Stop Being Your Coding Agent's QA Team

Today I've got a free tool that gives your coding agent a real browser, a 15-minute setup so it tests the page it just built, a free skills pack that makes your agent plan and test before it codes,...

Happy Friday,

Let's get into it.

In This Email

» Save This: Playwright CLI, a free real browser for your coding agent

» Build This: have your agent click through and check its own work in 15 minutes

» Try This: Agent Skills, 25 free skills that make your agent spec and test first

» AI Update: LFM3, a 1 GB model that runs on your laptop

» Here's My Take: stop being your agent's QA team

🔖 Save This: Let Your Agent Check Its Own Work

Playwright CLI

In this lesson you'll give your coding agent a real browser so it can click through and check the page it just built.

→ Open the free Playwright CLI lesson here

Here's the problem it solves. Your agent says "done," you open the page, and the signup button does nothing. So you click around, take screenshots, paste them back and explain what broke. You've become the tester.

Playwright CLI is Microsoft's free command-line tool that lets a coding agent open a real browser, click, type, take screenshots and read what's on the page. It works with Claude Code, GitHub Copilot or any other coding agent. Microsoft's README says agents increasingly prefer a CLI like this over the Playwright MCP server because it's more token-efficient: it doesn't load big tool schemas or whole page trees into the chat.

What it replaces: you clicking through your own app after every change and describing the bugs back to the agent.

What it costs: nothing. It's open source under Apache 2.0 and needs Node.js 18 or newer. Your agent's own usage still counts against whatever plan you're on.

The trap: the browser forgets you by default. The README says the profile lives in memory, so cookies and logins are "lost when the browser closes." Add --persistent if your test needs to stay logged in. Also, a headless session shuts itself down after an hour without commands, so if it vanishes, just open it again.

One number: 14,843 GitHub stars as of Nov 1, up from 14,607 a week earlier.

🛠️ Build This: Your Agent Tests What It Built

Here's the exact setup. You need Node.js 18+, a project you can run locally, and your coding agent. Copy the commands as they are.

Install the CLI.
npm install -g @playwright/cli@latest
Install the skills so Claude Code, Copilot and others know how to use it:
playwright-cli install --skills
Start your app the way you normally do, so it's running on a local address like localhost:3000.

Ask your agent to test one flow. Pick the flow you care about most and paste this:

Use playwright-cli to open localhost:3000 and test the signup flow end to end. Take a screenshot of every step that passes or fails. Check playwright-cli --help for the commands. Then list what broke and fix it.

Watch it if you want. It runs headless by default. To see the browser, open it yourself with:
playwright-cli open localhost:3000 --headed
Make it a habit. Add one line to your project instructions file (for example CLAUDE.md) so every change ends with a check:
After any UI change, use playwright-cli to open the page, click through the change, and take a screenshot before saying done.

That's it. Two installs, one prompt, and your agent stops handing you broken buttons. The lesson covers four more free plugins too.

→ Follow the step-by-step lesson here

🧠 Try This: Make Your Agent Plan Before It Codes

Agent Skills by Addy Osmani

Agent Skills is a free, MIT-licensed pack of 25 engineering skills that make your agent write a spec, plan, test and review before it ships. Commands like /spec, /plan, /test and /ship map to each step. It has 112,407 GitHub stars as of Nov 1.

How to try it Monday: the README says one command installs it into 70+ agents, including Claude Code, Cursor, Codex, Copilot and Cline.

npx skills add addyosmani/agent-skills

The trap: in Claude Code, the built-in /review wins the name clash, so use /agent-skills:review instead. And if the plugin install fails with an SSH error, the README says to use the full HTTPS URL.

→ Get the Agent Skills lesson here

🔊 AI Update: A 1 GB Model for Your Laptop

LFM3 by Liquid AI

Liquid AI released LFM3 on Oct 27: six open-weight 2B-class models covering text, vision, audio, omni and decision tasks. Liquid says the text model "fits in about 1 GB at 4-bit precision," has a 256K-token context window, and runs with llama.cpp, MLX and vLLM.

Why it matters: you can run a small agent model locally, with no per-token bill, for jobs like sorting tickets or reading documents.

The trap: it's open-weight, not open for everyone. The LFM Open License v1.0 allows commercial use only for companies under $10M in annual revenue. LFM3 isn't in the Academy yet.

→ Try LFM3-1.7B on Hugging Face here

💬 Here's My Take

Agents got fast at writing code. They're still bad at noticing when it doesn't work, and that gap lands on you.

Playwright CLI closes most of it for free. Two installs, one prompt, and your agent checks its own work before it tells you it's done. Add Agent Skills if you want it to plan first, too.

My advice: set it up on one project this weekend and let it test one flow. Once you see it catch its own broken button, you won't go back.

📚 Go Deeper

The full walkthrough is in the AI Resource Vault: 5 Claude Code Plugins Every Vibe Coder Needs, with Playwright CLI and four more free plugins. For the planning side, see Make Claude Code Work Like a Senior Engineer. Every Friday pick comes from the free AI Academy, where 5,000+ members share guides, tools and what's working for them.

Join the Free AI Academy →

Free, no card. Guides, tools and the people running them live here.

That's it for this week. Which flow are you having your agent test first? Reply and tell me, I read them all.

See you Tuesday for the news.

From Drew and The AI Debrief team

P.S. Send us the first bug your agent caught on its own. Hit reply, we read every one.

P.P.S. New here, or missed one? Every past edition is in the archive.