Skip to main content
SCREENWEB AGENTS

Browser Navigation and Web Task Flows

Web tasks are sequences of decisions made against changing page state. Browser Navigation and Web Task Flows preserve that sequence, connecting navigation, interaction and outcome rather than treating each webpage as a standalone visual example.

The collection can represent clicks, scrolling, form interaction and page transitions alongside the screen recording and task context. Where visible text matters, OCR-derived page text can add another layer of supervision for models that need to reason about both what is shown and what action should follow.

Technical structure

  • Browser sessions organised around task objectives and outcomes
  • Navigation and interaction events linked to the relevant page state
  • Page or screen text represented where it contributes to the task
  • Temporal records connecting decisions to subsequent state changes

Typical applications

Web agents, browser automation, navigation modelling, form interaction and multi-step task evaluation.

Delivery note

Task design, browser context and event granularity are defined around the behaviour the agent needs to perform.

Delivery

The exact package depends on the collection scope. Where relevant, delivery can include task-specific records, manifests, provenance fields, stable identifiers, SHA-256 hashes, MLCommons Croissant 1.0 metadata and loading instructions.

Related datasets

Explore all collections