playwright-cli — agent browser control on cluster nodes

Playwright CLI (@playwright/cli) is a daemon-based browser driver designed for coding agents: the first command starts a persistent browser process, every command prints a trimmed page state plus an accessibility-tree snapshot file with element refs (f1e10), and follow-up commands operate on those refs deterministically.

Upstream usage doc is the single source of truth for command syntax — read it before writing any command you have not used before:

• ~/tools/playwright-cli/node_modules/@playwright/cli/skills/playwright-cli/SKILL.md (all commands, quick start)
• Same directory references/: session-management.md, storage-state.md, tracing.md, running-code.md (one file per topic, load only the one you need)

This skill adds what the upstream doc does not know: running headless on a proxied cluster node, attaching observation tooling to the daemon, and the raw-script fallback.

Setup facts (this node)

• Binary: ~/tools/playwright-cli/node_modules/.bin/playwright-cli (npm-installed latest; version check in scripts/check-env.sh)
• Playwright library for scripts: require('~/tools/playwright-runner/node_modules/playwright') (v1.63, chromium headless shell in ~/.cache/ms-playwright)
• Cluster nodes require --no-sandbox; internal portals use self-signed certs (ignoreHTTPSErrors)

Node pitfalls (verified by experiment — do not rediscover)