This extension provides a set of Language Model Tools for Copilot agent mode, enabling one Copilot instance to initiate, monitor, and receive results from another asynchronous Copilot task.
It allows a primary Copilot agent to delegate a task to a secondary Copilot agent running in a separate chat session, track the progress of this sub-task, and retrieve its outcome.
Features
Asynchronous Task Delegation: Copilot can start a new Copilot agent session (sub-agent) with a custom prompt without blocking its own execution.
Session Tracking: Uses unique session IDs (UUIDs) to track each delegated task.
Status Monitoring: Provides a tool for the primary agent to poll the status (pending, completed, error) of the sub-task.
Result Retrieval: Allows the primary agent to retrieve the final result or error message from the completed sub-task.
Completion Signaling: Includes a tool for the sub-agent to signal its completion or failure back to the tracking system.
Workflow
Initiation: The primary Copilot agent calls the copilot-task-delegate_start tool with a prompt.
Session ID: The tool immediately returns a unique sessionId (UUID) to the primary agent.
Sub-Agent Spawn: In the background, the extension prepares a modified prompt (instructing the sub-agent to report back upon completion using copilot-task-delegate_complete) and opens a new VS Code agent chat session with this prompt, effectively spawning the sub-agent.
Polling (Optional): The primary agent can periodically call the copilot-task-delegate_status tool with the sessionId to check if the sub-task is still pending or has finished.
Completion: Once the sub-agent completes its assigned task, it must call the copilot-task-delegate_complete tool, providing the original sessionId and either a result or an error message.
Result Retrieval: The primary agent calls copilot-task-delegate_status again. The status will now be completed or error, and the corresponding result or error message will be included in the response.
Provided Tools
copilot-task-delegate_start
Purpose: Starts the asynchronous sub-agent task.
Input:prompt (string, Markdown supported).
Output: Confirmation message with the unique sessionId (string, UUID).
copilot-task-delegate_status
Purpose: Checks the status and result/error of a sub-agent task.
Input:sessionId (string, UUID).
Output: JSON string representing the current state ({status: 'pending'|'completed'|'error', result?: any, error?: string, startTime: number}).
copilot-task-delegate_complete
Purpose: Used by the sub-agent to signal completion or failure.
Input:sessionId (string, UUID), result (string, optional), error (string, optional). One of result or error must be provided.
Output: Confirmation message.
Usage
This extension is designed for interaction between Copilot agents via tool calling. Direct user invocation of these tools is not the intended use case.
Example Usage Scenario
A primary agent could be configured with a persona prompt to manage complex tasks by delegating them to sub-agents. See the example prompt file:
