codex exec: run Codex from a script or CI

Short answer: codex exec "your task" runs Codex once with no interface: progress goes to stderr, only the final answer to stdout, so codex exec "write release notes for the last 10 commits" > notes.md just works. It starts in a read-only sandbox; add --sandbox workspace-write when it should change files. --json streams every event, and -o file saves the final answer.

The Codex terminal app is made for a person who reads and approves. A script has nobody to approve anything, so codex exec changes three things: it never asks, it starts with the least permissions, and it separates the answer from the progress so other tools can use it.

AIGenerated with AI.

#How do you run Codex without the interface?

Pass the task as one argument:

codex exec "summarize the repository structure and list the top 5 risky areas"

Codex streams its progress to stderr and prints only the final message to stdout, so you can redirect or pipe it:

codex exec "generate release notes for the last 10 commits" | tee release-notes.md

Piped input becomes context, and the argument stays the instruction:

git diff main | codex exec "review this diff for bugs and missing tests" > review.md

#What can codex exec change?

Nothing, by default: codex exec runs in a read-only sandbox, and it never stops to ask for approval. Give it only what the job needs:

--full-auto is no longer supported; old scripts that use it need --sandbox workspace-write instead. Codex also refuses to run outside a git repository, to prevent destructive changes; --skip-git-repo-check turns that off when you are sure.

Codex sandbox and approval modes explains the sandbox in detail.

#How do you read Codex's output in a script?

With --json, stdout becomes a stream of events (thread.started, turn.started, item.started, item.completed, turn.completed, turn.failed, error); the last turn.completed carries the token usage:

codex exec --json "summarize the repo structure" | jq

With a JSON Schema, the final answer is data your next step can read without guessing:

codex exec "Extract project metadata" \
  --output-schema ./schema.json \
  -o ./project-metadata.json

#How do you continue a previous run?

With resume, for a pipeline in two steps:

codex exec "review the change for race conditions"
codex exec resume --last "fix the race conditions you found"

codex exec resume <SESSION_ID> picks a specific run. Add --ephemeral to a run you do not want saved to disk at all.

#How do you sign in on a CI runner?

codex exec uses your saved login by default. In automation, OpenAI recommends an API key in CODEX_API_KEY, set only for the one command that needs it:

CODEX_API_KEY=<api-key> codex exec --json "triage open bug reports"

Never set OPENAI_API_KEY or CODEX_API_KEY for a whole job that also runs the repository's own code: its build scripts, tests and dependencies can read the job's environment. On GitHub Actions, OpenAI's Codex GitHub Action keeps the key behind a proxy instead.

#Which settings does codex exec use?

The same ~/.codex/config.toml as the terminal app, and --profile picks a profile, so a CI profile can hold the model and sandbox for every run. --ignore-user-config runs without your config file, and --ignore-rules without your .rules files, for a run that must not depend on the machine it is on. Codex profiles shows how to set one up.

#Where can you check this?

• OpenAI Codex docs: non-interactive mode

#Can you build the command instead of typing it?

Config Studio for Codex has a codex exec page: write the task, pick where commands may write (read-only by default), a model and a profile, JSON events and a file for the answer, and it writes the command. Next to it are the config, profile, sandbox and rules files the run reads, explained setting by setting.