Automate Multi-Step Coding Tasks with the /goal Command
Execute complex multi-step coding tasks autonomously without constant manual intervention.
Use the /goal command in Qwen Code to trigger autonomous execution of complex coding tasks overseen by a secondary 'judge' model.
The Scenario
You have a multi-step coding task, like refactoring a legacy module while ensuring all tests pass, but you don't want to babysit the AI through every sub-command.
Before & after
Developers manually break down a task into five steps, execute each, and troubleshoot failures, which usually takes 30-45 minutes.
By typing /goal in the Qwen Code terminal, the agent uses an independent judge model to monitor progress and automatically detect impossible goals, completing the task in 2-5 minutes.
The Prompt
/goal [DESCRIBE_YOUR_COMPLEX_CODING_TASK_E.G._REFACTOR_THIS_COMPONENT_AND_ADD_UNIT_TESTS]
The /goal feature is designed for autonomous coding. It uses an LLM-based Auto Approval system and a separate 'judge model' to ensure the agent doesn't get stuck in loops or wander off-track, making it safe for hands-free operations.
Source
Blog | Qwen Code Docs"/goal autonomous coding... plus an independent judge model and impossible-goal detection."