AI coding assistants are powerful but often unpredictable. A developer might ask for a simple permission control feature, and the AI starts modifying thirty files, choosing an approach that doesn't match the user's intent. Even with clear planning documents like proposal.md, design.md, and tasks.md, the AI can still go off-script. This article presents a practical solution: combining two open-source frameworks to build a plugin that keeps AI code generation under control. The plugin enforces direction by validating that the AI's changes align with the user's stated requirements and the predefined task plan. It acts as a guardrail, preventing the AI from making unauthorized modifications or deviating from the agreed-upon architecture. The author shares their experience of building this plugin, highlighting the key design decisions and how the frameworks complement each other. For teams struggling with AI code generation reliability, this approach offers a way to harness AI's productivity while maintaining control over the codebase.
The article addresses a common pain point: AI coding assistants making unwanted changes or ignoring predefined plans. It proposes using two open-source frameworks to create a plugin that enforces direction and task constraints. This is valuable for teams integrating AI into their development workflow.