I Built a CLI to Replace Expensive AI Directive Generation

There has to be a way to get the Coding Agent and Design Agent into a loop, even within AI IDE Programs, to remove the friction of acting as a go-between for Claude and Windsurf.

That was the thought that led to OpenAgent. I was tired of pasting the same context into different AI tools, watching them miss the architectural patterns I’d established, and then having to course-correct them back toward my actual codebase.

So I built a CLI that reads my actual codebase, understands the structure, and generates strategic directives shaped to my development style.

The honest part: I didn’t know if it would work. The first version just dumped file paths and a generic “analyze this” prompt. The AI ignored it and gave me generic advice.

The breakthrough was the SOUL.md interview — eight questions about how I actually work, what I value, what patterns I use. That profile shapes every directive the tool generates. Now when OpenAgent suggests an improvement, it sounds like me, not like a generic best practices blog.

It’s on PyPI now as openagent-directive. It’s not perfect. Sometimes it misses the mark. But it saves me hours of context-switching every week, and the directives it generates are actually useful.

What didn’t work: Trying to make it too smart initially. I wanted it to understand everything about my codebase instantly. That led to slow responses and hallucinations. Scaling it back to focus on structure and patterns made it actually usable.

The lesson: Sometimes the most useful tool is the one that understands your constraints, not the one that tries to solve everything.