Browser Automation
Agent Buildingbrowser-automationBrowser Automation is an Agent Skill for agent building published in the agent0ai/agent-zero repository on GitHub: a SKILL.md instruction file an AI agent can load to do this task.
Use for complex Agent Zero browser automation, including multi-tab browsing, screenshots, forms, uploads, raw pointer/keyboard actions, host-vs-container browser mode, and visual verification workflows.
How to use this skill
Bring this guide into your coding agent with a prompt tailored to the tool you use.
- Open your project in Codex.
- Copy and paste the prompt below.
- Review files and risks before approving any action.
Inspect the Agent Skill at https://github.com/agent0ai/agent-zero/blob/HEAD/plugins/_browser/skills/browser-automation/SKILL.md and all supporting files. Report provenance, permissions, scripts, dependencies, and security risks. Propose an exact installation into .agents/skills/browser-automation, but wait for my approval before writing files or running scripts.
Copying this prompt does not install or run the skill. Review third-party files before use.
Browser Tool
Use this skill after the compact browser tool prompt points you here. It is the progressive-disclosure workflow guide for rendered pages, multi-step browser work, logins, downloads, JavaScript-heavy sites, screenshots, host/container browser mode, and visual inspection. Prefer search_engine or document_query for plain text research.
For fragile forms, load browser-form-workflows with skills_tool:load before acting when selects, checkboxes, radios, file uploads, contenteditable fields, validation, or final submission state are central to the task.
Core Workflow
opencreates a browser tab and returns abrowser_id.contentreturns readable markdown plus typed refs like[link 3],[button 6],[input text 8].- Interact with refs using
click,type,submit,scroll, etc.; iframe/shadow targets may return frame-chain metadata in action results. - Use
navigateon an existingbrowser_idfor serial browsing. - Keep only a small working tab set; close pages when finished.
- If the user asks for an existing tab, page title, or already-open URL, call
listfirst, match bytitleorcurrentUrl, then useset_activeornavigateon thatbrowser_idinstead of opening a new tab.
Modes
The same tool may run in Docker container mode or A0 CLI host-browser mode, depending on project/plugin settings.
- Container mode: browser and upload paths resolve inside the Agent Zero container.
- Host mode: browser and upload paths resolve on the connected A0 CLI host machine.
In host mode, page content and screenshots may be blocked by host-content policy when remote models are active.
Screenshots And Vision
Screenshots are explicit only; the browser does not automatically load images into model context.
- Call
browserwithaction: "screenshot". - Call
vision_loadwith the returnedvision_load.tool_args.pathsvalue. - Reason from the latest loaded screenshot.
Screenshot args include quality, full_page, and optional path. Without path, the screenshot is saved as a chat-scoped artifact and returned through vision_load.tool_args.paths; with path, PNG is used when path ends with .png, otherwise JPEG is used.
Forms And Files
select_optionworks for native selects and detectable ARIA listbox/combobox controls.set_checkedworks for checkbox, radio, switch, and toggle-like refs.upload_fileworks for file input refs or associated labels; verify the file exists in the active browser environment.- For fragile forms, call
skills_toolwithaction: "load"andskill_name: "browser-form-workflows", then follow that form-specific workflow before filling or submitting.
Pointer And Keyboard
- Prefer refs/selectors and DOM/CDP actions over viewport coordinates.
hover,double_click,right_click, anddragaccept refs or viewport coordinates when no reliable ref exists.- Coordinates are Chromium viewport CSS pixels and match screenshots; treat them as visual fallback, not the default interaction path.
key_chordpresses keys in order and releases in reverse.clipboardactions are copy, cut, or paste.set_viewportresizes the page viewport.
Tabs And Popups
- Popups and target-blank tabs are auto-registered.
listshows open tabs; passinclude_content: truesparingly.set_activedeliberately changes focus.- Operations on a non-active tab do not steal focus unless browser rules require it.
Browser Action Multi
multi is only a browser action, never a top-level tool. Use:
{
"tool_name": "browser",
"tool_args": {
"action": "multi",
"calls": [
{"action": "content", "browser_id": 1},
{"action": "screenshot", "browser_id": 2}
]
}
}
Use browser action multi for parallel reads across tabs. Avoid mutating the same tab twice in one batch unless serial order is intended.