Improve with Agent
Choose a skill
Select the kind of help you need while keeping one conversation and one deliverable.
Skills give the Agent a method for a particular job. They run inside the same Agent workspace and use the same access and evidence boundaries.
Open Skills to see the available labels, descriptions and output types. You can also select a skill when starting a chat.
Start from a workflow
New chat lists workflows grouped by job, such as creating content, social and video, improving your site, AI visibility, and planning and measurement. Each workflow runs one skill, and content workflows also fix the format, such as a blog post or a LinkedIn post. Pick one, fill in its short form and choose Start.
After a deliverable is ready, the chat suggests refinements and next steps. A next step opens a new chat with its workflow selected and the current revision as the brief.
Choose by the outcome
| What you need | Look for a skill producing |
|---|---|
| A prioritized approach to a problem | A plan or diagnosis |
| A way to evaluate work later | A measurement deliverable |
| A review of saved findings | Research |
| Proposed changes to an existing page | Page edits or a technical fix |
| A proposed internal-link approach | A link plan |
| A brief for an external publisher opportunity | An earned brief |
| A new long-form piece | Content |
| A social post, short script or forum answer | Content |
| A plan for your next prompt generation run | A prompt plan |
Use the descriptions in the current catalog to choose the exact skill. Availability can change as the product evolves.
Let an Action guide the choice
An attached Action can recommend a skill based on its diagnosis. Review the recommendation alongside the evidence and the work you intend to do.
If you choose a skill or a workflow explicitly, that choice takes precedence for the turn. Without an explicit choice, the Agent can use the Action’s recommendation, the previous chat skill or select a skill for the request.
Keep one output kind per chat
A chat owns one kind of deliverable. You can refine it over multiple turns, including after making your own edits.
If you need to move from a technical-fix deliverable to a different output kind, start a new chat. This keeps the output history understandable and avoids replacing one team’s work with an unrelated format.
Review before using
A skill helps structure the work; it does not make the result automatically correct. Check factual claims, sources, omissions and fit for your audience.
A prompt plan writes no prompts: apply it in the Prompt Builder, which generates suggestions you accept or reject in Prompts. A measurement plan is an explanation, not permission to add arbitrary checks to an Action. A content draft is ready for your review, not automatically published.