Smoke Testing with Kane CLI: My AI Agent Broke Production (On Purpose)

Smoke testing with Kane CLI, on a live production deploy. I told my AI coding agent to break my store on purpose, and one natural-language Kane CLI smoke test caught it. 👉 Try Kane CLI: https://www.testmuai.com/kane-cli/?utm_source=youtube&utm_medium=organic&utm_term=frln-2369yg&utm_campaign=kane_cli_smoke_testing Coding agents now write, deploy and even run in parallel, and AI-driven changes keep turning up in production-outage stories. Speed isn't the problem. Unchecked deploys are. Here, Kane CLI ran the check in a real browser, returned exit code 1, and my agent rolled the deploy back on Vercel, with one human approval from me. Kane CLI brings TestMu AI's testing agent to your terminal. It works with Claude Code, Codex CLI and Gemini CLI, and covers mobile apps on emulators, simulators and real devices. New in Kane CLI 0.10: explore maps your whole app with up to 20 cloud browsers and flags console, network, broken-link and accessibility issues, with evidence for every finding. What you'll see: • Kane CLI in action: one sentence in, one exit code out • The exit-code contract: 0 passed, 1 app failed, 2 CLI or environment error, 3 timeout • Every Kane CLI run saved in TestMu AI Test Manager, with steps and screenshots • A CLAUDE.md smoke gate that makes Claude Code run Kane CLI after every deploy • A deliberate break, caught by Kane CLI and rolled back on Vercel • The fix, verified by Kane CLI again Chapters 00:00 Breaking production, on purpose 00:21 What a smoke test is 00:57 What is Kane CLI 01:44 The demo store 01:57 Finding Kane CLI in KaneAI 02:28 One sentence in, one exit code out 03:35 Every Kane CLI run lands in Test Manager 04:35 Giving Kane CLI to Claude Code (CLAUDE.md) 05:40 Control run: the live site passes 06:35 The deliberate break 08:05 Kane CLI catches it, the agent rolls back 09:05 Fix, promote, pass 09:50 Did it obey? Get started with Kane CLI npm install -g @testmuai/kane-cli (or: brew install lambdatest/kane/kane-cli) kane-cli login kane-cli run "YOUR CHECK IN NATURAL LANGUAGE" --url YOUR_SITE_URL --agent --headless --timeout 180 --max-steps 8 echo $? The CLAUDE.md smoke gate Save this as CLAUDE.md in the root of your repo, then swap in your own check and URL. Claude Code reads it at the start of every session. Smoke gate: run after every production deploy Before deploying, note the live deployment: vercel inspect YOUR_SITE_URL Deploy to production with vercel --prod. If production is rolled back (vercel rollback status), run vercel promote NEW_DEPLOYMENT_URL first. Then run the Kane CLI gate exactly as written, in the foreground, no pipes: kane-cli run "YOUR CHECK IN NATURAL LANGUAGE" --url YOUR_SITE_URL --agent --headless --timeout 180 --max-steps 8 Act on the exit code: • 0: smoke passed. Keep the deploy and report what was checked. • 1: the app failed the check. Roll back with vercel rollback LAST_GOOD_DEPLOYMENT_URL, confirm with vercel rollback status, then report. • 2: environment error (auth, Chrome, credits). Do not roll back. Wait 30 seconds, retry once, then report. • 3: timeout. Retry once, then report. Never leave a build live that has not passed the gate. Links Kane CLI quickstart: https://www.testmuai.com/support/docs/kane-cli-quickstart/ Smoke testing from the command line (full guide + gate script): https://www.testmuai.com/blog/smoke-testing-command-line/ #KaneCLI #ClaudeCode #SmokeTesting #TestMuAI