Getting Started

Getting started

Install

The plugin, for Claude Code:

git clone https://github.com/m4bwav/brandwright
claude plugin marketplace add ./brandwright
claude plugin install brandwright@brandwright

Other agents: link or copy the four folders under skills/ into the agent's skills folder. On its own, the CLI is python skills/brandwright/scripts/bw.py. Below, BW stands for that.

What each part needs:

Part	Needs
new, tokens, train (config only), status, pick	Python 3.10+
login, capture, render	python -m pip install playwright, then --channel msedge or --channel chrome to use an installed browser, or python -m playwright install chromium
prep, score, sheet	python -m pip install pillow
gen, sweep, ref	a ComfyUI server with the model files (Training)
train --run	ostris ai-toolkit and an NVIDIA GPU
caption	Ollama with a vision model (optional)

BW doctor shows what this machine has.

A first kit

Keep the kit outside every repository: it will hold your company's screenshots.

BW new D:\kits\acme --brand "Acme" --site "portal=https://portal.acme.example/" --site "careers=https://careers.acme.example/"

The brand and each site get a trigger word, for example acmeui and acmeportalui. The kit gets a .gitignore of *.

Pictures in

Pick one per site:

BW capture D:\kits\acme --site careers --crawl 15 --channel msedge                    # public pages
BW login D:\kits\acme --site portal https://portal.acme.example/ --channel msedge      # you sign in, then close the window
BW capture D:\kits\acme --site portal --crawl 15 --channel msedge                     # reuses that session
BW import D:\kits\acme --site portal D:\screenshots\portal                            # your own screenshots

The first mock-up (code route)

BW tokens D:\kits\acme

This writes tokens.json, brand.css, tokens.md and palette.png. Ask your agent for a mock-up, for example "a new order-tracking card for the portal". It writes an HTML page that links brand.css and renders it:

BW render D:\kits\acme\out\tracker.html --width 1440 --height 900

Then pictures and LoRAs

See Dataset and captions, Training and Mock-ups, or the full run in Recipes.