Fable 5.2 Prompting: Delete ‘Always’ and ‘Never’

I used to think the best AI prompts were strict ones. More rules meant more control, or so I assumed. Then I read a breakdown of Anthropic’s official Fable 5.2 prompting guide that turned that idea upside down.

The author went through the whole guide and boiled it down to three things: what to delete, what to replace, and one prompt to copy. I read the breakdown twice. Most of us are probably working against Fable without knowing it.

The main idea is simple. Fable follows your words as written. That sounds great until you notice how loosely most of us write instructions. Here’s the process the expert lays out, one step at a time.

✦ Step 1: Delete the wrong rules you’ve given Claude

Start with what’s already sitting in your setup. Most of us have instructions and memory entries we wrote months ago and never looked at again. The original poster suggests a quick cleanup:

1. Go to your Claude instructions and memory.
2. Hunt for these 2 words: “always” and “never”.
3. Replace those words with “ask me first”.

Here’s the before and after the creator shared:

Old: “Never change my prices.”
New: “Ask me before you delete anything.”

And here’s the cleaned-up instruction block from the post:

Finish the whole job before you stop. Don’t offer to continue: continue.
Ask me before you delete anything.

Why this step matters: Fable obeys “never” to the letter, even after you say yes. So if you once told it to never touch something and later you actually want it changed, the old rule can still block it. “Ask me first” keeps you in control without locking the door for good.

The post’s author flags a second trap: style rules like “keep every sentence under 15 words.” They feel tidy, but they can make the writing sound like AI. As the author points out, no one actually writes like that. If your memory has rigid formatting rules, this is a good time to loosen them.

✦ Step 2: Say exactly what Claude should change

Once the old rules are gone, look at how you give new tasks. The expert’s formula has three parts:

1. State the goal.
2. Give it the files and context it needs.
3. Say what it can change.

Old: “Go.”
New: “Yes, rewrite the 3 welcome emails. Leave the invoice template alone.”

Why this step matters: “Go” isn’t enough. Fable applies your words as written, so a vague green light gets a vague result. The better version names the exact scope (3 welcome emails) and the exact boundary (the invoice template). Nothing is left to guess.

I like this because it works anywhere. Updating a landing page? Name the sections it can touch and the ones it can’t. Cleaning up a spreadsheet? Say which columns are fair game. That one extra sentence saves a lot of cleanup later.

✦ Step 3: Tell Claude to finish the entire job

Now you have clear rules and a clear scope. The next problem is the half-finished answer that ends with “Want me to continue?” The creator’s fix is to add this to your prompt:

“Finish the whole job before you stop. Don’t offer to continue: continue. Ask me before you delete anything.”

Then ask for a report you can actually read:

“End with a recap for someone who saw none of the work: key points first, plain words.”

Why this step matters: The first line keeps Fable working until the task is really done. The second keeps one safety check in place. The recap line means you get a summary you can skim in a minute instead of digging through a wall of output.

And lastly, the author reminds us to check what changed before you use it. Even a well-prompted model needs a quick human review.

✦ Step 4: Put it all together with a plan and a handoff

This is the part I found most useful. The original poster describes a workflow that uses Fable for planning and Opus 5.5 for building:

1. Write a planning prompt. Fill in your goal, context and limits.
2. Let Fable plan the project. Don’t ask it to build yet.
3. Still on Fable, ask it to write a handoff brief for the project.
4. Open a NEW chat on Opus 5.5. Attach the brief. Say: “Take it from there.”

Why this works: Fable handles the thinking-heavy planning, which is where it shines. Then a fresh Opus 5.5 chat does the execution with a clean, complete brief. You also save your Fable limit for the work that really needs it.

PS: Don’t switch models inside the same chat. Opus can’t read Fable’s thinking.

That tip is easy to miss and important. If you switch models halfway through a chat, Opus loses Fable’s reasoning. A separate chat with a written brief means nothing gets lost.

✦ A quick bad vs. better comparison

The expert wraps it up with a contrast that sums up the whole approach:

BAD PROMPT: “Fix my email. Don’t change anything else. Go.”

BETTER PROMPT: the task + what you approve + what needs your permission.

Here’s how I’d put that into practice. Instead of “Fix my email,” try something like: “Rewrite the subject line and first paragraph of this email to sound warmer. You can change wording and order. Ask me before you remove any links or the pricing section.” Same task, but now Fable knows the goal, the freedom, and the limits.

✦ Your quick-start checklist

• Audit your memory: Search for “always” and “never” and swap in “ask me first”.
• Drop rigid style rules: Word-count limits per sentence can make writing sound robotic.
• Scope every task: Goal, context, what can change.
• Add the finish-the-job line: Let it continue instead of offering to.
• Ask for a plain-words recap: Key points first.
• Plan on Fable, build on Opus: Use a new chat and a written brief.
• Review before you ship: Always check what changed.

What I like most is that none of this is complicated. It’s about writing what you mean, because Fable takes you at your word. Check out the full LinkedIn post for the author’s complete breakdown and examples.