Seventy-one consecutive AI sessions. Zero data loss between any of them. That’s not a benchmark from a research lab — it’s what happened when a structured framework met Claude’s Cowork desktop platform and refused to let the conversation end.
Today, the CRAFT Framework — Configurable Reusable AI Framework Technology — is available as a free public beta on GitHub. It’s the result of months of development that transformed Claude Cowork from a stateless chat tool into a persistent, version-controlled, quality-assured working environment.
This post introduces what CRAFT does and why it matters. Over the next eight weeks, we’ll go deep on each capability — one per week — so you can see exactly how each one works and decide which ones solve problems you’re already dealing with.
The Problem Nobody Talks About
Every Cowork user knows the feeling. You’ve spent an hour getting Claude up to speed on your project — the context, the constraints, the decisions you’ve already made. The session ends, and tomorrow you start from scratch. Your AI has no memory of yesterday.
This isn’t a minor inconvenience. It’s a structural limitation that compounds with every session. Project context evaporates. Naming conventions drift. Decisions get remade. Files can vanish when the sandbox resets. The AI produces confident output whether or not it’s accurate, and there’s no built-in mechanism to catch errors before they reach your deliverables.
If you’re running a multi-session project, you’re managing context restoration, file backup, quality checks, naming consistency, and decision tracking manually. These workarounds cost time, introduce errors, and scale poorly.
CRAFT was built to eliminate these problems entirely — not with a single feature, but with an integrated system that handles all of them as a unified workflow.
Eight Capabilities That Change How You Work with AI
CRAFT is a structured communication framework that sits between you and your AI. It standardizes how sessions initialize, how context persists, how quality gets verified, and how your work stays safe. Here’s what it covers — and what we’ll explore in depth over the coming weeks.
You’re reading it. CRAFT is a complete framework for turning Claude Cowork into a persistent working environment. It’s free, it’s on GitHub, and it was built using itself — 71 handoffs, zero data loss. The development process is the proof of concept.
At the end of every session, CRAFT captures a structured handoff snapshot: priorities, decisions, persona state, unfinished work, and contextual notes. When the next session starts, the initialization recipe reads the handoff and restores full context. Your AI picks up exactly where it left off — across crashes, compaction events, and even machine switches.
Cowork’s sandbox resets between sessions. Files vanish if they aren’t saved deliberately. CRAFT integrates git version control directly into AI workflows — every significant change triggers an automatic checkpoint tied to the recipe that caused it. Your work survives because it’s committed before the sandbox disappears.
CRAFT’s device-switch handoff lets you move a project between machines without losing context. The handoff captures everything the new session needs to continue seamlessly — project state, file locations, decision history — so switching devices doesn’t mean starting over.
AI produces confident output whether it’s correct or not. CRAFT’s QA framework runs verification gates on every output — checking source traceability, detecting reconstructed-from-memory data, flagging constraint violations, and surfacing untested assumptions. During CRAFT’s own development, this pipeline caught a licensing error that had propagated across 9 content files.
CRAFT provides a standardized project structure that organizes your work from the first session. Folder conventions, naming rules, and file placement are all handled by recipes — so you spend time on your actual work instead of deciding where to put things. Nine projects were consolidated into a single governed repository during development.
Brand guidelines are useless if the AI ignores them. CRAFT’s Brand Identity cookbook encodes your visual identity, tone, and messaging rules directly into the framework — so every piece of content the AI produces is on-brand by default, not by accident.
CRAFT tracks every recipe version, maintains changelogs, and provides update-checking tools so your framework never falls behind. When recipes improve, you know about it. When breaking changes land, you’re warned. The versioning system kept hundreds of files synchronized across months of active development.
Built With Itself
The strongest evidence for CRAFT is that CRAFT was built using CRAFT. The framework’s own handoff system maintained continuity across 71 consecutive sessions. Its own QA recipes caught errors in its own promotional content. Its own git integration protected its own source files through sandbox resets. Its own versioning system tracked its own recipe evolution.
If you want to know whether it works for real projects, the answer is that the framework you’re reading about was produced with it.
Each week, we’ll go deep on one capability — how it works, what problems it solves, and how to start using it in your own Cowork sessions. Follow along to see what a structured AI working environment actually looks like in practice.
