| name | visible-browser-lab |
| description | Use for lease-scoped visible browser navigation, semantic page interaction, diagnostics, emulation, performance, audits, memory analysis, screencasts, and browser artifacts. |
Visible Browser Lab
Start each browser task with start_session and retain its agent_session_id. Use only tab_id values owned by that session.
Inspect an unfamiliar page with snapshot, then act through its element references. Use fill to replace one ordinary field. Use fill_form for two or more controls, including one operation that combines select and checkbox updates. Use type_text for contenteditable controls and insertion at an established caret; it preserves application focus and does not require focus_tab. Use press_key only for named keys or shortcuts after focus_tab.
Use wait_for for asynchronous state and screenshot for visual appearance. Use console and network for runtime diagnosis. Use help to select an operation in a specialized domain.
click, press_key, and native pointer operations return focus_required until focus_tab makes the owned document visible and focused. CSS and evaluate are escape hatches for state the accessibility snapshot and named semantic tools cannot represent. Do not use them to verify a semantic action.
Use performance, audit, memory, screencast, and artifacts for their corresponding investigation outputs. Release a tab that another agent may continue, or close a tab whose browser work is complete.