03 | Codex Commands: Entry Points for Plan, Goal, and Review

Commands are not a shortcut list

The upstream guide explains Commands as entry point, state, and action. They provide discoverable, repeatable ways to change working posture; they do not replace a task description. Establish the current state, choose the appropriate entry point, and still inspect the result.

For App users, planning, long goals, and review are the stable core. MCP, Apps, and Plugins are capability-discovery and management entry points. CLI slash commands are a supplement, not the desktop workflow’s center.

Plan before writing

Use /plan to turn an ambiguous request into reviewable steps. Ask Codex to state its understanding, scope, sequence, verification, and risk before it edits files:

/plan
Analyze possible causes of the failed payment callback.
Do not edit files. List code, logs, and tests to inspect,
then propose the smallest repair and its verification steps.

A plan is not a commitment. If it exposes missing requirements, untestable assumptions, or an overly broad impact area, return to clarification rather than carry uncertainty into a write.

Long goals and review need different postures

/goal is for work spanning multiple steps. Its goal needs a boundary and acceptance conditions: what counts as complete, what evidence must remain, and when Codex should stop for confirmation. A long task must not grow just because it is asked to continue.

/review shifts attention from implementation to trust. Review diffs, test and manual evidence, risk, omissions, and the next step—not another explanation of work already done.

Commands, Skills, and Automations

Commands enter the current task; Skills preserve reusable process knowledge; Automations run work from triggers. First prove a workflow manually in a thread, then decide whether it should become a Skill or an Automation. When a command cannot be found, check App version, workspace, permissions, and nearby feature entry points before concluding the capability is absent.