openclaw-rpa — Locale router (read this first)

GitHub: https://github.com/laziobird/openclaw-rpa — source, README, install, sample scripts under rpa/.

What this skill does

openclaw-rpa is an LLM-based RPA Agent framework. You describe a task in plain language; the AI executes it step by step in a real browser, on your computer, or via API services — with screenshot proof at every step — then compiles everything into a standalone Playwright Python script. Replay runs that script directly — no model call, no token burn, no hallucination risk — faster and cheaper than having the AI click every time.

Why this matters

1. Saves compute and money — Having a large model drive the browser on every run can cost roughly single-digit to tens of US dollars per heavy session (tokens, tools, long context). After you record once, repeat work does not call the model—replay is much faster and near-zero LLM cost for those steps.
2. Verify once, run the same way every time — During recording you confirm the flow works; later, replay executes the saved steps deterministically. You avoid asking the AI to “do it again” on every run, which hurts consistency and raises hallucination risk.

What you can automate (record once, replay many times — follow each site’s terms and local law):