plumb_scan_references

Scan several live reference URLs and extract a per-role style digest — concrete nav/hero/footer/card exemplars (box size, layout, colour, type size/family, alignment) from whichever references actually have that section. Streams live progress to Plumb Studio as each reference is scanned.

Not a build tool. This scans and reports; it doesn't compose or apply anything to Figma. Feed the result into a plumb_design DSL or a plumb_studio brief yourself.

Not a palette tool either. plumb_brand already extracts a synthesized colour palette from references — this is the structural counterpart: what shape do their sections actually take, not what colours do they use.

Inputs

Field	Type	Notes
references	string[] · required	Reference URLs to scan live, 1–8.

Returns

{
  "references": ["https://linear.app", "https://stripe.com", "https://mercury.com"],
  "patterns": {
    "nav": [
      { "url": "https://linear.app", "box": { "w": 1280, "h": 64 }, "layout": { "flow": "row", "pad": [0,24,0,24] }, "textSizes": [14], "fontFamily": "Inter" }
    ],
    "hero": [
      { "url": "https://linear.app", "box": { "w": 1280, "h": 560 }, "fillColor": "#08090a", "textSizes": [56, 20], "textAlign": "center" }
    ],
    "footer": [ /* … */ ],
    "card": [ /* … */ ]
  },
  "misses": [{ "url": "https://example.com/unreachable", "error": "…" }],
  "next": "Read `patterns.<role>` for concrete exemplars…"
}

Each exemplar carries whatever the source page actually had — every field is optional and simply absent when that reference's section didn't set it. textSizes is every distinct text size found within the section (up to 3 levels down), largest first — a hero's [56, 20] reads as "a 56px headline over 20px body copy," not just one number. A role with zero exemplars means none of the references had a confidently-classified section of that kind (the classifier abstains rather than guessing — same discipline as everywhere else in Plumb).

When the agent should use it

• Before plumb_studio or plumb_design, when you want the generated sections to actually resemble the references structurally — typical hero height and layout, card-grid density, nav style — not just share their colour palette.
• Comparing how several competitors structure the same kind of section (e.g. pricing-page card grids) before designing your own.

Notes

Reuses the exact same structural pipeline plumb_import_web and its role classifier (nav/hero/footer/sidebar/card) run on — one page per reference, one desktop-viewport capture each (pass references one at a time through plumb_import_web with viewports if you need responsive exemplars too). Each reference is scanned independently; one failing URL is recorded in misses and does not abort the others.