web-browsinglisted

# Web Browsing `browse` drives real Chrome, locally or on Browserbase. Use it for any real webpage. WebFetch is for raw files, JSON APIs, and plain text only, and is never retried after returning empty content or a JavaScript shell. Run `browse skills show` before the first browse command in a session and follow it for command mechanics. These rules apply on top. - Pass the mode explicitly on every `browse open`. Use `--local` on a machine with Chrome. Use `--remote` when no local Chrome exists or the site needs Browserbase infrastructure. - Give each independent task its own `--session <name>`. When `BROWSE_SESSION` is set, every command already targets that session. Pass no `--session`, and never stop it. The user owns it. - Stop sessions you created when their task ends. - Remote sessions need `BROWSERBASE_API_KEY` in the environment. Never print it, and never export a secret yourself. Sessions never need a project ID. - Bot-protected sites need paid Browserbase features. Say so plainly and offer another source instead of retrying a block page. - If the CLI is missing, do not install it. Fall back to WebFetch, complete the task as static HTML allows, and note the limitation in the report. ## Remote budget Remote sessions spend a metered monthly budget. The numbers below describe the owner's current plan (100 browser hours a month). Scale them to the account's actual plan when it differs. Local Chrome is free, which is one more reason to prefer it. Be