No need for playwright/puppeteer.
Script the browser from the command line and let your LLM do the rest.
npm install -g playd
playd session create my-run --headed
playd goto https://example.com
playd llm "Does this page have a search input? What's its ID?" --on-html
# npx -y playd ← works tooRequired for llm command: Set your OpenAI API key
export OPENAI_API_KEY=sk-...Press # in Claude Code (or similar) and paste this line to store playd usage instructions:
Information about `playd` command: it controls Chrome via CDP; create session: `playd session create browser1 --headed` (other commands use last session by default); navigate: `playd goto URL`; interact: `playd click ".selector"`, `playd type "text"` or `playd type "text" --selector ".input"`, `playd type "text" --clear` (clears first), `playd fill ".selector" "text"`, `playd press "Enter"`; wait: `playd wait-for ".selector"`; extract: `playd text`, `playd dom`, `playd attr ".selector" "href"` (html attr, not current value!); AI questions: `playd llm "question"` (analyzes title+text) or `playd llm "question" --on-html` (analyzes full HTML); run JS: `playd eval "code"`; capture: `playd screenshot`; cookies: `playd cookie-get <name>`, `playd cookie-set <name> <value> [--domain <domain>] [--path <path>]`; storage: `playd storage [--export file.json] [--import file.json]`; `playd session close browser1`; `playd wait-for "textarea[name='q']"`; `playd help`; `playd shutdown` (closes all sessions) `npm install -g playd` if not installedplayd session create demo --headed
playd goto https://news.ycombinator.com
playd llm "Give me 3 bullet points from the page"
playd session close demo
playd session create browser1
playd goto https://google.com
playd wait-for "textarea[name='q']
playd click "textarea[name='q']
playd type "search"
playd press Enter
playd llm "what is this page about, summarize in 2-3 sentences"
playd llm "what query was used" --on-html
playd shutdown # closes all sessions