Where does the context before an agent task live?

A lot of discussion around agents is about handoff: one agent does part of the work, another one continues it. But I keep getting stuck one step earlier. Before an agent can do anything useful, there is usually a pile of context behind the task: pages I read, images I saved, files on my computer, decisions made in earlier chats, and half-finished ideas. That context often exists before the first agent is involved. It also needs to survive after the task ends. Right now I usually end up rebuilding it: finding the tabs again, dragging files together, copying old messages, and trying to remember why a reference mattered. How are people handling this? Do you keep a project folder? A Notion page? A second brain? A repo? Do you just start a new chat and reconstruct it each time? I’m working on this problem in a project of my own, but I’m more interested in the workflows people have actually kept than in promoting a tool. ✨update I’d like to clarify that I am both the product owner and the creator: notes(context.md etc.) organize background, decisions, and context for handoff, while projects also accumulate raw materials—screenshots, web pages, images, files, posts, conversations, and AI outputs—that may inform decisions without belonging in the documentation. How should these resources and AI outputs be organized and managed? Should I use a local folder, asset library, knowledge base, or structured context system? The key is to ensure that all process data ultimately serves this project, because progress at every stage is valuable, especially for creative work. When starting a task, I usually provide both project notes and relevant source materials. Do you retain raw resources alongside updated notes, or keep only summaries and revisit the originals when needed?