Codex CLI workflows
Mini-workflows using verified Codex commands. Cross-tool, tool-agnostic versions of these workflows live in Example: Understand a new repository (and the other walkthroughs such as Example: Build a new feature end to end); for capability comparisons across agents see Feature Matrix.
Understand a codebase
codex --sandbox read-only --ask-for-approval on-request
> /init # scaffold AGENTS.md so future sessions know the repo
> Explain the request lifecycle in this repo, from HTTP entry to database
> /mention src/server.ts
> Where is authentication enforced? List file:line references
Tips: read-only sandbox keeps exploration side-effect-free; @ attaches files; /status confirms the active model and writable roots. For large repos, ask Codex to delegate exploration to explorer subagents (see Codex CLI subagents).
Build a feature
codex --sandbox workspace-write --ask-for-approval on-request # the recommended "Auto" combo
> /plan Propose an implementation plan for adding retry-with-backoff to the payment client
> (approve the plan, then:) Implement it, with tests
> /diff # inspect exactly what changed, including new files
Tips: /plan gets you an approach to approve before edits; a good AGENTS.md (build/test commands, style rules) pays off here; /compact when the session gets long; /goal for multi-step objectives with automatic continuation.
Debug
codex --sandbox workspace-write --ask-for-approval on-request
> Tests are failing with "connection refused" in src/db/pool.test.ts — find the cause and fix it
> ! npm test -- --run pool # run a local command under the current sandbox yourself
> /ps # check background terminals; /stop kills them
Non-interactive alternative for CI logs:
cat failure.log | codex exec --sandbox read-only "Find the root cause in this log" -
Refactor
> /new refactor-session
> /plan Propose a stepwise refactor of the config module into typed submodules
> Implement step 1 only
> /diff
Tips: work in small, verifiable steps; use /fork to try an alternative approach without losing the current transcript; use /new to keep refactors out of unrelated chats. For sweeping renames across a monorepo, subdirectory AGENTS.md files keep per-package conventions in scope (see Codex CLI project instructions (AGENTS.md)).
Code review
Interactive, before committing:
> /review # reviews the working tree; never modifies it
Non-interactive, for branches, commits, or automation:
codex review --uncommitted # staged + unstaged + untracked
codex review --base main # diff against a branch
codex review --commit abc123 --title "hotfix"
codex review "Focus on SQL injection risks" # custom prompt instead of a target
echo "Focus on SQL injection risks" | codex review - # custom prompt from stdin
Notes: set review_model in config.toml to use a different model for reviews; codex apply <TASK_ID> pulls the diff from a Codex cloud chat into your local tree.
Last verified: 2026-09-07
Official sources:
- https://developers.openai.com/codex/cli/
- https://developers.openai.com/codex/cli/slash-commands
- https://developers.openai.com/codex/noninteractive
- https://developers.openai.com/codex/sandbox
- https://developers.openai.com/codex/guides/agents-md
- https://developers.openai.com/codex/subagents
AI coding tools evolve quickly. If you find outdated information, please open an issue or pull request.