GitHub radar
Fable Orchestrator: Claude as Planner Inside Codex
A Codex skill that puts Claude Fable 5.1 in charge of planning — it never writes code itself, just decides what to do and who does it, then delegates to GPT or DeepSeek.
Fable Orchestrator is a local-first routing skill for Codex that adds Claude Fable 5.1 as a dedicated planning and adjudication layer. Claude never writes code or owns the workspace — it builds a plan, decides which model handles which work, and adjudicates when decisions arise. GPT-5.6 Luna handles standard implementation; DeepSeek V4 Flash takes repeated loops and high-throughput iteration. Every implementation node must use one of those two routes — if neither is reachable, the workflow reports the blocker instead of improvising a solution. Install runs from the repository root with ./install.sh --copy, landing the three skill files in ~/.codex/skills/fable. Invoke with '$fable <objective>' from any Codex task.
Why a vibe-coder should care
With Claude handling planning separately from implementation, the agent stops misreading requirements and diving into code prematurely. The separation of concerns — plan once, then delegate — reduces the feedback loops that happen when an agent misunderstands the task and has to redo everything.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Install this Codex skill from https://github.com/codejunkie99/fable-orchestrator — follow the README, run ./install.sh --copy, then invoke $fable on a small test task and show me how Claude's planning step looks
Runs on a regular laptop — this is a skill for Codex; no special hardware requirements.
Open on GitHub▌ More finds