Title option 1
3 prompts to run on Fable 5 before July 7
Title option 2
Do this with the best model before it gets expensive
The best model writes the instructions. Cheaper models run them for months.
Fable 5 is the most powerful model Anthropic has ever put inside a normal plan. Right now it's included. On 7 July it moves to metered usage credits, roughly double what Opus 4.8 costs.
So don't just chat with it. Make it write things down that keep working after it's gone.
Three prompts. Copy, paste, run them before the 7th.
Why this window matters
Fable 5's lead over Opus 4.8 gets bigger the longer and harder the task. That means the best model writes the most detailed, most reliable instructions.
Instructions are a file. Save the file and a cheaper model runs it for months. I did this across my own skills library last week. A bug that had blocked me for weeks got solved in one pass.
You have until 7 July on your current plan. Use it on the three things below.
The 3 prompts
Paste these into Fable 5 (claude-fable-5). Swap anything in [brackets] for your own detail.
PROMPT 1 — AUDIT
Point Fable 5 at your skills, prompts, or SOPs and have it make them better once, so a cheaper model runs them correctly forever.
You are Claude Fable 5, the strongest model I have until 7 July. Audit and rewrite the skills and workflows in [PASTE FOLDER OR PASTE THE FILES].
For each one:
1. Read it in full and work out its real job: what it should make a weaker model do reliably, with no help from me.
2. Score it 1-10 on clarity, specificity of steps, error handling, and how well Opus 4.8 could follow it cold.
3. Rewrite it so a weaker model gets it right first time. Add the missing edge cases, the exact commands, input and output examples, and a short "done when..." checklist. Detailed and unambiguous beats short and elegant.
4. Keep the original intent, format and filename. Save the rewrite over the file, or to [file].v2 if you want me to diff it first.
Work through the whole set. After each file give me one line: filename, old score, new score, biggest single fix. Start with the ones I use most; if you can't tell, go alphabetically.
PROMPT 2 — SOLVE
For the bug you keep circling. First export the whole thread to a single markdown or text file (use your tool's export option, or select all and paste into one file). Then start a fresh chat or project and paste this with the export attached.
You are Claude Fable 5. Below (or attached) is the full history of a problem I've been stuck on for [HOW LONG]. Everything I've tried is in there.
1. Read all of it before you respond. Don't skim.
2. Restate the exact problem in one paragraph and list what's already been tried and ruled out, so I know you actually read it.
3. Find the root cause, not the symptom. If the history doesn't give you enough to be sure, tell me exactly what to run or paste to confirm before you guess.
4. Give me the fix as concrete steps or a diff I can apply now. Then tell me how to check it worked and the one thing most likely to break next.
Assume I've been circling this for weeks and missed something obvious. Find it.
PROMPT 3 — BUILD
The 90-day plan, the content calendar, the roadmap. Fable 5 writes it, then a cheaper model executes it week after week.
You are Claude Fable 5. Build me a [90-day content calendar / product roadmap / GTM plan] for [ONE SENTENCE: what it's for, who it serves, the goal and the deadline].
Context: [paste your situation: audience, offer, constraints, what's worked, what hasn't].
Deliver:
1. The end goal as a measurable outcome with a date.
2. A week-by-week plan. Each week: the objective, the specific tasks, and what "done" looks like.
3. For every recurring task, write the exact prompt or checklist a cheaper model (Opus 4.8 or Sonnet) can run to do it without you, because you won't always be in the room.
4. The 3 assumptions most likely to be wrong and the early metric that flags each one.
Make it a standalone document I can hand to a weaker model and it still runs right. Detailed over pretty.
Why it still works after Fable 5 gets pricey
You're not renting the model, you're keeping what it writes. The rewritten skills, the fix, the plan are all files. Opus 4.8 runs them at a fifth of Fable 5's output cost.
Fable 5 is $10 in / $50 out per million tokens from 7 July. Opus 4.8 is $5 / $25. You pay the top model once for the thinking, then run the output cheaply for months.
If you only do one
Do Prompt 1. Auditing what you've already built is the fastest win, because you keep every rewrite.
Prompt 2 if you've got a specific blocker bleeding time right now.
Reply and tell me which one you ran and what it fixed. I read every reply.
If you'd rather someone built these systems into your business for you, that's what we do at OptiMAX (optimax-ai.com).
P.S. Systems built for you, not by you: optimax-ai.com