The fable-goal skill: turn a ramble into one great goal prompt

The prompt is the bottleneck on a long autonomous run. You describe the thing you want in fragments, out loud, half of it in the wrong order, and what comes back is a decent version of a vague ask. The model isn't weak. Nothing in the prompt told it what excellent looked like or gave it a way to check its own work.

fable-goal fixes the prompt, not the build. You hand it your ramble and it writes one prompt back. Fable is Anthropic's most capable generally available Claude model, the one you point at a long job you want run start to finish, and a goal prompt is how you hand that session a single outcome and walk off instead of sitting there answering questions. The whole idea behind the prompt it writes is to get out of the model's way: say what you want clearly, hand over the tools, grant real creative freedom, then demand it check its own work before calling anything done.

It's one of the 55 free skills in our freeskills repo, and it's the one people run right before they walk away from the laptop for two hours.

Three lines. Clone the repo, make sure your skills folder exists, copy this folder into it.

Start a new Claude Code session and the skill is live. That's the install. No API keys and nothing to configure.

Type /fable-goal and then just talk. Voice-to-text is fine and half-sentences are fine. The skill pulls out what your ramble already contains: the deliverable, the quantity, who it's for, any tools you named, the quality bar you implied, and where the results should end up.

Then it fills the small gaps itself. It only asks when the answer would genuinely change the prompt, and when it does ask, it asks everything in one batch instead of interviewing you in rounds. Before it names any file path, skill or tool in the prompt, it checks that thing actually exists on your machine, because a prompt pointing at something imaginary sends the fresh session on a dead-end hunt.

What lands is one copy-ready block of about 150 to 350 words with a short list of assumptions underneath. Nothing gets saved to a file and nothing gets built. Copy the block, start a fresh Claude Code session with Fable selected as your model, type /model if you need to switch, paste it, and leave it alone. If an assumption is wrong, correct that one line instead of re-rambling the whole idea.

Say it the way you'd say it to a friend. Don't tidy it up first, that's the skill's job.

The self-check loop is what separates a good run from a rough draft. Tell it what checking means for your medium.

Naming a couple of real tools raises the ceiling. Naming ones you don't have wastes the run, so let it verify before it writes them in.

Here is what this skill does with a ramble and why the prompt it writes runs better than yours.

1. 1Hook: there's a brand new free Claude skill that takes you from rambling about a thing you want built to one paste-ready prompt that runs the whole job while you're away from the laptop.
2. 2Mechanism: you ramble about the thing you want built and it pulls out the deliverable, how many, the quality bar you implied, and where the finished work should land. It fills the small gaps itself and only asks when the answer would genuinely change the prompt. What you get back is one block of prose that hands the job over, grants freedom on how it gets done, and demands three self-checking passes before anything is called finished.
3. 3Transformation: you stop babysitting long builds. You describe the outcome once, walk away, and come back to something that already checked itself before it called done.
4. 4Close: it's free and it's one of 55 in the repo. This one writes the prompt for you. If you want the path that tells you which jobs are worth handing over in the first place, Claude Code Club is $9 a month. Link below.

Take the idea you've been describing to people for a week, run it through the skill tonight, and paste the block into a fresh session before you go to bed.