I Built Claude's "Auto Dream" Skill for Authors (And It's Free)

This is one of the worst things about using Claude for writing a long book.

Every session, Claude's gonna forget your manuscript every single time.

Now Anthropic just shipped a fix.

It's called Auto Dream.

Claude consolidates its memory between sessions.

But Auto Dream only works on their agent platform, and most of us authors don't have access to that.

So I taught Claude to dream about our books.

Here's how it works.

you drop a skill file into your book's folder.

You can run it manually inside Claude AI with no setup, or you can run a small script that does the same thing automatically every night while you sleep.

Either way, the skill reads your manuscript your notes and a memory file it's been building over time.

It pulls out everything durable, your character decisions your style preferences, your corrections, the rules you've made about your world.

It merges what's new with what it already knew.

Then it writes a clean memory file that the AI loads at the start of every future session.

You wake up, your AI knows your book.

That's the gift.

That's what every long-form writer needs and almost nobody has.

Now to be clear, this is not a port or clone of Anthropic's Auto Dream.

Theirs runs on Claude's memory layer.

Mine runs at the project's memory layer.

Theirs is automatic for the AI agent.

Mine is automatic for your book.

They're not competing.

If Anthropic ever opens up Auto Dream for general use, and they probably will, you might wanna use both.

They're not contradictory, they're complementary.

Let's see this thing working.

What you're seeing on screen is a sample project that I built that comes with the free bundle.

We're about to run the dreaming skill on this folder for the first time, and there are three ways to do it.

I'll show you the easiest first.

Method one is a manual mode inside of Claude AI.

You open a fresh Claude AI session, drag in Skill MD from the bundle, drag in the manuscript and notes file, And then type the prompt on screen.

You hit send, and that's it.

The skill's gonna run four phases.

Inventory, that's what files are there.

Signal extraction, what's worth remembering.

Consolidation, merging the new stuff with what it already knew, and writing the outputs, a memory file the AI will load into every future session.

Plus, a dreamlog entry that tells you what changed.

Notice how it's pulling things from the memory file preferences corrections.

One of my favorite parts is this open questions section that the AI's gonna flag for you automatically.

It's even gonna show you these Chekhov's guns sections.

These are open loops or open threads that you'll need to close down eventually.

Then the dreamlog entry that shows you every single thing it learned.

And notice its dated structure.

That's manual mode, about forty-five seconds of work.

Now method two, the launcher.

On Mac, it's a file called dream.command.

dream.command.

On Windows, it's dreams.bat.

You double-click it.

The first time, the launcher checks if Python is installed.

It installs a small library it needs and asks you for your Anthropic API key.

It's only gonna do that one time.

Every time after that, it just runs the pass.

Same outputs as manual mode, memory file the AI loads, dreamlog entry telling you what changed.

I can't tell you how much I love this dreamlog.

It's like a journal of what your AI is learning about your book.

Here's the thing that makes the skill feel exactly like AutoDream.

You can put that launcher on a schedule.

That's a cron job in Macs or Linux or a task scheduler on Windows.

3:00 AM every night, while you sleep, the script runs on your books folder, dreams on it, updates the memory file, writes a new dream, and writes a new dream log entry.

You wake up the next morning, and there's a new dream log waiting for you.

Open it, you skim what the AI learned overnight and you start writing.

It's cool because the AI already knows your book the way it did when you stopped working yesterday, plus whatever it figured out from the overnight pass.

You can even take this to the next level by giving this author dream skill to an AI agent like Author Claw, Open Claw, Hermes, or whatever super powerful autonomous, definitely not going to end human civilization digital god someone vibe codes in their parents' basement next.

Simply hand your agent this skill like you would any other skill, and voila, your agent can now dream about your book.

If you're writing short stories or just anything short, then you probably don't need this.

You can hold all that information in your head.

And you can re-explain it to the AI in two paragraphs in a new session.

But if you're writing something longer, say eighty thousand words or more, a trilogy, a series with continuity between the books, a memoir spanning decades, or a nonfiction book with deep research notes, that's where this becomes a tool that you actually need.

Because at that scale, most AIs are gonna start contradicting things that you set weeks ago.

It will reference subplots that you already cut.

It may even name characters wrong.

It ignores preferences that you stated months ago.

And none of that is AI being stupid.

It's that you never gave it a reliable place to put what it learned about your book.

The dreaming skill is that place.

Here's the deal.

Free download, it's on Ko-Fi, link's in the description.

No email gate, no upsell, just the skill, the script, launcher files for almost any operating system.

Integration guides for Claude AI or AI agent of your choice.

Even a sample project so you can test it on that before trusting it with your own book.

Of course, if you want deeper paid tools like Author Swarm, Author OS, and a lot more, those are on my Ko-Fi.

Check those out.

Now, I'm giving away this for free on my Ko-Fi because Anthropic's gonna release it to everyone eventually.

At least I think they will.

And until then, though, I still want you to have access, and I want access too.

I've already added this skill to my own projects.

Now it's your turn.

Try it on a project you've been working on for a while.

Run one pass.

You'll know in five minutes if this is a tool that you'll actually use and if it's worth keeping.

I'm building all kinds of fun, cool things on Ko-Fi.

I'm updating tools that I've already created, and if you like that sort of thing, go ahead and like the video.

If this is the kind of content you like to watch, then please subscribe.

And if you have stayed around to this part of the video, thank you very much, because that really helps me out, and it really helps the channel out, too.

Thank you so much for watching this video.

I have a whole playlist with these kinds of tests and tools that I've built.

Check that out.

I think the playlist will be right here on screen.

I'll see you in the next one.