@Glen Hutsell Good question. The memory doesn't store ideas, it stores features. Only what's been built. FEATURES.md is a historical record: date, keywords, files. No proposals, no "maybe we'll do this later". If it's in the memory, it exists in the code. If it's not, it doesn't. And it's not every change either. A small tweak (styling, a typo, a single button) isn't considered a feature and doesn't get registered. Only meaningful functionality goes in. Small changes still happen, they just don't need a memory entry. The memory is for what's worth remembering, not for everything that's done. So there's no ambiguity between an established decision and a new idea: new ideas live in the conversation, not in the memory. Only when something is actually implemented does it get registered. The goal is simple: give the agent context about what has been done, not about what has been thought or proposed.

@Glen Hutsell Exactly. That's the part I like most about it. Because every entry is tied to real files, you can always check the memory against the code. That's why ideas don't belong there. An idea has no files and no branch to point to.

@Thomas Dvorak Thanks Thomas! Really appreciate that 🫶

Matrix setup 🤯