Efficient Fable Orchestrator — AI skill for Claude Code

efficient-fable-orchestrator

[](LICENSE) [](CONTRIBUTING.md) [](https://docs.claude.com/en/docs/claude-code/skills)

A [Claude Code](https://claude.com/claude-code) skill that turns a session into a quota-conscious orchestrator. It plans at a high level and hands the actual implementation work to other AI CLIs you're already paying for, instead of burning your Claude usage on the grind. Codex CLI is the default path; Grok CLI is optional.

"Fable" is what I call the orchestrating Claude model here, the one planning and delegating instead of coding. It's not a placeholder name, it's literal: Claude Fable 5 is the actual model I run day to day, and limiting its usage specifically is the entire reason this skill exists.

That said, the mechanism itself doesn't care which model sits in that seat. Whatever Claude model is running your session when you activate this skill is Fable for that session, Sonnet, Opus, whatever you're on. Nothing here checks or requires a specific model line, it only cares that the orchestrating model stops implementing and starts delegating. So if you're not on Fable 5, or Anthropic ships something else next, swap the name in your head, the skill still does the same job.

Where this came from

I got the idea watching [Theo (t3.gg)'s "A proper guide to Fable 5"](https://youtu.be/8GRmLR__OGQ?si=g_xN55h9h6hSCsui). He runs Fable almost entirely as an orchestrator, scouting, planning, kicking off workflows, reviewing PRs, rarely touching the diff itself. Watching that, the obvious question was: if the orchestrator barely codes, what's stopping me from routing the coding part somewhere that isn't metered against my tightest plan? This skill is that question, answered.

The goal

Protect one specific resource: m