browsinglisted

# Browser Automation in CCY CCY ships **one CLI under three names, and you must pick one every time**. The bare `agent-browser` command is deliberately blocked: it prints this decision matrix and exits non-zero. Whatever you pick, the subcommands and flags are identical. | Will a human watch it happen? | Needs pixels or geometry? | Command | | ----------------------------- | --------------------------------------------------------- | ----------------------------- | | **yes** | acceptance tests, UI checks: a real window on the desktop | `agent-browser-headed` | | no | **yes**: screenshots, PDFs, geometry, layout | `agent-browser-headless` | | no | no: parsing text or content, scraping, research | `agent-browser-lite-headless` | There is no `agent-browser-lite-headed` and there cannot be: Lightpanda has no renderer, so it has nothing to show. **Announce:** "I'm using the browsing skill with `<command>` because `<one-line reason>`, and I will close it when I'm done." The reason matters. A headed window is the right thing when the user asked to watch an acceptance test and the wrong thing when you are quietly reading documentation, and saying which case you think you are in lets the user redirect you before the window appears rather than after. ## Decide in this order 1. **Did the user ask to see it, or is the w