domains / agent-tooling / browser-automation / browser

browser

Skill card

combined

The two copies describe different browser automation stacks. The browserbase/skills copy is a detailed procedure for the browse CLI with local/remote modes, CDP tailing, and a large command surface. The LeoYeAI copy is a minimal Puppeteer-based skill that only documents a read command. They overlap in the goal of browsing and extracting content, but each contains material steps and constraints the other lacks, and they disagree on the underlying tooling, so crowning one would hide a real alternative.

Repositories

Other copies

Combined skill

# Browser Automation

Automate web browsing to navigate pages, interact with elements, and extract content. Choose the tooling that matches the environment.

## Option A: browse CLI (Browserbase/local/CDP)

Use when the browse CLI is available or can be installed.

1. Verify the CLI is available:
   ```bash
   which browse || npm install -g browse
   ```
2. Choose the environment:
   - Local clean state: `browse open <url> --local`
   - Reuse a running debuggable Chrome: `browse open <url> --auto-connect`
   - Attach to a specific CDP target: `browse open <url> --cdp <port|url>`
   - Remote Browserbase session: `browse open <url> --remote` (requires `BROWSERBASE_API_KEY`)
3. Prefer `browse snapshot` to understand page state; use `browse screenshot --path <path>` only when visual context is needed.
4. Interact using snapshot refs or selectors:
   - `browse click <ref>`
   - `browse type <text>`
   - `browse fill <selector> <value>`
   - `browse select <selector> <values...>`
   - `browse press <key>`
   - `browse wait <type> [arg]`
5. Extract content with `browse get text`, `browse get html`, `browse get markdown`, or `browse get value`.
6. Stream CDP events when needed with `browse cdp <url|port>`.
7. Stop the daemon with `browse stop`.

## Option B: Puppeteer via openclaw

Use when the openclaw browser command is available.

1. Render a page and return its text content:
   ```bash
   openclaw browser read <url>
   ```

## Conflicts

- browserbase/skills::skills/browser/SKILL.md uses the browse CLI with local, remote, and CDP modes; LeoYeAI/openclaw-master-skills::skills/browser/SKILL.md uses Puppeteer and only documents `openclaw browser read <url>`.
- browserbase/skills::skills/browser/SKILL.md requires the browse binary and optionally `BROWSERBASE_API_KEY`; LeoYeAI/openclaw-master-skills::skills/browser/SKILL.md requires Puppeteer and the openclaw browser command.