Every morning, your agent meets your project for the first time.

This is documented behaviour, not a defect. Anthropic's own memory page opens with it:

"Each Claude Code session begins with a fresh context window."

code.claude.com, How Claude remembers your project

Two mechanisms are offered to carry things across. An instructions file you write, and an automatic memory the model keeps for itself. Both are useful. Both have a shape worth knowing before you lean on them.

The automatic memory is loaded with a budget: "The first 200 lines of MEMORY.md, or the first 25KB, whichever comes first, are loaded at the start of every conversation. Content beyond that threshold is not loaded at session start." So it is an index, not an archive.

And it does not travel. In the same page: "Auto memory is machine-local. All worktrees and subdirectories within the same git repository share one auto memory directory. Files are not shared across machines or cloud environments." Your laptop remembers things your desktop never learns.

Write the handoff yourself. A file in the repository, called something like HANDOFF.md, updated at the end of each session and read at the start of the next.

This is a real answer, not a stepping stone. Give it credit before you go looking for anything else.

What the hand-written version gets right

• It costs nothing and takes ten minutes to set up.
• It works with every tool, and it keeps working when you change tools.
• Committed to the repository, it is versioned, reviewable and shared with the team.
• You control exactly what goes in, which is more than most memory systems offer.

People go further than a file. The author of issue #25999 built a whole memory layer with lifecycle hooks, and reported more than fifty engineers using it. The instinct is correct: the state belongs outside the conversation.

Where it breaks, in the order it usually breaks

• It depends on remembering. The handoff has to be written at the end of the session, which is the moment you are tired, or interrupted, or the terminal has already closed on you.
• It drifts. Nobody deletes from it. After a month it is a long file describing three versions of the project at once, and the agent reads all of them.
• It is one file for the whole project. Everything in it loads for every task, whether the task is the payment flow or a button colour.
• It says what you remembered to say. It cannot tell you what was actually completed yesterday, because it is a note, not a record.

None of those are reasons not to start. Start there today. They are the reasons people eventually want the writing to happen without them.

Mission HQ is the handoff file, written for you, kept on a server, and cut to the task in front of you.

One call returns the state, not a file to read

At the start of a session the agent asks for the handoff and gets back where the project stands, the single next action, and the warnings that matter: failing tests, tasks gone stale, decisions still waiting on you.

The note is written while you work

Checkpoints are posted on every prompt and at the end of every turn, and the session closes with a summary, the state, the next steps and the warnings. Nothing depends on you remembering at the end.

Knowledge arrives scoped, not as one growing file

What the project has learned is stored per module and per kind of work, and delivered when a task of that kind starts. Working on the database means you get the database conventions, not four hundred lines about everything.

It knows what was actually finished

Tasks close against their own criteria, so the record of yesterday is what completed, not what someone summarised. That is the part a note can never give you.

A hand-written handoff file is the right tool for a small project, one person, a few weeks. It is free, it is yours, and for that shape of work it is genuinely enough. We would rather you used it than bought something you do not need.

It stops fitting on three axes: time, because the file grows and nobody prunes it; people, because your automatic memory stays on your machine and your teammate's stays on theirs; and trust, because a note describes what someone believed, not what actually closed.

That is the whole decision. Not a better notebook against a worse one, but a note against a record.