I have seen this feature request for Codex quite a few times and now it’s actually coming:
Here is a link to the rough documentation.
And an explanation what skills are and why you may want to use them.
Happy building and please remember that this is an experiment. Share some feedback with the team here, or on GitHub
Was expecting an issue or discussion link.
Found this issue but not sure if it is the correct one.
Plan mode and Skills form a feedback loop.
Plans are often one-off and may already use existing Skills, but execution exposes what those Skills (or the plan itself) missed. Those gaps and fixes are what justify creating new Skills, and the next plan benefits from them immediately. You get the full payoff only when missed details in real plans are turned into reusable Skills.
Mental model:
Plan (uses Skills) → Execute → Miss → New Skill → Better Next Plan
This is just one possible use case with skills. There are so many more, this is one that I use daily as a software developer and find that some daily grind workflows become so much more effective with Plan mode and skills with the feedback loop of possible new skills.
Do not read anymore into that than just some feedback to the OpenAI team developing the skills for Codex to also consider that as a use case.
This is the correct link, to the best of my knowledge. Since Codex CLI is an open source project, it makes sense to have these conversations on GitHub. It could also be worth a try to invite the team to engage with the community here more often.
And, thanks for the feedback! I understand skills as collections of more advanced tools and prompts that are exposed to the model when they are useful or required to complete a task.
My expectation for the near future is something closer to specialized agents with defined skills, orchestrated by a plan or execution manager to handle larger tasks.
For now it seems the intent is to rely on a single agent with additional capabilities. But let’s see what the team will come up with.
I do apricate your effort to get this correct.
The link to https://github.com/openai/codex/blob/main/docs/skills.md is the link to the documentation for Codex skills and is the one I would also publish for the documentation.
What I am seeking is something like the Plan mode discussion to give feedback and one in which the OpenAI developer team will engage with replies as needed. Make sense?
Where is the source of that idea? 
Curious if that is noted in the official documentation somewhere.
agent with additional capabilities is one possible use case but there are so many more.