I work on my portfolio, two UI starters, and the tools I use with AI. I move between agents too. A decision made while working on the portfolio might matter to a starter, but the agent working on that starter won't necessarily have seen the conversation.
I use Taskosaur to keep track of those connections. My customized setup has a workspace called Shared Work, with a project for each codebase. Alongside the tasks, Docs holds the guidance I want to use again.
Picking up where I left off
Chat works well while I'm in the middle of something. I can try an idea, ask for a change, and work out what I meant. Coming back later is harder. The latest reply may say what changed without explaining why.
I wanted a place to check before starting another conversation. In Shared Work, I can find the task under its project and see where we stopped. Keeping the projects together helps me follow related work without putting everything into one task list.
What I put in a task
I want the task to tell me what we're doing, who's working on it, and what needs to happen next. Checkpoints record progress and link to the files or review. They also say what was checked and what's still uncertain.
These articles are an example. A draft can render correctly and pass the code checks while still needing my review. The task keeps that distinction visible: ready to read doesn't mean ready to publish.
What goes in Docs
Docs is where I keep instructions and explanations that apply beyond one task. My collaboration guidance belongs there, as do project conventions and reasons for decisions I expect to revisit.
A document can belong to the organization, a workspace, or a project. That lets me keep shared guidance in one place and project details with the project. Tasks link to those documents.
Underneath the interface, they're Markdown files. I can edit a document in Taskosaur, and an agent with access can read that same file. I don't need to maintain a separate copy for each tool.
Following a change through
For an interface change, the task describes what needs changing and links to the project's design guidance. The code stays in the repository. After the change is checked, the task gets a checkpoint with the result and anything left to do.
If we made a decision that will affect later work, it belongs in the guidance too. A temporary detail, such as the port used for a local preview, can stay in the task. I want to be able to find it without making every future agent read it.
Switching agents
When I move to another agent, I can point it to the task and the documents it needs. I still have to tell it to read them. Having access to Taskosaur doesn't mean it knows what's in there.
I've also had to deal with an agent using old instructions. A shared source only helps if the agent reads the current version. When several agents are working at once, I need to know which one is changing which files.
Keeping it useful
I don't want maintaining the workspace to become another project that gets in the way of everything else. I need enough detail to resume the work, with links when I need to go deeper.
The test for me is what happens when I return after a break. Can I see where we stopped and what needs my attention? That's what I want this setup to make easier.