browserbase-sessions_skill

Overview

This skill creates and manages persistent cloud browser sessions using Browserbase, preserving authentication state (cookies and storage) and recording sessions for later review. It automates browser actions, solves CAPTCHAs by default, and provides live remote-control links so you can interact manually when needed. Use it to keep logged-in sessions across chat turns, scrape authenticated pages, or run multi-site automated workflows.

How this skill works

The skill talks to Browserbase via a manager script that creates sessions, contexts, and workspaces. Workspaces combine a persisted Browserbase Context (auth state) with tab snapshots so the agent can stop and later resume where you left off. Commands perform navigation, click/type interactions, screenshots, recordings, and handoffs for required human steps (SSO/MFA/captchas).

When to use it

• Automating interactions on sites that require login and need persistent auth between runs.
• Scraping or extracting data from pages behind authentication or SSO flows.
• Maintaining a long-running research or task workspace across chat turns (multi-site workflows).
• When manual intervention is required (MFA/consent) and you need a live browser handoff.
• Testing and debugging complex browser flows with replayable session recordings.

Best practices

• Provide BROWSERBASE_API_KEY and BROWSERBASE_PROJECT_ID before requesting browser actions.
• Prefer workspaces (create/start/resume/stop) to keep browser state and tab snapshots across turns.
• Use handoff with explicit completion checks for any manual SSO/MFA/captcha steps.
• Stop workspaces (not terminate sessions) to persist tabs and auth state.
• Install dependencies via the provided install command when ImportError appears (browserbase/playwright).

Example use cases

• Create a workspace per app (e.g., github) to keep separate logged-in profiles and restore tabs later.
• Run an invoice-run workspace that opens multiple sites, performs actions, and preserves session state between steps.
• Resume a stopped workspace so the agent can continue a scraping job without re-authentication.
• Hand off a live URL to the user to complete SSO, then automatically detect completion and continue.
• Take screenshots and extract page text from authenticated pages for reporting or audits.