Codex mobile CLI docs your team can read anywhere

Reading a Codex run from your phone is mostly a docs problem, not a hardware one. If your repo holds its own rules and receipts, you can check state from a train; if it keeps that knowledge in a chat thread, no screen size will save you. Codex CLI is OpenAI's coding agent that runs in your terminal, and "mobile docs" here means the repo-visible artifacts a teammate can read away from the desk: AGENTS.md rules, run notes, and verification transcripts.

So the picture people are searching for, controlling Codex from a phone, is really about reading state and trusting receipts. You rarely need a remote joystick. You need a repo that can answer questions on a small screen without you scrolling through a chat export.

Pin the model so reviews stay predictable

The first thing distance amplifies is a quiet model swap. Different models carry different risk appetite, so when one run uses one model and the next uses another, review expectations wobble and nobody notices until a merge goes sideways.

Pin the default model and the escalation rule right inside AGENTS.md. Then whoever opens the run, on whatever device, can reason about the blast radius before approving. The pin is two or three lines, and it travels everywhere the repo does.

Keep a connector roster in the repo

MCP connectors accumulate quietly, and they accumulate fastest when nobody is at the desk to ask why. Each server you add widens what the agent can touch, so least-privilege erodes one convenient connector at a time.

Check a plain Markdown roster into the repo root, one line per connector with its allowed actions. Use the MCP specification as your shared vocabulary so the roster reads the same to everyone. Security reviews then start from a real list instead of a guess.

Cap AGENTS.md at a done checklist

AGENTS.md grows until it stops being readable, and a long file is exactly where ambiguity hides. When the definition of done gets buried, Codex optimizes for the wrong "done" and the CI green tells you nothing.

Put a Definition of Done at the top of the file, ten lines or fewer, in plain bullets. Short enough to read at a bus stop, precise enough to review a PR against. Here is a snippet you can paste straight in:

# AGENTS.md verification snippet

- Every Codex CLI run ends with the transcript snippet reviewers can replay.
- Pair browser evidence with the project's normal CLI checks before merge.
- If MCP servers are enabled, list allowed actions beside each connector name.

Document slash commands once

Undocumented slash commands turn into private folklore. One person knows a keyboard shortcut, another invents their own, and during an incident everyone is speaking a slightly different dialect.

Keep a slash catalog at docs/codex-commands.md and link it from AGENTS.md so every operator can find it. The slash commands reference covers the built-ins; your catalog adds the team-specific ones. After that, the team converges on one set of moves instead of guessing.

To sanity-check the four artifacts, open your repo on a phone and try to answer these without zooming into a chat:

One thing the docs cannot do for you: hard calls still belong to humans wherever the session runs. Threat models, customer promises, and blast-radius decisions stay off autopilot. openai/skills is worth watching for how packaged skills change the setup, but the ownership line does not move.

The quickstart and the Codex CLI docs assume you are at a desk, while the features page describes the tool wherever it runs. The companion piece on starting Codex CLI from anywhere covers the operating side, and the broader practice area lives at Codex CLI workflows.

Common questions

• What are Codex CLI mobile docs?

  Codex CLI mobile docs are the repo-visible artifacts a team can read away from the desk: AGENTS.md rules, run notes, and verification transcripts. This note keeps four of them current: a pinned model rule, a Markdown connector roster, a ten-line Definition of Done, and a slash-command catalog. Together they let a phone-sized screen answer real questions about a run.

• How short should AGENTS.md stay?

  Short enough that the Definition of Done fits in ten lines or fewer at the top of the file. When AGENTS.md grows unchecked, Codex starts optimizing the wrong done. The checklist is the fix that keeps output aligned with the same vocabulary the team reviews against, and it is the one doc guaranteed to fit every screen.

• Where do Codex slash commands get documented?

  In a slash catalog at docs/codex-commands.md, linked from AGENTS.md so every operator can find it. Undocumented keyboard shortcuts become private folklore and operators drift apart. Once the catalog exists, the team converges on one shared set of commands instead of inventing private variants in the middle of an incident.

• Why pin the default model in AGENTS.md?

  Pinning the default model plus an escalation rule keeps review expectations steady when models get swapped casually. Different models imply different risk appetite, so the pin lets a lead reason about blast radius before a merge rather than after an incident. It costs two or three lines and saves a postmortem.

Where to start

If your team reads agent runs from trains more often than desks, the done checklist is the first artifact to add, since ten lines at the top of AGENTS.md fit any screen. When you want help wiring the rest into a real workflow, tell us where it breaks.