I built this because Claude Fable 5 is the most capable model and also the most expensive. Running everything on it adds up fast. This skill makes Fable the architect and hands the token-heavy work to Opus 4.8.
The loop is simple. Opus 4.8 researches, Fable plans, Opus 4.8 builds, and Opus 4.8 checks the work. Fable only touches the planning step, so you pay its rate only for the thinking that needs it.
What it does
- Gives two ways to set it up: advisor mode, where Opus writes and Fable steps in when it's stuck, or architect-and-delegate, where Fable plans and parallel Opus 4.8 agents build.
- Treats the /effort setting as the cost lever and says to cap it at high.
- Includes 9 ready-to-paste instruction blocks, covering things like acting without overplanning, delegating, verifying and setting boundaries.
- Lists the mistakes that quietly reroute Fable to Opus, such as asking it to explain its reasoning in the response.
- Covers running GPT-5.6 Sol through Codex as a lower-cost architect, alone or alongside Fable.
How it works
- Opus 4.8 does the research.
- Fable takes the results and writes the plan.
- Opus 4.8 agents carry out the plan.
- Opus 4.8 verifies the work against the spec.
What you need
- Claude Code
- Optional, for the Sol setup: the Codex CLI, or a ChatGPT plan or OpenAI API key plus Node 18.18+ for the Codex plugin
Install it
mkdir -p ~/.claude/skills/fable-orchestration
curl -sL https://raw.githubusercontent.com/per-simmons/fable-orchestration/master/SKILL.md -o ~/.claude/skills/fable-orchestration/SKILL.mdThat installs it for every project. For one project only, put SKILL.md in that project’s .claude/skills/fable-orchestration/ instead. Claude Code loads it automatically.
The full instructions, and the skill itself, are on GitHub.
Try asking
- “Set up advisor mode with Fable for this build.”
- “Make Fable the architect and have Opus 4.8 agents build the plan in parallel.”
- “Check my system prompt for anything that would reroute Fable to Opus.”
Good to know
- The skill advises against xhigh and max effort on Fable, because it tends to over-reason and cost more.
- Some of its advice, like the prompt that trips the reroute to Opus, depends on how the models behave now and may change.
Written by
Pat Simmons. Pat tests AI models and tools on the AI for Mortals YouTube channel.
