Click accuracy: return bounding box and viewport context on failure

[rafiki270]

Problem

When click or tap fails (element not found, element obscured, coordinates miss), the error response gives minimal context. LLMs can't diagnose whether:

• The selector was wrong
• The element moved (dynamic page)
• The element was off-screen
• The coordinates were calculated incorrectly
• Another element was overlapping

Current behavior

{ "success": false, "error": { "code": "ELEMENT_NOT_FOUND", "message": "..." } }

Proposed behavior

On failure, include diagnostic context:

{
  "success": false,
  "error": {
    "code": "ELEMENT_NOT_FOUND",
    "message": "No element matches selector '#submit'",
    "diagnostics": {
      "selector": "#submit",
      "viewport": { "width": 390, "height": 844 },
      "scrollPosition": { "x": 0, "y": 1200 },
      "similarElements": [
        { "selector": "#submit-btn", "text": "Submit", "rect": { "x": 120, "y": 400, "w": 150, "h": 44 } }
      ]
    }
  }
}

On click with coordinates, if the click lands on an unexpected element:

{
  "diagnostics": {
    "targetSelector": "#login-btn",
    "targetRect": { "x": 100, "y": 300, "w": 200, "h": 48 },
    "clickedPoint": { "x": 200, "y": 324 },
    "actualElementAtPoint": "div.overlay-modal"
  }
}

Impact

This is the #1 failure mode LLMs hit when automating browsers. Better diagnostics would let them self-correct immediately instead of retrying blindly.