1. Configuration

1.1. Agents

Caret allows users to configure the execution order of agents responsible for classifying and processing tasks. These agents do not necessarily have to use the same technology. Additionally, you can configure the agent sequence for code completion (Content Assistant).

For each LLM agent, the user must configure the API Key, the API URL, and the Model.

1.2. Database

Configure MongoDB access by entering the Connection String and the Database Name. This database stores the assistant's usage traceability model and the newly defined task plugins.

1.3. Tasks

Users can configure tasks by defining the number of pre-validation and post-validation cycles, as well as the context information to be sent with each request for a specific task.

1.4. Task Repository

Newly defined tasks can be shared for install via a Task P2 Repository. To configure access, the user must enter the Repository URL. Optionally, a bearer token can be added if one has already been defined on the plugin server.

1.5. Version Control

Caret automatically detects the current Git user and completes the Git User and Git Email fields. The user can manually edit these fields if he wishes to use different credentials for traceability records.

2. Task Definition

Caret utilizes the TaskGroup extension point to add new tasks. Users can easily define one or multiple tasks using the standard Eclipse extension UI.

The following mandatory information must be defined:

• Code, Name, Description, Bind type.
• Pre-validations and Post-validations (if applicable)

Optional fields include the commandId and any additional instructions to be appended to the task description. Furthermore, the user must define the necessary parameters, the context to be sent, and the action to be deployed.

3. Usage

Caret allows users to request a task through defined commands (tasks) in the contextual menu, which can be accessed by right-clicking on the Eclipse code editor or the Package Explorer.

Additionally, the user can request a task through the text field in the chat window.