10 Things to Do With Fable Before It Leaves Your Plan
The exact prompts for all ten — plus one correction — before the window closes on Sunday 12 July
The Window
Claude Fable 5 — the smartest model Anthropic have ever shipped — is currently included in Pro, Max and Team plans. The included window was due to close on 7 July; it's been extended, and now runs until Sunday 12 July. After that it moves to metered usage credits (around $10 per million tokens in, $50 out), paid on top of your subscription. It's not vanishing — but the included window closes, and that's exactly why these few days are worth emptying.
The list below gets increasingly niche on purpose. The first few are for anyone with a project; the last few are the quiet winners — because plans, frameworks and claude.md files written by the smartest model keep working long after the window shuts.
The Ten, With Prompts
Run /security-review — on Opus, not Fable (a correction)
This list originally said run it with Fable. I was wrong: Fable ships with tighter guardrails around security work, and it blocked my review mid run. The command is the same, the model isn't — switch to Opus first, and save Fable's window for the other nine.
/model opus
/security-review
Then: "Fix everything you found, highest severity first,
and show me each change before you make it."The thing you gave up optimising
The slow page, the flaky build, the query you made peace with. One last proper go, from the model most likely to crack it.
This is slow and I've never managed to fix it: [describe the
slow page / build / query]. Profile it, find the real
bottleneck, and fix it. Prove the improvement with before
and after numbers.The feature stuck in the backlog
The one that's defeated every previous attempt. Plan first, build second, no guessing.
Here's a feature I've never managed to deliver: [what it
should do, and where past attempts stalled]. Plan it first
and show me the plan. Then build it end to end - and flag
anything you're unsure about instead of guessing./deep-research the decision you've been putting off
The market, the pricing, the "should I even build this." A report is only as good as the model that wrote it.
/deep-research Should I [the decision]? Cover the market,
the competition, realistic pricing, and the strongest
argument AGAINST doing it. Cite sources.Your business idea, judged brutally
You want the truth from the smartest model in the room, not the most agreeable one.
Here's my business idea: [the idea]. Judge it brutally -
no encouragement, no softening. Where does it fail? Who
beats me to it? Why might customers not pay? End with the
single biggest risk and whether you'd bet on it.Ultracode, once
A fleet of Fables sweeping your whole codebase at the same time - the months-overdue cleanup, done in an afternoon. It deliberately burns tokens; that's rather the point this weekend.
ultracode - sweep this whole codebase: dead code,
inconsistencies, bugs, out-of-date docs. Fix what's safe
to fix, and give me a list of everything that needs my
call. Verify your changes as you go.Your next big project, planned
The smartest model does the thinking; the cheaper ones follow the plan for months afterwards. Wrong plans are the expensive kind.
Plan [the project] in full: architecture, milestones,
risks, and step-by-step tasks written so a less capable
model can execute each one without extra context.
Interview me first about anything unclear, then write
the whole thing to PLAN.md.The landing page glow-up
Animations, hero, colour scheme. Taste is exactly what you want from the best model you can get.
Redesign my landing page hero: a modern animation, a
colour scheme built around [your brand colour], and make
the buttons feel expensive. Show me three directions
before you build anything.Your own testing framework
The niche one. Whatever you make or ship repeatedly, have it build the system that decides what you do next - with predictions you can score.
Build me a testing framework for my [content / product /
work]: for every piece, a hypothesis, a prediction, and a
review step that scores the result against it. Set it up
as files in this project so I can keep running it.Restructure your claude.md
The one that actually matters. The rules every future session follows, written by the smartest model while you've still got it. The meter starts on the 13th - what it writes down, you keep.
Rewrite my claude.md from scratch: how this project is
structured, the rules every future session should follow,
what good looks like here, and the mistakes to avoid.
Interview me first - ask about anything you can't infer
from the code. Make it the file you'd want to inherit.Don't Burn It on Small Edits — The Handoff Line
The flip side of the list: renames, one-line fixes and quick lookups don't need the biggest model, and Fable eats your usage limit fastest. You don't have to police it yourself — one line in your claude.md tells Fable to hand the small stuff to a smaller model and keep its own effort for the thinking.
For small mechanical tasks (renames, one-line fixes,
simple lookups), hand the work to a subagent on a smaller
model instead of doing it yourself. Save your effort for
the thinking.The Honest Bit
Fable isn't being switched off — after the 12th it's available on usage credits, and some enterprise plans keep it included. This list isn't "use it or lose it forever"; it's "the included window closes, so spend it on the things that benefit most from the extra intelligence."
And the included allowance is part of your plan's weekly limits, not unlimited fuel — so if you only run a few of these, run the last four. Plans, frameworks and claude.md files are the ones that keep paying you back after the meter starts.