Naka-pin na Tweet

Codex for Claude Code Users
Heard good things about Codex and want to give it a try? Here's a cheat sheet to smooth the transition from Claude Code:
Basic setup + usage
Rename CLAUDE .md to AGENTS .md. Add a new CLAUDE .md whose contents are just `@AGENTS.md`. Now both apps will read the same instructions.
The following things are the same in both apps:
- Ctrl+C to clear current input. (Double Esc also clears input in Claude but not Codex.)
- Double Esc opens chat history browser if text input is empty.
- /init to generate starter AGENTS/CLAUDE md file.
- Ctrl+g to edit prompt in your default $EDITOR
- Ctrl+v to paste an image.
- !command to execute a shell command directly (skipping the LLM)
- `/model` to switch the current model.
When viewing chat history, jump between messages with left/right arrows in Codex (up/down in Claude.)
Codex doesn't yet have an equivalent of Claude's "stash prompt" (ctrl+S) or reverse history search (ctrl+R).
In Claude use opt-T (alt+T on Windows) to toggle thinking mode. Codex doesn't have a separate "thinking mode" toggle; instead you configure thinking when selecting the model with `/model`.
Resuming/browsing old sessions
`claude --resume` → `codex resume` (or `/resume` in either app)
`claude --continue` → `codex resume --last`
`claude --resume ` → `codex resume `.
`claude --resume --fork-session` → `codex fork `.
Steering vs queuing
Codex lets you "steer" the model's output like Claude: send a new message with Enter while the previous response is still generating, and it'll be added to the ongoing reasoning; the model sees your message immediately and can adjust what it's doing mid-response.
Unlike Claude, Codex also lets you queue new messages. Send them with Tab instead of Enter and Codex will wait until the current response is fully complete instead of injecting them immediately.
Permissions
Codex has a "plan mode" similar to Claude's. Like Claude you can activate plan mode using shift-Tab. (Alternatively use `/plan`)
`claude --dangerously-skip-permissions`
=
`codex --dangerously-bypass-approvals-and-sandbox`
(or equivalently `codex --yolo` 😎)
Skills, Hooks, MCP
Codex skills go in `~/.codex/skills`. Claude Code skills go in `~/.claude/skills`.
Trigger a skill with `/skill-name` in Claude Code, `$skill-name` in Codex.
Codex doesn't have any equivalent to Claude's hooks.
MCP servers work similarly in both apps, e.g. `codex mcp add` vs `claude mcp add`; use `/mcp` to manage your installed MCPs in either app.
----
Let me know if I've missed anything!
x.com/craigzLiszt/st…
English




























