feat: add playwright-cli skill

[royosherove]

Browser automation skill using @playwright/cli. Direct CLI — no MCP protocol needed.

[github-actions]

Claude finished @royosherove's task in 2m 2s —— View job

---

PR Review: feat: add playwright-cli skill

• Gather context
• Read repo structure and other SKILL.md patterns
• Review playwright-cli/SKILL.md
• Post review feedback

---

Summary

The skill is well-structured and clearly written — good command coverage, sensible workflow guidance, and useful tips. A few issues worth addressing before or after merge:

---

Issues

1. Potentially incorrect package name — Medium severity

playwright-cli/SKILL.md:13

npm install -g @playwright/cli

The standard Playwright package is playwright (or @playwright/test), not @playwright/cli. The @playwright/cli scoped package may not exist or may differ from what's described. If the binary is installed via playwright or @playwright/test, the setup instructions will fail.

Fix this →

---

2. README.md not updated — Low severity

The new skill is not listed in the skills table in README.md. All other skills appear there under either the upstream or custom section. This makes the skill harder to discover.

Fix this →

---

3. Frontmatter is incomplete compared to other skills — Low severity

playwright-cli/SKILL.md:1-4

Other skills (e.g., neon, checkout) include displayName, version, author, and keywords in frontmatter. This skill only has name and description. Missing fields may affect skill discovery or display in consuming systems.

Example from neon/SKILL.md:

name: "neon"
displayName: "Build a database with Neon"
description: "..."
version: "..."
author: "..."
keywords: [...]

Fix this →

---

4. Undocumented commands referenced only in Tips — Minor

playwright-cli/SKILL.md:134-135

Two commands appear only in the Tips section without a proper entry in the Commands reference:

• playwright-cli eval "document.title" — arbitrary JS execution, significant capability
• playwright-cli resize 1920 1080 — viewport sizing

eval especially deserves a dedicated section given its power and the security reminder that it runs arbitrary JS.

---

What's good

• Clear open→interact→close session model reduces agent confusion
• The [ref=eN] accessibility-tree targeting approach is explained well and the "refs change on re-render — don't cache" tip is exactly the right guardrail
• Parallel session support with -s=name is documented and exemplified
• Auth state save/load pattern is a good practical inclusion
• --json output format flag is called out for programmatic use cases

Overall: solid skill doc. The package name issue is the most important one to verify before agents run the setup step.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 91d5208c10

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Use the documented browser install subcommand

The setup step currently tells users to run playwright-cli install to download Chromium, but Playwright CLI’s installation docs use playwright-cli install-browser for browser binaries and reserve install --skills for skill files. In a fresh environment this instruction can leave the browser uninstalled, so subsequent open/automation commands fail despite following the skill exactly.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Remove unsupported network response command

The Network section documents playwright-cli response-body 3, but the published CLI command set lists request inspection via requests and request <index> and does not expose a response-body subcommand. Users following this guide will hit an immediate CLI error when trying to inspect responses, breaking the advertised network-debugging workflow.

Useful? React with 👍 / 👎.