Keep a running task context that Codex automatically reloads at the start of every new session
Track architecture decisions and per-turn notes as plain editable text files inside your project folder
Use the /goal command in Codex to activate goal mode and anchor a session to a specific long-running objective
Requires Node.js 18 or later and OpenAI Codex with hooks support enabled, targets Windows environments.
This repository provides a skill and installer for OpenAI's Codex coding assistant that adds persistent goal tracking across sessions. The core problem it addresses is that AI coding sessions normally have no memory between restarts: if you set a long-running task (map the architecture, find fragile areas, make an improvement, add tests), that context is lost when you close and reopen Codex. Goal-hook works around this by maintaining a folder called .codex-goal/ in your workspace that stores the ongoing goal, accumulated memory, decisions made, and per-turn notes as plain files on disk. When Codex starts a new session, a hook script reads those files and loads the stored context back in, so the assistant picks up where it left off. The hook integrates with Codex's hooks system, which allows external scripts to run at defined points in the session lifecycle such as when a session starts. Installation uses npm and PowerShell. The installer writes the necessary hook configuration into Codex's config file automatically when the --apply-config flag is passed. After restarting Codex, a /goal command becomes available that activates goal-mode for the current task. The runtime files it creates (context.md, memory.md, decisions.md, and others) are plain text and can be read or edited by hand. The project targets Windows environments given its reliance on PowerShell, though the README includes a brief note referencing a Linux community forum. Node.js 18 or later is required alongside Codex with hooks support.
← last-emo-boy on gitmyhub — every repo by this author, as a profile.
Verify against the repo before relying on details.