BrowserSAFE
Browserbase's official collection of agent skills to access the web.
Overview
Browserbase's official collection of agent skills to access the web.
45d21dc75974OBSERVED · 2026-10-22Host compatibility
What the documentation claims. We have not run a compatibility test.
| Host | Status | Notes |
|---|---|---|
| openclaw | mentioned |
What it tells the agent
The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.
---
name: browser
description: Automate web browser interactions using natural language via CLI commands. Use when the user asks to browse websites, navigate web pages, extract data from websites, take screenshots, fill forms, click buttons, or interact with web applications. Supports remote Browserbase sessions with Browserbase Identity, Verified browsers, automatic CAPTCHA solving, and residential proxies — ideal for protected websites and JavaScript-heavy pages.
compatibility: "Requires the browse CLI (`npm install -g browse`). Remote Browserbase sessions need `BROWSERBASE_API_KEY`. Local mode uses Chrome/Chromium on your machine."
license: MIT
allowed-tools: Bash
metadata:
openclaw:
requires:
bins:
- browse
install:
- kind: node
package: "browse"
bins: [browse]
homepage: https://github.com/browserbase/skills
---
# Browser Automation
Automate browser interactions using the browse CLI with Claude.
## Setup check
Before running any browser commands, verify the CLI is available:
```bash
which browse || npm install -g browse
```
## Environment Selection (Local vs Remote)
The CLI supports explicit per-command environment flags. If you do nothing, the next session defaults to Browserbase when `BROWSERBASE_API_KEY` is set and to local otherwise.
### Local mode
- `browse open <url> --local` starts a clean isolated local browser
- `browse open <url> --auto-connect` attaches to an already-running debuggable Chrome; use `--local` when no debuggable Chrome is available
- `browse open <url> --cdp <port|url>` attaches to a specific CDP target
- Best for: development, localhost, trusted sites, and reproducible runs
### Remote mode (Browserbase)
- `browse open <url> --remote` starts a Browserbase session
- Without a local flag, Browserbase is also the default when `BROWSERBASE_API_KEY` is set
- Provides: Browserbase Identity, Verified browsers, automatic CAPTCHA solving, residential proxies, session persistence
- **Use remote mode when:** the target site has bot detection, CAPTCHAs, IP rate limiting, Cloudflare protection, or requires geo-specific access
- Get credentials at https://browserbase.com/settings
### When to choose which
- **Repeatable local testing / clean state**: `browse open <url> --local`
- **Reuse your local login/cookies**: `browse open <url> --auto-connect`
- **Simple browsing** (docs, wikis, public APIs): local mode is fine
- **Protected sites** (login walls, CAPTCHAs, anti-scraping): use remote mode
- **If local mode fails** with bot detection or access denied: switch to remote mode
## Commands
Most driver commands work across local, remote, and CDP sessions after the daemon starts.
### Navigation
```bash
browse open <url> # Go to URL
browse open <url> --local # Go to URL in a clean local browser
browse open <url> --remote # Go to URL in a Browserbase session
browse reload # Reload current page
browse back # Go back in history
browse forward # Go forward in history
```
### Page state (prefer snapshot over screenshot)
```bash
browse snapshot # Get accessibility tree with element refs (fast, structured)
browse screenshot --path <path> # Take visual screenshot (slow, uses vision tokens)
browse get url # Get current URL
browse get title # Get page title
browse get text <selector> # Get text content (use "body" for all text)
browse get html <selector> # Get HTML content of element
browse get markdown [selector] # Get page content as markdown (defaults to body)
browse get value <selector> # Get form field value
```
Use `browse snapshot` as your default for understanding page state — it returns the accessibility tree with element refs you can use to interact. Only use `browse screenshot` when you need visual context (layout, images, debugging).
### Interaction
```bash
browse click <ref> # Click element by ref from snapshot (e.g., @0-5)
browse type <text> # Type text into focused element
browse fill <selector> <value> # Fill input; add --press-enter if Enter is needed
browse select <selector> <values...> # Select dropdown option(s)
browse upload <selector> <files...> # Upload file(s) to <input type="file">
browse press <key> # Press key (Enter, Tab, Escape, Cmd+A, etc.)
browse mouse drag <fromX> <fromY> <toX> <toY> # Drag from one point to another
browse mouse scroll <x> <y> <deltaX> <deltaY> # Scroll at coordinates
browse highlight <selector> # Highlight element on page
browse is visible <selector> # Check if element is visible
browse is checked <selector> # Check if element is checked
browse wait <type> [arg] # Wait for: load, selector, timeout
```
### CDP event tailing
```bash
browse cdp <url|port> # Stream CDP events as NDJSON from any target
browse cdp 9222 # Attach to local Chrome on port 9222
browse cdp ws://localhost:9222/devtools/browser/... # Full WebSocket URL
browse cdp <url> --domain Network # Only Network events
browse cdp <url> --domain Network --domain Console # Multiple domains
browse cdp <url> --pretty # Human-readable output
browse cdp <url> > events.jsonl # Pipe to file
browse cdp <url> | jq '.method' # Filter with jq
```
The `cdp` command connects directly to any Chrome DevTools Protocol target and streams events. It does **not** use the daemon — it's a standalone, long-running process. Press Ctrl+C to stop. Default domains: Network, Console, Runtime, Log, Page.
### Session management
```bash
browse stop # Stop the browser daemon
browse status # Check daemon Trust audit
SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.
| Layer | What it checks | Result |
|---|---|---|
| L0 | Provenance & inventory | PASS |
| L1 | Static analysis of the code | NA |
| L2 | Instruction surface (what it tells the agent) | PASS |
| L3 | Class-specific surface | PASS |
| L4 | Behavioural (sandbox) | SKIPPED |
What the source does
- Filesystem
- none-observed
- Network
- none-observed
- Shell
- none-observed
- Dependencies
- pinned
- Secrets in source
- none-found
Findings (0)
No findings outside the package's declared scope.
Gates applied: no_behavioural_pass.
45d21dc75974full audit observations/trust-audit/skill/browserbase__browser.json · Report an issue / request a re-scanAudit history
Every audit this skill has had.
| Date | Source | Verdict | Grade | Score | Change |
|---|---|---|---|---|---|
| 2026-10-22 | 45d21dc75974 | SAFE | B | 89 | first audit |
Questions
What does the browser — browserbase/skills skill do?
Browserbase's official collection of agent skills to access the web.
Is browser — browserbase/skills safe to install?
The audit found nothing in the source that contradicts what it says it does, and graded it B (89/100). Grade A is held back for packages that have also passed a sandboxed behavioural run, which is why a clean skill reads B.
What can browser — browserbase/skills access on my machine?
The audit observed no filesystem, network or shell use at all in its source.
Which assistants does browser — browserbase/skills work with?
Its documentation mentions openclaw. That is what the text claims, not a compatibility test we ran.
How current is this page?
The grade is for one exact copy of the source (45d21dc75974), read on 2026-10-22. The repository is watched, and a new audit runs when it changes — this is the first audit.
Add this badge to your README
Shows this page's audit result and links back to it. It updates when the audit does.
[](https://modelsatlas.com/skills/browserbase/browser/)
Not your listing, or something out of date? Claim or update your listing.