pi-webwright

A small Pi package that ports Microsoft's Webwright skill into Pi.

Webwright treats browser automation as code-as-action: Pi writes and debugs Playwright scripts in a local workspace, then finishes only after a fresh rerun leaves auditable artifacts (plan.md, final_script.py, logs, and screenshots).

Install

From npm:

pi install npm:pi-webwright

From this checkout:

pi install /Users/mal/projects/pi-webwright

Or use it temporarily for one Pi session:

pi -e /Users/mal/projects/pi-webwright

Runtime prerequisites

Install Python Playwright and Firefox in the environment Pi uses for bash commands:

python3 -m pip install playwright
python3 -m playwright install firefox

Commands

• /webwright — show help and setup notes.
• /webwright:run <task> — solve a concrete browser task and leave a one-shot final_script.py.
• /webwright:craft <task> — create a parameterized reusable CLI script with argparse defaults from the task.
• /skill:webwright <task> — invoke the skill directly via Pi's built-in skill command.

Example

/webwright:run Search Google Flights for flights from SEA to JFK on 2026-08-15 to 2026-08-20

Attribution, credits, and license

This package is MIT licensed. It includes adapted skill text and reference documentation from Microsoft's Webwright repository, also MIT licensed.

The Webwright-derived files are primarily under skills/webwright/ and were modified for Pi's tool names, setup instructions, and slash-command workflow. See THIRD_PARTY_NOTICES.md for the original Microsoft copyright and MIT license notice.

Webwright's README asks that users cite the repository when using Webwright in research or building on it:

@misc{webwright2026,
  title        = {Webwright: A terminal is all you need for web agents},
  author       = {Lu, Yadong and Xu, Lingrui and Huang, Chao and Awadallah, Ahmed},
  year         = {2026},
  howpublished = {\url{https://github.com/microsoft/Webwright}},
  note         = {GitHub repository}
}

Original Webwright credits:

• SWE-agent/mini-swe-agent — design inspiration for the minimal agent loop.
• Playwright — browser automation.