Skills
- First public date: 2026-02-25
- Official references:
- Note: This entry records when Skills were officially introduced as an AI engineering mechanism. Concrete Skill design and authoring practices should be expanded in the Guide.
What It Is
Skills are a mechanism for packaging methods for completing specific task types.
By OpenAI's definition, Skills are reusable and shareable workflows that tell ChatGPT how to complete a specific task more consistently. They can include instructions, steps, examples, and even code plus dependency resources. Official docs also mention structured files such as SKILL.md.
The problem Skills solve is not "can the model answer," but "how to make AI follow the same method reliably in repeated tasks."
What Step It Moved AI Application Engineering From and To
Skills moved AI application engineering from "rewrite prompts and restate requirements every time" to "package expertise and process into reusable execution units."
Before Skills, many AI workflows were still one-off prompting: if the user asked well, outcomes were stable; once operators changed, quality and format drifted. Skills began turning expert know-how, team conventions, and fixed procedures into installable, reusable, shareable, and portable workflow objects.
This means teams are no longer only consuming model capability; they are building a reusable process layer on top of models.
What Stage It Is In Now
I currently mark Skills as emerging.
Officially, Skills are already presented as a product capability and are described as relevant to ChatGPT, Codex, and API ecosystems. But based on Help Center positioning, they are still in beta or early rollout, and scope, cross-product consistency, and team governance patterns are still forming.
In short, Skills are no longer a vague concept. They are a formal workflow mechanism at product level, but not yet at the stage where every AI product supports them by default with fully stable cross-platform behavior.
What It Might Replace
Skills can replace part of scattered prompt templates, internal prompt docs, and non-reusable word-of-mouth workflows.
If this path holds, teams will store not only "how to ask," but directly "how to do this task" as a Skill that bundles prompts, process, format, and checks. This pushes AI applications from conversation tools toward workflow runtimes.
What Might Replace It
If stronger standards for execution units appear, Skills may be superseded.
For example, stricter agent workflow formats, more native multi-tool orchestration protocols, or standards that directly encode permissions, state, checkpoints, and observability may evolve reusable workflows into fully executable task units. At that point, Skills may continue evolving or be absorbed by a next workflow layer.