Skip to content

playwright-cli install --skills downloads browsers unnecessarily #295

@inertjade

Description

@inertjade

Description

playwright-cli install --skills downloads Chromium even when only skill files (markdown) are needed. This breaks CI/CD pipelines that have no network access to cdn.playwright.dev.

Steps to reproduce

PLAYWRIGHT_SKIP_BROWSER_DOWNLOAD=1 playwright-cli install --skills

Expected: Only installs skill markdown files to .claude/skills/playwright-cli/.
Actual: Also downloads Chromium (~174MB) from cdn.playwright.dev. The PLAYWRIGHT_SKIP_BROWSER_DOWNLOAD env var is ignored.

Error in CI

> PLAYWRIGHT_SKIP_BROWSER_DOWNLOAD=1 npx playwright-cli install --skills
✅ Workspace initialized at `/local/p4clients/pkgbuild-const/workspace/src/Kairos`.
✅ Skills installed to `.claude/skills/playwright-cli`.
BEWARE: your OS is not officially supported by Playwright; downloading fallback build for ubuntu24.04-x64.
Downloading Chrome for Testing 146.0.7680.0 (playwright chromium v1212) from https://cdn.playwright.dev/...
Error: getaddrinfo ENOTFOUND cdn.playwright.dev

Use case

We use playwright-cli in an LLM-driven test runner where the system prompt includes SKILL.md content. The skills need to be installed at build time (npm postinstall), but the browser should only be downloaded at runtime on the test host.

Proposed solution

Either:

  1. playwright-cli install --skills should not trigger browser download (only copy markdown files)
  2. Respect PLAYWRIGHT_SKIP_BROWSER_DOWNLOAD=1 env var during install --skills
  3. Add a --no-browsers flag: playwright-cli install --skills --no-browsers

Workaround

We commit the skill files directly to git and skip the install --skills step in CI.

Version

@playwright/cli 0.1.1

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions