Basic browser actions

Basic actions use agent-browser command names directly. Tasks compose these commands with current screen targets and input values. Filtering and message preview are sequences of these same commands.

Observe and interact

1. Open the target page.
2. Read interactive elements with snapshot -i.
3. Match role, name and record context, then act on the ref.
4. Inspect a new snapshot after changes. Read results with get or is and compare with expectations.

Do not reuse refs after navigation or element replacement. Surviving DOM elements can retain refs, but documentation never stores them. Use full snapshots for content and annotated screenshots for visual context.

Choose a task from feature and prerequisite sequences.

Basic commands

• open: Open a page
• snapshot: Read the screen and refs
• click: Click an element
• fill: Replace an input value
• press: Press a key
• select: Choose a native select value
• check: Check a checkbox
• uncheck: Uncheck a checkbox
• wait: Wait for a condition
• get: Read a displayed value
• is: Read an element state
• reload: Reload the page
• upload: Upload a file
• screenshot: Capture the screen
• record: Record the screen

Targeting and verification

Scripts can use a semantic locator when its accessible name is unique, such as find role button click --name "Name" --exact. Agents interpreting current screens prefer snapshot refs. Use CSS or positional selectors only after inspecting the actual screen.

For searchable comboboxes, click the input, press the required key, inspect a new snapshot and click the option. Use check and uncheck for desired checkbox state.

get and is return values. Compare them with task expectations. Verify receipt and payment against the original request ID.

Testing and recording

Execute the same sequence with fixed inputs and expected values. Assert observed results and stop on failure. Surround execution with record start and record stop for video evidence. A recording alone does not establish success.

External approvals, phone calls and terminal work identify the operator and evidence at the relevant point. They are not browser commands. Ordinary UI interaction needs no custom business-command plugin.

Official references: Selectors, Snapshots, Commands, Recording.