappings (with optional tolerance and time options)
Write mappings
Wait
Execute Tool and Port Jobs parameterized
Calculate and check expressions
Package call (workspace or library Packages with input/output parameters)
If-Then-Else
MultiCheck
In case the agent cannot generate suitable steps, it adds a Todo, informing you which additional information may be missing. If the agents response can’t be processed, a comment is inserted so you can adjust the Block description and try again.
Which information the agent considers¶
The agent compiles a context from your project to steer the generation. Items are filtered using the limits in .
Current Block description: Action, Parameterization and Expectation.
Current Package implementation: Other Blocks from the open Package.
Package variables: Parameters and returns of the current Package, including type and description.
Signal mappings: Signal mappings from the global mapping with their names, paths, types, and descriptions.
Tool and Port Jobs from the global mapping with names and descriptions.
Package mappings: Mapped Packages from the global mapping with name, description, and path. The agent also considers parameter/return names and descriptions of those Packages.
Packages from library workspaces: Packages discovered in configured library workspaces.
Global constants: Global constants of the current configuration with values and descriptions.
Reference implementations: The most similar Blocks from the selected reference implementation context.
Your Custom instructions, as defined in the settings.
How to improve the quality of the generated test steps¶
Be specific in Blocks¶
Action: Describe the goal, not the solution.
Parameterization: Provide relevant inputs, constraints, ranges, and units.
Expectation: State observable outcomes, thresholds, tolerances, and timing.
Use Custom instructions¶
Add style and naming conventions (e.g., variable prefixes, required units, allowed mappings).
Emphasize or restrict the use of certain signals, variables, or constants.
Apply test-specific strategies (e.g. waiting after reading, adding tolerances, performing a certain action always at the beginning/end of a test case etc.).
Apply domain-specific logic or rules.
Keep it clear, specific and consistent, use bullet points or short sentences.
Tune Filter settings¶
Increase limits for categories that matter most to your Block, but keep the total below 50.
If only Package calls should be generated, set the limit of Signal mappings to 0, vice versa if only Signal mappings are desired.
Lower limits can reduce latency and cost.
Maintain good metadata¶
Add clear descriptions to mappings, variables, global constants and Packages.
Use Reference implementations¶
Generate references for representative Packages and select that context in the toolbar. Regenerate when source Packages change.
Reference Implementations¶
Reference implementations let the agent learn from your existing Packages. Reoccurring patterns and solutions are picked up and applied to the current Block. This can significantly improve the quality of the generated steps.
Add a context¶
Open .
Click the + tab, enter a Name, and set the Path (workspace-relative paths are supported).
Click Generate to create the references for that folder.
The status shows Up to date with the generation date. It becomes Outdated if source files change.
Instead of manually clicking on Generate, you can also enable Update automatically to automatically update references when generating test steps in the test case editor. This setting applies to the currently selected context.
Select a context¶
In the test case editor, open the dropdown and select the desired context. Only contexts with generated references are enabled.
The agent considers the top matching Blocks (per your limit). The match is based on the semantic similarity between the current Block description and the reference Blocks.
Info
If references are missing or outdated for the selected context, you’ll be asked to generate them, before the agent can proceed to generate steps.
Info
Packages or directories prefixed with a dot (e.g., .hiddenPackage.pkg) are considered hidden
and are excluded from the references.
Warning
Generating references for large folders with many Packages can take a while. We recommend to select only representative Packages that cover your typical use cases.