← Back to all free guides
A free guide by Mariah Brunner ↗ INSTAGRAM ↗ TIKTOK

The Fable 5
Usage Limit Fix

Stop burning your usage limit on your best model. Build a loop in Claude Code where your top model plans and judges, and a cheaper, faster model does the grunt work.

This works with whatever your top model is, Fable 5.1 or Opus 5.5. Either way, it’s too powerful to waste on tiny tasks. Use it as the planner and judge, and let Haiku 5.5 or Sonnet 5.5 do the labor.

Why It Matters For Fable

On Pro, Fable models aren’t included in your plan’s limits, so every request bills to usage credits. On Max and Ultra you can use up to 50% of your weekly limit on Fable, then it’s credits or another model. (Fable on your plan)

Build The Loop

01

Set your top model as the planner

In Claude Code, run /model fable for Fable 5.1 (or /model opus for Opus 5.5). Set /effort high for planning. This model writes the plan and reviews the result. It doesn’t do the grunt work.

02

Create a cheap executor subagent

Ask Claude to create one for you, or save this file at .claude/agents/executor.md in your project. The model line is what keeps it cheap. (Subagent docs)

.claude/agents/executor.md
--- name: executor description: Implements a written plan step by step. Use for all coding, editing and file changes once a plan exists. model: haiku --- You are the executor. You implement exactly what the plan says. - Work through the plan in order. Don't redesign it or add features it doesn't ask for. - After each step, run the relevant check (tests, build, lint) if one exists. - If a step is unclear, blocked, or looks wrong, stop and report it instead of guessing. - When you finish, return a short report: steps done, files changed, checks run and their results, anything skipped and why.

Swap haiku for sonnet when the work is harder.

03

Run the loop

Now your top model plans, hands the work to the executor and reviews it. Kick it off with this.

Plan, Delegate, Review
Task: [describe what you want built or changed] Done means: [how we'll know it worked, e.g. "the signup form saves to the database and all tests pass"] 1. Read the relevant files and write a numbered implementation plan. Keep each step small enough for a cheaper model to do without judgment calls. Don't build anything yet. 2. Show me the plan and wait for my OK. 3. Hand the approved plan to the executor subagent to implement. 4. When it reports back, review the result yourself against the plan and the "done means" line. Run the checks. 5. Send any fixes back to the executor with specific instructions. Only fix things yourself if they need real judgment. 6. Finish with a short summary: what changed, what you checked, and anything I should look at.

Your expensive model only spends tokens on the strategy and the quality check. The cheap model does the heavy lifting.

Make It Run Itself

01

/goal

Sets a finish line. Claude keeps working across turns until the condition is met, and after every turn a separate small, fast model checks whether it’s done. It judges from the conversation, so write a condition Claude can show, like /goal all tests in the tests folder pass, or stop after 20 turns. (Goal docs)

02

/loop and /schedule

/loop re-runs a prompt on an interval while your session stays open. /schedule sets up a cloud scheduled task, so it keeps running when your laptop is closed.

The Shortcut: /advisor

Claude Code also has a built-in version of this idea. Run your session on a cheaper model, then run /advisor fable (or /advisor opus). Claude checks in with the stronger model at key moments, like before picking an approach, when it’s stuck, or before calling the job done. Anthropic says this typically costs less than running the strong model the whole time. (Advisor docs)

Good To Know

The advisor is experimental and works on the Anthropic API only. The advisor has to be at least as strong as your main model, and on Pro a Fable advisor bills to usage credits.

Bonus Savers

Use /effort low for simple steps, /clear between unrelated tasks, and put standing context in a CLAUDE.md file so you’re not re-explaining yourself. Run /usage to see where your usage went.