About this Skill
Work through a page from top to bottom, identify a concrete element, open an annotation at its position and leave a short actionable critique. The instructions cover toolbar state, coordinate interactions, shadow-root access and confirmation after every note.
Why we picked it
The review workflow requires an installed Agentation toolbar and the agent-browser CLI. It creates annotations; automatic code fixes require a separately connected agent and MCP workflow.
What it helps with
- Add review notes to a page while a user watches the browser.
- Review hierarchy, navigation, spacing, typography and calls to action.
- Pass precise element context to a separate implementation agent.
Run the annotation loop
Check the browser and toolbar
Open a visible browser, confirm the toolbar exists and expand it only if collapsed. Keep page-interaction blocking enabled.
Target a specific element
Inspect the page, scroll the element into view and obtain its bounding box. The upstream CLI workflow uses coordinate-based pointer input for the annotation overlay.
Write and verify the note
Give a short critique with a concrete change and a named design principle. Submit it and confirm the annotation count increased before continuing.
Keep feedback specific
The workflow covers the hero, navigation, illustrations, content rhythm, typography and final actions. It suggests five to eight notes for a typical page, while individual notes remain concise and actionable.
Reviewing and changing code are separate steps. The optional two-session flow sends annotations through MCP to another agent that edits the code. Installing this Skill alone does not establish that connection.
Install after toolbar setup
Use the companion setup Skill first if the page has no toolbar. The upstream instructions also describe troubleshooting stale browser sessions, collapsed toolbars and unsaved notes.
| Command | What it helps with |
|---|---|
npx skills add benjitaylor/agentation --skill agentation-self-driving | Select the autonomous annotation workflow. |
Before you use it
- The target page already has the Agentation toolbar installed.
- The agent-browser CLI and a visible desktop browser are explicit upstream dependencies.
- The assistant must be able to run the browser commands and interact with the page.
- An MCP connection and a separate listening agent are required for the optional review-and-fix loop. Annotation-only review can work without that connection.
Supported agents
Sources & references
Original sources are linked below. Practical starting points are editorial suggestions from SkillHey.
