Journal

At the start of a session your AI opens the project file and knows what shipped, what broke and what was already decided.

Every decision, every failure and its cause go in the same turn they occur, not at the end of the day when they are already half forgotten.

The file is plain text on your disk. A closed window, a lost session or a new laptop changes nothing, and another AI model or a teammate can open it and keep going.

Your AI reads the file when it starts and writes to it as it works.

One command installs the habit, so it happens without you asking.

Ask where you fixed something before and it finds the entry across every project you ran.

It is a plain file: it survives a closed window, a lost session and a new laptop.

Hand it to another AI model or a teammate and they pick up exactly where you left off.

In Claude Code, before your AI changes a file it is shown the journal of that part, so it cannot skip what was decided.

A status written in the journal (paused, verified, out of credit) is flagged to be checked live before your AI repeats it.

Open every journal in Obsidian with a map: a start page, a table of all of them and a linked graph.

Ask in your own words or in another language: a small model on your own computer finds the entry that means the same thing.

When a later entry changes a decision or its price, the old one is flagged first, so your AI never repeats it as current.

Before your AI rebuilds something a shipped entry already did, it is shown that entry and stopped once.

Under your prompt it says active, and one plain sentence only when something real happens: it found your earlier decision, it saved what was done.

An old rule that a newer entry changed is shown with the newer word next to it, so your AI never follows it as current.

Optional: before your AI hands you a reply, it is checked against what you already decided and how your tools really work, and sent back to fix it if it breaks either.