Repository navigation
[Bug]: When CLI was installed along with 1.62, newly installed skills still reference playwright-cli instead of npx playwright cli #42135
Description
Version
1.62
Steps to reproduce
npm init playwright@latest(without installing browsers)npx playwright install chromiumnpx playwright cli install --skills agents
In GitHub Copilot in vscode enter a prompt similar to
Using the playwright-cli skills, open a new (headful) browser, navigate to https://www.playwright.dev
Expected behavior
I expected the browser to open and navigate to the Playwright website.
Actual behavior
It's trying to execute playwright-cli and not finding it, because it's only installed as playwright cli
Additional context
No response
Environment
System:
OS: Windows 11 10.0.22631
CPU: (16) x64 11th Gen Intel(R) Core(TM) i7-11850H @ 2.50GHz
Memory: 40.71 GB / 63.67 GB
Binaries:
Node: 21.0.0 - C:\Program Files\nodejs\node.EXE
npm: 10.2.0 - C:\Program Files\nodejs\npm.CMD
IDEs:
VSCode: 1.128.0 - C:\Users\Eric.Veltman\AppData\Local\Programs\Microsoft VS Code\bin\code.CMD
Languages:
Bash: 5.1.16 - C:\WINDOWS\system32\bash.EXE
npmPackages:
@playwright/test: ^1.62.1 => 1.62.1Reacted by Adi Gal
Activity
Skn0tt commentedon Aug 5, 2026
I can see that your model read lines 1-260 of the Skill file, and the note about the installation is in Line 343:
playwright/packages/playwright-core/src/tools/skills/playwright-cli/SKILL.md
Lines 343 to 357 in 19ecad6
How unfortunate. What model are you using, out of curiosity?
We can solve this in multiple ways:
refactoreric commentedon Aug 5, 2026
Ah I hadn't noticed that the skills file already contains that 'npx playwright cli' vs 'playwright-cli' note.
I'm not 100% sure which model ran into the issue originally. I have my model set to Auto, and by now I changed the skills file as a workaround, replacing all 'playwright-cli' by 'npx playwright cli'.
But it looks like most of my today's playing with Playwright CLI was routed to GPT-5.3-Codex.
Skn0tt commentedon Aug 5, 2026
Out of curiosity, could you try out one of the newer models with larger context and see how many lines that pulls in? Opus 5 / GPT 5.6 Sol / Grok 4.5 etc. GPT-5.3-Codex is a couple months old at this point and we'd rather optimise towards newer models. I'm also curious whether other harnesses, like the Copilot CLI, behave differently here.
refactoreric commentedon Aug 5, 2026
Ouch, so many variables when testing these things 😅 . With GPT-5.6 Sol it worked better. It used the right command from the start, even though I had first deleted the skill and reinstalled the original unmodified version.
No time to test with Copilot CLI.
Skn0tt commentedon Aug 10, 2026
Sounds good! If it works with Sol we're fine.