Browser Annotations

---

The problem

Your agent writes UI it can't see, and writes content it can't fact-check. So you squint at the browser and type a paragraph back — *"the CTA's too big, move it left, wrong green"*, *"where does that 34% come from?"* Slow, vague, and the agent still has to guess which element or sentence you meant.

**Point at the real page instead.** Two gestures, no mode to switch:

Gesture	Note kind	For	Anchored by
Click an element	🟢 Design	"make this green", "move it left"	source file:line (dev build), else data-testid / id / text
Select a run of text	🔵 Content	"verify this number", "expand this"	the quote itself + surrounding context + heading path

Install (30 seconds)

Not on the Chrome Web Store yet, so load it by hand:

1. Download this repo (green Code button → Download ZIP, then unzip) — or git clone it.
2. Open chrome://extensions and turn on Developer mode (top-right toggle).
3. Click Load unpacked and pick the extension/ folder — that subfolder, not the repo root.
4. P