Smart Drawing (Claude skill)

Turn an architectural PDF drawing set into one clickable HTML file. Click a door, see its schedule.

A door tag sits on the plan. Its size, frame and hardware are in the schedule three sheets away. Its finish is a code explained in a legend, and the wall it sits in is defined on yet another sheet. Reading one item means flipping between sheets.

This skill reads the set once and links it all together:

• Doors: click one and see its full schedule row (size, frame, glass, hardware, fire rating).
• Rooms: click one and see floor, base and wall finishes, with the codes decoded into products.
• Wall tags: click one and see how the wall is built (studs, width, fire and sound rating).
• Detail bubbles: click one to open that sheet.
• Search: type a tag and every instance lights up on the plan.
• Whole-set counts and filters: doors, rooms, windows, wall types, finishes.

The output is one HTML file. No install, no login, no upload. Double-click it and it works offline.

Results

Set	Sheets	Found	Linked on plans
Public safety facility (firm 1, 122 arch sheets)	122	468 doors, 453 rooms, 12 window types, 10 wall types, 79 finish codes	466 / 468 doors, 451 / 453 rooms
IndyGo Training Facility (firm 2, public bid set)	15	26 doors, 24 rooms, 17 finish codes	26 / 26 doors, 24 / 24 rooms

The two firms lay out their schedules differently. Nothing was configured per set: the skill reads each schedule's own columns.

Try it: open examples/indygo/smart-drawing.html.

Install

pip install -r requirements.txt

Copy this folder into ~/.claude/skills/smart-drawing/ (or your project's .claude/skills/), then ask Claude: "make a smart drawing from inputs/my-set.pdf".

Or run it directly:

python scripts/smart_drawing.py my-set.pdf --out outputs/my-project --project "Office building"

How it works

1. Index: build_index.py reads the PDF's text and table lines (no AI vision, no OCR) into SQLite: sheets, schedule rows, legend codes, wall types, plan tags and callouts. An item is HIGH when its schedule row and its tag on a plan are both found.
2. Viewer: export_viewer.py renders the sheets and writes the index and images into one HTML file.

Every value you see when you click comes from the drawing's own text.

Limits

• Needs a CAD-exported PDF. Scanned sets won't work.
• Architectural only for now.
• A big set takes 10-25 minutes to index the first time.
• Windows listed only as elevation types (no window schedule) are not picked up on every set (IndyGo: 0).
• Two-column legends can put a code under the wrong category.

Source-available licence (see LICENSE). Part of the AutoConst tools by Hamza Jabbar.