Question: Workflow guidance how to use agent skills

[gcapnias]

Hi, I was thrilled when I saw the announcement on WordPress.com—it’s about time!

However, I haven’t been able to find a guide or video explaining how to use these skills. From the repository, it looks like there’s a workflow, but I can’t seem to find any documentation anywhere. Am I missing something?

Regards,
George J.

[thejimbirch]

Once skills are made available, either globally, or in your project, your LLM will automatically be aware of them, and use them when needed. Skills are for agents to decide when to use.

If you want to call them explicitly, in Claude Code you can create custom slash commands and agents to use the specific skills. But for the most part, think of it like the Matrix. When the agent needs to learn how to fly a helicopter, it can reference the "Fly a helicopter" skill.

[gcapnias]

I understand that the agent automatically loads skills based on the context, but ultimately, success depends on the wording. One word less, one skill does not load, one word more, the wrong skill loads.

Let’s say I issued: “I want to create a new block theme called demo-theme.” However, only the wp-block-themes skill was activated, and the script included in that skill, never ran. I’m pretty sure another one should have been activated, between wordpress-router or wp-project-triage.

There are scenarios that were created to test these skills, and in the disclosure released alongside them, it clearly stated, “We tested skills with AI assistants and iterated based on results.”

Although there are scenarios, there are no instructions on how to use them to run tests and verify the results. If the scenarios run and we knew which skills should be activated, we could use them as guides to apply these skills.

[justlevine]

your LLM will automatically be aware of them, and use them when needed. Skills are for agents to decide when to use.

@thejimbirch sadly this is rarely - if ever - the case.

@gcapnias if you're looking for some suggestions on how to improve the likelihood of a skill being trigged, check out https://vercel.com/blog/agents-md-outperforms-skills-in-our-agent-evals

(While I also hope we'll get the actual evals from the team, in the interim you can use https://github.com/WordPress/wp-bench yourself if you've got the tokens to spend. )