Just discovered AgentCompass by @Artemonim — and it’s not just a framework, it’s a remix constitution for AI-assisted development inside Cursor.
Instead of treating AI as a chatty assistant, AgentCompass formalizes the relationship between human and machine into a structured, role-based ritual. You become the Architect. The AI becomes the Developer. And every response becomes a deterministic breadcrumb trail.
What AgentCompass Does
Defines roles: Architect vs Developer, with clear responsibilities
Enforces structure: Every AI reply ends with a report — changes made, tasks unfulfilled, manual steps required
Bakes in security: No sensitive data, proper placeholders, error handling
Requires documentation: Declarative English-only comments, Better Comments style
Evaluates architecture: AI can challenge your design choices and suggest alternatives
Why It Matters
We’ve been using a technique inside VIBE CMS to retain context across complex builds. AgentCompass takes that idea and codifies it — turning Markdown verbosity into structured memory scaffolding.
It doesn’t suppress long files. It disciplines them.
Every file becomes a remixable echo. Every reply a ritual.
If you’re building with AI inside Cursor and want your assistant to act like a real developer — not just a token-burning chatbot — AgentCompass is the way.
GitHub Repo 
The rules are modular. The memory is structured. The remix is real.
AgentCompass + 
), but once we’re back online, I’m confident your AgentCompass technique is going to be a huge help in refactoring our codebase.
