Browser Execution completes direct browser tasks with Playwright, computer controls, secure vault handling, recovery tactics, and an explicit verified result.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install browser-execution
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install browser-execution using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Browser Execution?
Browser Execution is an Openclaw Skills workflow for completing browser-based tasks reliably through Kernel browser sessions. It combines Playwright for navigation, inspection, extraction, and deterministic interaction with computer controls for visual reasoning and coordinate-level input.
The skill emphasizes bounded execution, secure secret handling, human approval for transactions, and clear terminal reporting. It reuses one browser for the full job, responds to blocked pages with limited tactic changes, and calls complete_task exactly once with a verified success or failure outcome.
Browser Execution Use Cases
- Complete form submissions using user-provided names, contact details, addresses, and other non-credential values.
- Navigate websites, inspect page content, extract information, and interact with deterministic controls.
- Use screenshots and computer actions when visual reasoning or coordinate-level interaction is more reliable.
- Fill saved login, payment, address, phone, identity, or token data without exposing raw secrets to browser calls.
- Handle blocked pages, timeouts, authentication challenges, and other browser execution failures with bounded recovery.
- Complete approved purchases or other transactions while validating the merchant, item, quantity, options, fees, and total.
- Provide an explicit success or failure result instead of leaving a task in an open-ended state.
How Browser Execution Works
- Read the current Kernel documentation for browser sessions, Playwright execution, computer controls, and live view rather than relying on assumed API shapes.
- Create one browser with
manage_browsersand reuse it throughout the task. - Prefer
execute_playwright_codefor navigation, DOM inspection, extraction, and deterministic interaction; usecomputer_actionwhen visual or coordinate-based control is more dependable. - Navigate with
domcontentloadedor wait for a specific locator, URL, response, or visible state, keeping waits bounded and avoiding fixed sleeps ornetworkidle. - Use directly supplied non-secret form values as provided. For saved secrets or saved identity data, inspect fields first, obtain an opaque handle with
list_vault, and inject values withfill_from_vault. - If a page is blocked, try no more than two materially different relevant approaches, such as a direct provider URL, a different interaction method, or a fresh tab.
- For transactions, request approval once for the exact payload and quoted total before filling payment secrets, then submit in the same run after approval.
- Preserve the browser when human approval or authentication completion is the only remaining blocker; otherwise delete it when the task is complete.
- Call
complete_taskexactly once, usingsuccessonly for an achieved and verified result orfailurefor a blocked, incomplete, or failed task, then return the same terminal message.
Browser Execution Setup
- Make the Kernel browser execution tools available to the Openclaw Skills runtime, including
manage_browsers,execute_playwright_code,computer_action,list_vault,fill_from_vault, andcomplete_task. - Review the current Kernel documentation before implementation:
- Browser sessions: https://kernel.sh/docs/introduction/create
- Playwright execution: https://kernel.sh/docs/browsers/playwright-execution
- Computer controls: https://kernel.sh/docs/browsers/computer-controls
- Live view: https://kernel.sh/docs/browsers/live-view
- Configure vault entries only when saved secrets or saved identity data are required. Supported setup kinds are
login,payment,address, andphone. - Start each task by creating a browser and keep the fast path bounded to approximately 90 seconds and six browser tool calls for uncomplicated work.
No package installation command is specified by the skill; use the current Kernel integration and documentation as the source of truth for runtime setup.
Browser Execution Data Schema & Taxonomy
Browser task state
| Area | Organization and rules |
|---|---|
| Browser session | One browser is created and reused for the complete job; it is deleted after completion unless a human action or approval remains necessary. |
| Interaction methods | execute_playwright_code handles navigation, inspection, extraction, and deterministic actions; computer_action handles visual or coordinate-level interaction. |
| Task result | complete_task is called exactly once with success or failure, followed by the same terminal message. |
| Timing metadata | The fast path targets 90 seconds and six browser calls; Playwright calls have a 30-second ceiling, locator waits are at most five seconds, and computer-action sleeps are at most two seconds. |
| Recovery state | A blocked page permits up to two materially different approaches before reporting the verified blocker. |
Vault taxonomy
- Supported vault item kinds:
login,payment,address, andphone. request_vault_setupacceptskind, optionallabel, optionalaccount, andtarget.fill_from_vaultsupportsusername,password,cardholder_name,card_number,expiration,expiration_month,expiration_year,cvc,billing_postal_code,address,phone,identity, andtoken.- Raw secrets must never be placed in a normal browser call. After vault injection, the skill must not inspect filled values or capture screenshots that could expose them.
- User-provided non-credential values can be used directly and do not need to be stored in the vault first.
Browser Execution Advanced Features
- Secure opaque-handle vault integration for login credentials, payment cards, addresses, phone numbers, identities, and tokens.
- Transaction approval gating that binds approval to the exact merchant, item, quantity, option, fees, and quoted total.
- Automatic continuation after approval, including vault filling and submission in the same run.
- Bounded blocked-page recovery with materially different tactics instead of repeated selector or code retries.
- Dual interaction strategy combining Playwright automation with computer controls and final screenshots when visual confirmation is useful.
- Human-in-the-loop preservation when authentication challenges, approval, or another manual action is required.
- Strict timeout, wait, browser-call, and terminal-result controls for predictable Openclaw Skills execution.
- Explicit distinction between verified success, incomplete work, and exact environmental blockers.
SKILL.md
Loading
Related Openclaw Skills
HTML/CSS to Image API
Render HTML, CSS, public webpages, and reusable templates into high-quality images or PDFs through HCTI MCP tools.
Chuhaijiang TikTok Live Commerce Intelligence
A read-only TikTok live commerce research skill for discovering livestreams, analyzing session performance, and identifying products sold through Chuhaijiang.
Chuhaijiang TikTok Product Market Intelligence
A public TikTok Shop market intelligence skill for discovering products, analyzing performance, tracing promotion channels, and finding visually similar items.
Cargo CLI — Segmentation
Cargo CLI Segmentation lets you define, save, inspect, export, and monitor named audiences over Cargo models.
Render on Hermes
Render on Hermes configures an AI agent to manage Render through its pre-registered MCP server without installing the Render CLI.
QingTian Free Psychological Resources
A multilingual OpenClaw skill for discovering free psychological support, crisis resources, AI companions, and self-assessment tools worldwide.
Featured*
