Guide · Building with Fable 5.1
Give it somewhere to remember things
Fable 5.1 works notably better with a place to write down what it learns.
One of the more surprising findings about Claude Fable 5.1: it performs measurably better when it has somewhere to record learnings for later — even just a Markdown file in your project. Tell it where, tell it to read the file at the start of each session, and give it a format.
Why it helps
Every session starts from nothing. Corrections you gave last Tuesday, the approach that didn't work, the reason a config value is odd — all gone unless written down. A memory file turns a series of disconnected sessions into something cumulative.
This is the same idea as a project brief, but maintained by the model as well as by you.
The setup
Create a NOTES.md or a notes/ folder, and put this in your project instructions:
You have a memory directory at ./notes/. Read it at the start of each session before doing anything else. Store one lesson per file with a one-line summary at the top. Record corrections I give you and approaches that worked, including why they mattered. Don't save what the repo or the chat already records. Update an existing note rather than creating a duplicate. Delete notes that turn out to be wrong.
That's it. The model will start writing to it when it learns something, and reading from it when it starts.
What ends up in there
Things you corrected: "Don't use the ORM for reports — the raw SQL in reports.php is deliberate." Things it discovered: "The test suite needs the dev server running first." Decisions and their reasons: "Chose SQLite over Postgres for this project because there's a single user and no server."
All the things that live in your head and cost twenty minutes to re-explain.
Keep it honest
Review the notes occasionally. Delete anything stale. A wrong note is worse than no note, because the model will trust it. If you find one, correct it in the file rather than in the chat — otherwise you'll correct it again next session.
In Claude Code specifically
The tool already reads a CLAUDE.md at startup. Use that for the stable facts — how to run the project, conventions, what's unusual. Use a separate notes directory for the accumulating lessons, so the stable file doesn't grow into a mess.
Anything you've explained twice belongs in the file. That's the whole rule.