06 | Codex Skills: Turn Repeated Prompts into Workflows

Skills solve stable repeated work

The upstream guide defines a Skill as reusable workflow knowledge. When a prompt repeats, its steps are relatively stable, and its output can be specified, it should not have to be rewritten in every thread. A Skill lets Codex load the right process for the right situation instead of overloading project instructions with every detail.

It is not for one-off work, and it does not replace project-wide safety or code rules in AGENTS.md.

Every Skill needs four answers

The course centers the structure around SKILL.md:

# When To Use
Task signals and explicit non-use boundaries.

# Workflow
The order for checking, analyzing, acting, and verifying.

# Output
Required conclusions, evidence, and next steps.

# Rules
Allowed scope, prohibitions, and stop conditions.

Scripts, templates, and reference material can sit beside the Skill when they serve that workflow. Do not copy an entire repository into it.

When to promote a prompt

Ask three questions: Does this task recur? Are its steps stable? Does the team want a consistent output? When the answers are all close to yes, promotion is justified. The upstream guide uses code review, documentation updates, migration checks, and domain-specific verification as examples of valuable Skills.

Relationship to Commands and Automations

A Command enters a task; a Skill explains how to perform this class of task; an Automation triggers it on a schedule, event, or thread follow-up. Verify a Skill in human-run threads before moving it into background work.

Avoid giant catch-all Skills, unclear triggers, and steps without an output standard. A Skill is useful only when its execution demonstrably removes repeated work.