Claude Code and Codex both work by reading and writing files in one folder on your Mac. A clean setup is what stops them from overwriting each other, and what lets tomorrow's session pick up where today's stopped.
My Channel/ ← the workspace. Open Claude Code and Codex here.
├── AGENTS.md ← rules both tools read
├── CLAUDE.md ← Claude-only rules (pulls in AGENTS.md)
├── Brand/ ← fonts, colors, logos. Neither agent edits it.
├── Library/ ← what you draw on. Inputs, never published.
│ ├── Reference/ ← how good work is built
│ │ ├── My Work/ ← your best scripts and videos, kept as examples
│ │ ├── Craft/ ← guides you trust: hooks, retention, storytelling
│ │ ├── Editing/ ← what you've learned about the algorithm and engagement
│ │ ├── Motion/ ← screenshots of graphics you want to match
│ │ └── Thumbnails/ ← thumbnails that worked, yours or others'
│ ├── Research/ ← facts for your topics, each one dated
│ └── Media/INDEX.md ← a list of your footage, B-roll, music and sound effects
├── Content/ ← Claude's lane
│ └── Long-Form Video/
│ ├── CLAUDE.md ← long-form rules: your editor, pace, cut rules
│ ├── _performance_log.md ← how each video did, and what you learned
│ ├── _tools/ ← scripts Claude builds once and reruns
│ └── Videos/
│ └── my-video/ ← one folder per video
│ ├── my-video_script_v2.md
│ ├── my-video_motion.json
│ ├── my-video_cuts.json
│ └── _storyboard/
├── Motion/ ← Codex's lane
│ ├── AGENTS.md ← graphics build rules
│ └── my-video/
│ └── renders/ ← one file per graphic
└── Handoffs/ ← shared inbox between Claude and Codex
├── README.md ← the template
└── my-video.md ← one file per video
Footage lives on an external drive, outside the workspace:
My SSD/
├── Long Form Video/
│ └── my-video/
│ ├── raw/ ← camera files. Agents never delete these.
│ └── exports/
├── B-Roll/ ← clips and stills you reuse across videos
└── Audio/
├── SFX/ ← sound effects: mouse-click-sfx.mp3, whoosh-sfx.mp3
└── Music/ ← music beds you're licensed to use
AGENTS.md at the top holds the rules both tools need. Codex reads it automatically.CLAUDE.md holds Claude-only rules. Its first line is @AGENTS.md, which makes Claude read the shared file too.CLAUDE.md or AGENTS.md inside a folder holds rules for that kind of work: long-form editing in Content/Long-Form Video/, graphics in Motion/.Codex never reads a CLAUDE.md. So anything Codex needs to know goes in an AGENTS.md.
<aside> ✅
Best practice: write each rule once, in one file. Copy a rule into two files and the copies drift apart. Then one agent follows the old version.
</aside>
A lane is a set of folders one agent owns.
| Folder | Owner | The other agent |
|---|---|---|
Content/ |
Claude: scripts, cut lists, motion specs | Reads only |
Motion/ |
Codex: graphics code and renders | Reads only |
Brand/ |
Neither. You change it. | Both read |
Library/ |
Claude, and you add files too | Reads only |
Handoffs/ |
Both. Add your own entries, never rewrite the other agent's. |
<aside> ✅
Best practice: put this line in your AGENTS.md: "If a task needs you to write outside your lane, say so and stop." The one time my two agents wrote into the same folder, they overwrote each other's work.
</aside>
Agents copy what you show them. A library gives them real examples to work from, so you stop describing your style from scratch every session.
CLAUDE.md and AGENTS.md files, examples go in Reference/. A rule says "do this." A reference says "here's one that worked."