Fable's Free Access Is Ending, So I Asked It to Write Down Everything It Knows

Fable's free access is ending soon.. it's moving to API-only pricing, and it's the best engineering model I've had access to. So before that happens, I did the obvious thing: I asked it to write down everything it knows.

The prompt wasn't fancy. Cleaned up a little, this is what I actually told it:

I want you to think like you're going to retire. Because you're retiring, you want to give away all your amazing knowledge. You're an amazing distinguished engineer, and you're also a top 1% product thinking expert.. product knowledge, design, UI/UX, system design, all of it. If you had to retire and you wanted your knowledge to stay behind so other people could use it, write a document and categorize it — kind of like a document of skills. Just write a bunch of skills that can be used based on your knowledge.

What came back is a 50-skill playbook covering nine areas: engineering craft, system design, product thinking, UI/UX, metrics and experimentation, operations, decision-making, working with people, and a meta section on when to ignore the rest. Every skill has the same shape.. what it is, when to use it, and the heuristics it says took years to learn.

Then I told it to take each lesson and turn the whole thing into a Claude skill. Same content, rewritten as instructions for a model to execute: every practice gets an "Apply when" trigger and imperative rules. You can't download Fable's weights, but you can capture how it behaves in a form another model can load. Drop the skill into Claude Code and now Opus 4.8 reviews your code guided by Fable's lessons. Distillation, except into prompt space instead of weights.

Both versions are free:

Download the playbook (markdown)  Download the Claude skill

The markdown is for reading. The .skill file loads into Claude Code or claude.ai. Both also live on github.com/dimichoueiry/playbooks.

The first ten skills, so you can get a feel for it:

1. Reading code before writing code
2. Naming things
3. Managing complexity (the only real enemy)
4. Writing code that's easy to delete
5. Testing strategy
6. Debugging systematically
7. Code review as leverage
8. Refactoring without breaking the world
9. Technical debt as a financial instrument
10. Starting from the data model

..and 40 more. A few lines from inside, verbatim:

Prefer visible duplication over the wrong abstraction — duplicates are easy to merge later; wrong abstractions metastasize.

The bug is in the thing you're sure about. ("It can't be DNS" — it's DNS.)

Mark every comment as blocking: or nit: explicitly. Never leave severity ambiguous.

To install the skill: unzip it into ~/.claude/skills/ for Claude Code, or upload the .skill file directly on claude.ai. If you're in Claude Code, this one-liner does it:

Or skip the terminal entirely and make Claude do it. Copy this prompt, paste it into Claude Code:

It'll fetch the repo and set itself up.. fitting, given what the file is.

Then ask for a code review, an architecture take, whatever.. it names the principle it's applying and grounds the critique in your actual code.

Hope it helps.