Goal-Driven Prompt Released

To address the issue of coding agents arbitrarily declaring work complete, a prompt tool called Goal-Driven has been released. This tool separately defines Completion Criteria and has a master agent verify them.

This project is not executable code but rather prompt text in Markdown format. When a user inputs a goal and completion criteria, the master agent compares the sub-agent's output against the criteria and, if unmet, instructs the creation of a new sub-agent to continue the task. The author stated that this approach enables over 300 hours of continuous autonomous coding.

Scope and Limitations

• Suitable Tasks: Complex system designs (e.g., compilers, DB architectures) where mechanical judgment is possible, such as passing tests or matching outputs.
• Unsuitable Tasks: Work requiring human judgment, such as UI polish.
• Caution: Token consumption is high, so cost verification is necessary. To prevent context pollution, paste the prompt directly into the conversation without registering skills or plugins.