← Back to Browser
OpenClaw browser

How browser actions work

Use this page when you need the core OpenClaw browser interaction model before building a workflow.

Core idea Reliable browser work starts with the right tab, a fresh snapshot, and one ref-based step at a time.
Interaction model

Tabs, snapshots, and refs beat brittle selector guessing

OpenClaw browser automation works best when you think in tabs, snapshots, refs, and targeted actions rather than improvised one-off selectors.

The reliable loop is simple: target the right tab, inspect the current page with a snapshot, use the returned refs for the next action, then re-check before continuing.

Stable targeting

Use the right tab anchor before any click happens

  • prefer stable follow-on targets such as suggested target IDs, tab IDs, or readable labels
  • do not assume a raw browser target ID will survive every navigation or form submission
  • refresh refs after meaningful page changes
Screenshots vs snapshots

Use the right artifact for the question

Action map

Snapshot

Shows the current interactive structure and yields refs for the next step.

Visual proof

Screenshot

Shows what the page looked like for review or reporting.

Hybrid proof

Labeled screenshot

Shows the visual state and the current refs together.

Next route

Read the right neighbor page next