openclaw-rpa

Record browser, Excel, Word, and API actions once, then replay them as standalone Playwright Python scripts.

Updated Aug 13, 2026
One-click install
npx skills add https://github.com/chenge886/Harness-RPA --skill openclaw-rpa-chenge886
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: openclaw-rpa
Source: https://github.com/chenge886/Harness-RPA
Command: npx skills add https://github.com/chenge886/Harness-RPA --skill openclaw-rpa-chenge886

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires playwright, httpx, openpyxl, python-docx, and includes scripts (resource) components.

What problem does it solve? Repeatedly driving a browser with an LLM for the same task is slow, expensive in tokens, and prone to hallucinated mis-clicks. This Skill records a workflow once in a real browser and compiles it into a deterministic Python script that replays without any model calls. ## Core Features & Use Cases - Record-to-script pipeline: Trigger with #rpa or #rpa-api, confirm each step with screenshots, then compile to a standalone Playwright Python file under rpa/ that runs with zero LLM cost. - Multi-surface automation: Supports browser actions (click, fill, extract), HTTP API calls via httpx, Excel (.xlsx via openpyxl), Word (.docx via python-docx), and vision-based extraction for SPA pages. - Session reuse: Save login cookies once with #rpa-login and auto-inject them into future recordings and replays to skip OTP, CAPTCHA, and QR-code flows. - Use Case: Record an Amazon best-sellers scrape (top 40 products with title, price, rating) once, then replay it daily via #rpa-run:amazonbestseller to append results to a Word report without any token spend. ## Quick Start Send #rpa-list to see registered tasks, then run one with #rpa-run: followed by the task name, or send #rpa to start recording a new automation.

Frequently Asked Questions about openclaw-rpa

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I record a browser automation and replay it without an LLM?▼

Send #rpa to start onboarding, provide a task name plus a capability letter (A-G or N), then describe the task. Each step executes in a real Chrome window with screenshot confirmation; sending #end compiles the steps into a standalone Playwright Python script you replay with rpa_manager.py run.

How do I automate a workflow that includes an HTTP API call?▼

Send #rpa-api and describe the API or paste its documentation parameters inside a ### block. The recorder adds an api_call step using httpx, embeds the API key directly into the generated script, and the replay runs without any model involvement.

Can Playwright automation handle sites that require login with 2FA or CAPTCHA?▼

Yes, through cookie reuse. Run #rpa-login with the login URL, complete the login manually once, then send #rpa-login-done to export cookies. Future recordings and replays inject those cookies automatically via #rpa-autologin, skipping the login flow.

Does this work on Windows, macOS, and Linux?▼

Yes, all three platforms are supported. The stack is Python 3 plus Playwright, httpx, openpyxl, and python-docx, all of which have official cross-platform support, and Excel or Word files are generated without requiring Microsoft Office.

Why does extraction fail on dynamic single-page applications like Airbnb?▼

Heavy SPAs use hashed CSS class names that change between builds, so DOM selectors break. For these pages the recorder switches to extract_by_vision, which screenshots the page and uses the Qwen3-VL vision model to read data directly from the screen.

What are the limitations of record-and-replay browser automation?▼

It is not suited for large ETL, databases, or heavy OS automation. Sites with CAPTCHA or bot detection still require manual intervention, and page redesigns can invalidate recorded selectors, requiring re-recording of the flow.