Claude Code 待辦清單：我不再靠記憶重建清單的那一天

A while ago, I was building another project in Claude Code. Vibe coding had surprised me; I could build an app myself, so I started a service.

The first features came together quickly, and using them, I kept thinking of new ideas. I laid out a development roadmap toward a beta release, and the work went smoothly.

But the more concrete the service became, the more there was to think about. I would tell Claude Code what I had noticed and work out a new strategy. Then I noticed something else, told it, and worked out another, and the planning documents kept piling up.

A beta plan, a v1 plan, a development plan v0.5, a next-milestone plan v0.1.

Eventually, I could no longer see where things stood.

How much of the beta plan was done? What was left, and in which planning document? Which parts contradicted each other?

I could not tell where to start, and the priorities drifted apart

Once the plan branches, a Claude Code todo list tends to end up living in your memory. For me, it became hard to know where to start.

Each piece of work raised new problems, so new tasks kept appearing. When I moved to another plan before finishing one, I lost track of how far anything had got. Across my documents, priority was what contradicted most often. Asking Claude Code to sort it all out again only added one more list, and before I knew it, I was holding my Claude Code todo list together from memory.

Building the board, using it, and leaving the plan to it

I wanted to make my conversations with Claude visible, and that is where Vondet came from. At first, I did not even know what to record; I kept trying things, kept using it myself, and kept fixing it.

The day I stopped rebuilding my Claude Code todo list from memory, and started reading the board instead. From that day on, my conversations with Claude began to change, little by little. Today I leave much of it to the board, and I talk with Claude, pick the task that matters most, and move it forward.

The board is where Claude writes, and where the two of you can point

Vondet is a board Claude writes to, connected over MCP, the shared method Claude uses to work with outside tools. There is no box for writing a plan from scratch; Claude writes the goals, the tasks, and the reasons.

When you ask for a tidy-up, every conversation leaves a fresh summary. With the board connected, each conversation adds to a single plan instead, and you just keep talking as usual.

The board is what the two of you point at when the plan comes up. Claude points at a move, and you open it when you want to change it.

Figure 1. Left: a few planning documents, priorities that disagree, and three questions you can only answer from memory. Right: Claude writes to one board, and the two of you point at the same place.

Once you connect, talk about the plan while pointing at the board

With the board connected, Claude can:

Read the board as you go, so you stop pasting the plan into the conversation.
Put anything that needs your decision under "Waiting on you", then move on to other work.

What it looks like together

In the example on our front page, you ask Claude to fix a double charge on an order form. Claude writes the task and its reason to the board before starting the fix, and leaves the undecided monthly price under Waiting on you.

The one decision Claude cannot make, left where both of you can see it. Images show a demo project.

After three days away, the board opens on how much moved, day by day, drawn as a chart. When you disagree with a call, tell Claude while you look at the board. Claude rewrites the call with its reasons, and the earlier call stays on record.

How much is done and how much remains, day by day. Images show a demo project.

Connecting takes one registration per app

In a repository, register a board token in Claude Code; for the Claude app, use Connect with permission. The guide has the steps: signing up, connecting your AI, and when connecting does not work.

Two ways to connect: a token for a repository, or permission for the Claude app. No token is shown. Images show a demo project.

What it covers today

Only Claude Code and the Claude app are tested end to end. The board lives on vondet.com, does not read your code, and there is no way yet to invite anyone else to your board.
Every plan exports the whole board as JSON.
On Free and Pro, content may be anonymized to improve suggestions; you can turn that off.

About this article

This article is based on my own experience and is subjective; nothing in it was measured objectively.

When you do not need Vondet

If you have one repository, one person writing, and work that ends within a few conversations, a file is enough, and you do not need to connect a board.

Can you answer the three questions this week?

Next time you come back after a few days, can you say right away how much is done, what is left, and what contradicts what? Open a new conversation, paste nothing, and ask Claude Code those three questions.

If its answers and your memory disagree, try keeping the answers to those three questions in one place.

This post was written with AI assistance from my own notes and experience, and reviewed by me. It was first published on vondet.com.

I build Vondet, the board described here.