Your AI project remembers. Do your projects learn from each other?

Single-project AI memory is getting solved — sessions that resume, context that survives. The question
that stays open: when you start project #2, does anything project #1 learned come along?

For most setups, the honest answer is no. The conventions you settled, the failure you diagnosed late
at night, the configuration that turned out to matter — all of it lives where it was learned. Project
two starts from zero.

Files make knowledge portable

CRAFT for Claude Cowork (a free-during-beta framework that runs on Claude Cowork — you'll need Cowork
available in your Claude plan; CRAFT adds no cost, account, or gate on top) keeps a project's working
knowledge in plain, versioned text files. That design choice has a consequence worth noticing: if state
lives in files, state is portable.

The framework ships a cross-project transfer registry — recipe-governed. One project packages what it
knows into a transfer file. Another receives it, records where it came from, and folds it into its own
working files. The registry keeps the accounting: what went out, what came in, from where. Inheritance
is auditable, not folklore.

A real transfer, condensed

While the source post was being written, one of our projects exported its metrics-automation setup to a
newer project: the task definitions, the exact formats, ten hard-won lessons — one file, delivered to
the receiving project's inbound folder, registered on receipt. Anything the records didn't hold was
marked NOT RECORDED rather than guessed. The whole move took an afternoon.

The mechanism, in three moves

1. Export — ask Claude to package what the new project should inherit; you name what moves.
2. Deliver — one file, dragged into the new project's transfers/inbound/pending/ folder.
3. Receive — registration is automatic: the registry records what arrived and where it came from.

No platform feature, no export API, no permission needed — moving knowledge is moving a file, and
auditing an inheritance is reading one.

Honest bounds

CRAFT for Claude Cowork is in beta. Transfer is a deliberate act with a recipe and a record — not
ambient sync; nothing moves unless you move it. AI can make mistakes: review what you inherit. This
article reflects the framework as it is today.

The full walkthrough — including the real (lightly condensed and redacted) export lines — is in the
original post: https://craftframework.ai/your-project-remembers-do-your-projects-learn-from-each-other/
· Download: https://craftframework.ai/craft-downloads/