cdp-browser skill

Install this file in your repo as .agents/skills/cdp-browser/SKILL.md .

Use this skill when a task needs real-browser execution rather than static HTTP fetching.

Quick workflow

bunx cdp-browser start bunx cdp-browser nav https://example.com --new bunx cdp-browser wait-network-idle bunx cdp-browser dismiss-cookies bunx cdp-browser eval "document.title" bunx cdp-browser screenshot bunx cdp-browser watch bunx cdp-browser logs-tail --follow bunx cdp-browser net-summary

Optional version pinning

bunx cdp-browser@0.1.3 nav https://example.com

Default operating pattern

• Start a managed browser session.

• Navigate to the target page and wait for network idle.

• Remove overlays when needed (dismiss-cookies ).

• Extract data with eval ; use pick when selector discovery is needed.

• Capture artifacts (screenshot , logs-tail , net-summary ) when requested.

• Return results with all generated file paths.

Commands

• start [--fresh] [--copy-profile [name]] [--browser <path-or-name>]

• nav <url> [--new]

• eval '<expression>'

• screenshot

• pick '<message>'

• dismiss-cookies [--reject]

• watch

• logs-tail [--file <path>] [--follow]

• net-summary [--file <path>]

• wait-network-idle [--timeout <ms>] [--idle-time <ms>] [--max-inflight <count>]

Response expectations

• Include exact paths for any generated artifacts.

• Include the failing command and concise stderr when a command errors.

• Keep extraction steps deterministic and reproducible.

Troubleshooting

• If connection fails, verify http://localhost:9222/json/version .

• If browser detection fails, run start --browser <path-or-name> .

• Use --fresh for a clean session.

• Use --copy-profile [name] for logged-in state.

• Use CDP_BROWSER_BASE_DIR to control managed data paths.