CodeX Project Setup Pack · Free · v5

Answer three questions,

and the folder is a project

Respects CodeX's native design — a plain-text interview that pauses for you, writes only the workspace AGENTS.md, uses CodeX's native mechanisms. Set up in one pass.

What's inside
  • AGENTS.md This workspace's root AGENTS.md — a working-conventions block (language, work type, explanation style + 2 ground rules), leaves your global config alone
  • Skill One reusable procedure — code-review or decision-record, pick one, written to the official .codex/skills/ — ready to use
  • Folders Five working folders: work / archive / library / log / inbox — plain descriptive names, no number prefixes
  • Memories CodeX native memories (~/.codex/memories/) — the official background mechanism; no hand-written memory file
  • Interview 3 plain-text questions (language / work type / explanation style) — CodeX pauses between each for your reply
  • No symlink The procedure lives in the in-repo official skills dir and travels with the repo — never touches ~/.codex/
  • Updates Version updates pushed to the GitHub repo — no sales pitches
How it works

Paste, say one line, CodeX-native

Paste codex/init.md into a fresh CodeX chat, say "Run LUCE setup", and it walks you through:

  1. 01/
    Environment check

    Reads your environment first (OS, existing AGENTS.md, git repo) and prints what it found — touches nothing

  2. 02/
    Three questions

    Conversation language / what this folder is for / your preferred explanation style — plain text, pausing after each for your reply

  3. 03/
    Write conventions + add a procedure

    Appends working conventions to this workspace's AGENTS.md (never the global one), then writes one reusable procedure into the official .codex/skills/ — no symlink

  4. 04/
    Create folders, wrap up

    Creates work / archive / library / log / inbox, prints a wrap-up, done

CodeX leans toward acting immediately, so it pauses between each part for your "continue" — making sure the questions are answered before moving on. Set up in one pass.

Not for you if

Three honest misfits

  1. 01/ People expecting a fully autonomous no-prompt agent — you still need to make judgments
  2. 02/ Heavy users with a mature workspace setup — this will feel too basic
  3. 03/ Teams wanting enterprise configuration — see Studio
FAQ

Design, CodeX philosophy, exits

  1. 01/ How is this different from the Claude Code Project Setup Pack?

    Same folder structure; the difference is the conventions file (CodeX uses AGENTS.md, Claude uses CLAUDE.md) and the native mechanisms. The CodeX version respects bias-to-action: it pauses after each question for your "continue", rather than running straight through.

  2. 02/ Why use CodeX native memories?

    CodeX has native memories at ~/.codex/memories/ that the system reads and writes in the background, carrying over across chats. The pack just points you at this official mechanism instead of building a hand-written memory file that fights with it.

  3. 03/ Will this touch my global CodeX config?

    No. The pack writes only this workspace's root AGENTS.md (which CodeX auto-loads); it never touches your global ~/.codex/AGENTS.md, so your other projects are unaffected.

  4. 04/ Does it touch my ~/.codex/ setup?

    No. The reusable procedure is written straight into the in-repo .codex/skills/ (CodeX's official location for project skills); it travels with the repo and needs no symlink. Your existing AGENTS.md is not overwritten either — a dated conventions block is appended, existing content untouched.

  5. 05/ What if I use both Claude Code and CodeX?

    Run the Claude Code Project Setup Pack separately. The two tools read different files (Claude Code reads CLAUDE.md, CodeX reads AGENTS.md), so they coexist cleanly.