In partnership with
How to Make Claude and Codex Work Together
WTF is file-based agent teamwork?: It means agents coordinate through files instead of one messy chat. Claude writes the plan. Codex builds from it. Claude reviews the diff. The next session reads the handoff. The file is shared memory. Chat is the work surface.
You ask Claude to plan a feature.
Then you paste the plan into Codex.
Codex edits files, hits errors, fixes tests, and explains what changed.
Then you paste that back into Claude.
Claude asks what files changed.
You paste more.
After 20 minutes, you are the courier between two agents.
A new paper called tap: A File-Based Protocol for Heterogeneous LLM Agent Collaboration shows a better pattern. Different agents can collaborate through files. Messages are saved as Markdown files with metadata. Notifications help, but the file stays the source of truth. The paper also used separate git worktrees so agents could work without overwriting each other.
The practical version is simple:
Make the project remember.
I am Alex, welcome to ShortCu8 by Innov8.
Lets Dive Deep 🐰
⭐Today's Shortcut
Create this folder in your repo:
AI_TEAM/
PLAN.md
BUILD_NOTES.md
REVIEW.md
HANDOFF.mdThen split the work:
Claude = planner + reviewer
Codex = builder + tester
Files = shared memory1. Add the standing rule
Use AGENTS.md for Codex. Use CLAUDE.md for Claude Code. If Claude needs the same rules, put this inside CLAUDE.md:
@AGENTS.mdStarter rule:
## AI team workflow
- Claude writes `AI_TEAM/PLAN.md` before implementation.
- Codex builds from `AI_TEAM/PLAN.md`.
- Codex updates `AI_TEAM/BUILD_NOTES.md` after editing.
- Claude reviews the diff and writes `AI_TEAM/REVIEW.md`.
- Before switching sessions, write `AI_TEAM/HANDOFF.md`.
- Do not claim done unless BUILD_NOTES.md has checks and results.Keep this file short. Put repo rules, test commands, and handoff rules in it. Do not paste your whole company wiki into it.
2. Make Claude write PLAN.md
Use Claude first when the task is unclear.
Prompt:
Read AGENTS.md.
Write `AI_TEAM/PLAN.md`.
Include:
- goal
- files likely involved
- risky parts
- build steps
- exact test command
- acceptance checklist
Do not edit code.
Make this a work order for Codex.Good PLAN.md:
Goal:
Fix mobile navbar overlap on checkout.
Files:
- apps/web/components/Navbar.tsx
- apps/web/styles/layout.css
Risk:
Navbar is shared with account pages.
Test:
pnpm --filter web lint
pnpm --filter web test
Done when:
- no overlap at 390px
- desktop nav unchanged
- tests passClaude's job in this step is reducing misunderstanding.
3. Make Codex write BUILD_NOTES.md
Open Codex in the repo.
Prompt:
Read AGENTS.md and AI_TEAM/PLAN.md.
Implement only the plan.
After editing, write `AI_TEAM/BUILD_NOTES.md` with:
- files changed
- what changed
- commands run
- test results
- skipped items
- uncertaintyGood BUILD_NOTES.md:
Files changed:
- apps/web/components/Navbar.tsx
What changed:
- Added mobile wrapping under 480px.
Commands run:
- pnpm --filter web lint: passed
- pnpm --filter web test: passed
Uncertainty:
- Account page still needs visual check at 390px.This is the point: Codex should leave state in the repo. Chat alone is too easy to lose.
4. Make Claude review the diff
Prompt:
Read:
- AGENTS.md
- AI_TEAM/PLAN.md
- AI_TEAM/BUILD_NOTES.md
- current git diff
Write `AI_TEAM/REVIEW.md`.
Check:
- plan followed?
- extra files touched?
- tests shown?
- edge cases missed?
- ship, revise, or restart?
Do not edit code.Good REVIEW.md:
Verdict:
Revise before shipping.
Problem:
Account page uses the same Navbar but was not checked.
Required fix:
- Check account page at 390px.
- Update BUILD_NOTES.md after the check.Now Codex gets the review:
Read AI_TEAM/REVIEW.md.
Fix only the required items.
Update BUILD_NOTES.md with new checks.5. Use HANDOFF.md before a new session
Long sessions get stale. Before switching agents or starting fresh:
Create `AI_TEAM/HANDOFF.md`.
Include:
- current goal
- status
- files changed
- commands run
- open issues
- next safest step
- which agent should continueNext session starts with:
Read AGENTS.md, AI_TEAM/HANDOFF.md, PLAN.md, and REVIEW.md.
Continue from the next safest step.
Do not assume old chat context.Yesterday's context canary told you when one session is going bad. HANDOFF.md lets the next session continue without guessing.
Pro version
The tap paper used separate git worktrees. Use that only when agents edit in parallel.
git worktree add ../myapp-codex codex-build
git worktree add ../myapp-claude claude-reviewRule:
Codex edits in myapp-codex.
Claude reviews in myapp-claude.
Main repo merges after review.For a small bug, the AI_TEAM/ folder is enough.
Now go build something great.
The ShortList
🛠️Cool Tools of the Week:
Runway: The AI video generator is now available inside of OpenAI's ChatGPT.
Dreamina Seedance 2.0 Mini: A low-cost AI video generator that's 30% more cost efficient than its predecessor, Seedance.
Sonic-3.5 and Ink-2: Cartesia's streaming models for text-to-speech and speech-to-text, now available for voice agents.
📩 Innathe Shortcu8 engane undarunnu 👇️?
We read every reply - just reply to this email and let us know how we can improve !
Appo adutha Shortcu8il kanaam bie…👋
If you read till here, you might find this interesting
#AD1
Your prompts are leaving out 80% of what you're thinking.
When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning — constraints, edge cases, examples, tone — and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.
89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Try Wispr Flow free — works on Mac, Windows, and iPhone.
#AD2
What happens when the S&P moves 3% during your commute?
We are living in volatile times. While you cannot control the state of international affairs, you can position your portfolio accordingly.
Liquid is one of the fastest growing trading platforms, allowing users to trade stocks, commodities, FX, and more 24/7/365 from their phone and computer.