132 / 168 · 01 Agent Skills for Browser Automation · How to Present Your Framework: 5, 15, and 30 Minute Versions← prev⊞ allnext →☰ Read as one page
18.4Common Follow-Up Questions (Be Ready)
| Question | Key Point |
|---|---|
| "What about visual testing?" | Screenshots on disk + vision-capable models for comparison; commercial visual AI where budget allows |
| "How do you version tests?" | Markdown plans in git, just like code; generated specs map 1:1 to plans |
| "What if the agent makes a mistake?" | Command log + trace for reproducibility, CI gates on pass/fail, healer patches are code-reviewed |
| "Cost at scale?" | ~4x cheaper than MCP per task; token spend tracked per suite as a first-class metric |
| "Why not just use Cypress?" | Cypress v15 added cy.prompt() and self-healing — credit it, then note our stack is agent-first end-to-end, not NL bolted onto a runner |
| "How do you handle pop-ups/alerts?" | The agent sees them in the snapshot and reasons about them; deterministic handling can be scripted via eval |
| "What about mobile testing?" | Viewport emulation for responsive checks; real mobile via Appium — different stack, same agent patterns |
| "Who maintains the tests?" | QA writes and reviews plans; agents handle execution and first-pass healing. Maintenance is updating intents, not selectors. |