Title: claude-flow/v3/@claude-flow/browser/README.md at main · ruvnet/claude-flow

Claude-flow/browser is a technical framework that extends AI agent capabilities into web automation and data extraction. The system combines intelligent browser control with trajectory learning—a pattern-recording mechanism where agents document successful interaction sequences for reuse across similar tasks. This learning capability means agents improve over time by building an internal knowledge base of effective workflows.

The platform emphasizes security throughout, with automatic scanning for phishing threats, personally identifiable information (PII), and malicious code injection attempts before any web interaction occurs. This is particularly important for enterprise teams automating sensitive processes like account access or data entry.

A standout feature is element referencing—instead of complex technical selectors, agents use simple labels like '@e1' and '@e2' to interact with page elements. This reduces context overhead by 93% and makes automation patterns more portable across different websites. The system also supports parallel agent coordination through 'swarms,' allowing multiple AI agents to work simultaneously on distributed web tasks like data collection or testing.

While powerful, the implementation requires developer setup including Node.js, TypeScript knowledge, and command-line tools. Pre-built workflow templates address common scenarios (login flows, form submission, data scraping) that teams might adopt directly, but customization demands technical expertise.