nam-cheol/codex-fable-mode

This repository is a plugin for Codex, an AI coding assistant. The plugin is called fable-mode and is described as a documentation-only plugin, meaning it contains no executable code of its own. Its purpose is to change how Codex approaches tasks before it starts working, making it pause and decide what kind of output is actually needed before doing anything. The core idea is called Output Lock. Before Codex writes code, edits files, or gives any response, it first classifies the task into one of several categories: answer, edit, implementation, review, audit, design artifact, research, clarification, or validation. This classification then shapes everything that follows. An answer stays an answer and does not turn into an implementation. An edit stays small and does not become a redesign. A simple question gets a direct response without a planning phase. Two additional constraints called Procedure Budget and Tool Budget limit how much work Codex does. Procedure Budget controls how much planning, questioning, and auditing happens before acting. Simple answers skip planning entirely. Small fixes check only the minimal relevant scope. Larger builds get a short plan before execution. Tool Budget limits which tools Codex reaches for, restricting file reads and searches to only what is needed to verify facts and check the affected areas. The plugin is written mostly in Korean and is aimed at Korean-speaking Codex users, though the underlying concepts apply regardless of language. You activate it by typing $fable-mode at the start of a message in Codex. Installation is done via two terminal commands that register the plugin from this repository. The README includes installation steps for both Codex CLI and Codex Desktop. The plugin is unofficial, not affiliated with Anthropic or OpenAI, and is provider-neutral in its guidance. It does not reproduce any proprietary prompts or internal system instructions.