A browser session lifecycle an agent can reason about
Create, inspect, connect and close Cloudflare Browser Run sessions without confusing access links with persistent browser state.
Lupin field guides
Browsers
Browser sessions, page actions and authenticated automation.
20 published articles
Choose browser selectors that fail safely
Use unique visible controls, inspect fresh page state and avoid turning selector recovery into repeated form submissions.
Choose research or an interactive browser session
Choose temporary public research for evidence gathering and a managed session when a task needs page state or human interaction.
Close browser sessions even when the task fails
Make browser cleanup observable, distinguish a closed page from an ended session and reconcile uncertain provider outcomes.
Control browser spending through task design
Bound runtime, avoid duplicate allocation and close idle work while keeping search action costs separate from browser seconds.
Handle browser access links as credentials
Protect session-scoped CDP and live-view links, refresh them safely and understand their expiry and revocation limits.
Keep browser actions attached to the right tab
Use tab identifiers, inspect popups and manage a small tab set without relying on tab order.
Keep browser evidence inside useful output bounds
Plan for truncated text, capped controls and bounded image results instead of assuming every page fits in one tool response.
Make a human sign-in handoff predictable
Pause automation, guide a private live-view sign-in and verify the returned account before continuing an authorized task.
Map the authentication boundaries around a browser
Separate Lupin API permission, session access and website sign-in so agents do not assume one credential grants every kind of access.
Migrate browser workflows without pretending sessions move
Plan a provider cutover around supported contracts, fresh sessions, human reauthentication and verified cleanup.
Plan concurrent browser work with separate sessions
Choose isolation boundaries, avoid conflicting controllers and respect workspace and provider capacity.
Preserve source provenance in browser research
Keep search results, opened pages, retrieval times and agent conclusions distinct so research can be reviewed.
Read public paginated pages with a stopping rule
Collect allowed public records with explicit page identity, deduplication and a finite budget.
Retry browser work without repeating unknown effects
Classify validation, permission, provider and uncertain-action failures before deciding what can be retried.
Separate browser timeouts and deadlines
Distinguish action deadlines, access-link startup expiry, inactivity and runtime leases when diagnosing browser failures.
Treat web page instructions as untrusted content
Keep a website’s text separate from the user’s authority when researching or acting through a browser.
Understand what a browser page read contains
Interpret visible text, links and controls without mistaking a bounded page snapshot for the entire DOM or a structured application API.
Use browser screenshots as visual evidence
Capture bounded JPEG screenshots, pair them with page observations and avoid treating a picture as proof of an action outcome.
Wait for page readiness without endless sleeps
Use bounded observations to distinguish a loaded document from an application that has finished presenting the needed information.