| name | design-annotations |
| description | Use when adding, repositioning, or improving dev note components next to Figma design screens. Triggers include requests to add dev notes, place annotations, improve dev note copy, or update existing notes. |
Design Annotations
Add, reposition, or improve dev note components next to design screens in Figma. Works with design system components or a default note template. Chains from design-organize but also works standalone.
Validation principle: Every use_figma write step must be followed by a get_screenshot check. Do not stack writes without visual verification.
Quick Reference
| Step | What |
|---|
| 0 | Prerequisites — Figma MCP required |
| 1 | Ask preferences: which screens, format, content |
| 2 | Read container, classify children, find existing labels |
| 3 | Version safety |
| 4 | Determine note width → reposition screens/labels → place notes |
| 5 | Validate and present result |
Step 0: Prerequisites
-
Figma MCP server — check if mcp__figma__use_figma tool is available. If not, tell the user to install it and restart.
-
figma-use skill — load figma:figma-use via the Skill tool before any use_figma call.
-
Figma Section gotchas — these are critical and cause silent failures:
- Page discovery:
getNodeById() works cross-page — it will find nodes on ANY loaded page, giving false positives. Use figma.currentPage.findOne(n => n.id === "<nodeId>") instead, which only returns nodes that actually live on the current page.
- Page-level TEXT: Figma stores TEXT nodes placed inside SECTIONs as page-level children — they won't appear in
section.children. Scan figma.currentPage.children for TEXT nodes whose absoluteBoundingBox falls within the section bounds. These are existing labels. Their x/y are canvas-absolute — use section.absoluteBoundingBox to convert.
- Overlapping children: SECTION nodes may silently discard children that overlap existing frames between script executions. Position new nodes to avoid overlaps.
- Stop on repeated failure: If the same operation fails twice, STOP and read back children before retrying.
Step 1: Ask Preferences
Extract fileKey and nodeId from the user's Figma URL (convert - to : in nodeId).
Detect intent from the message — "add dev notes" vs "improve copy" vs ambiguous. Then always confirm preferences before proceeding (skip only what the user already specified):
Which screens get dev notes?
- A) All screens
- B) Let me pick (I'll list them)
Format:
- A) Use a design system component (I'll search for it)
- B) Use a default note template
Content:
- A) Pre-fill with AI-generated descriptions
- B) Leave empty for manual editing
For "improve copy" intent, skip format/screen questions — go straight to Step 2 → Path C.
Step 2: Read Container and Classify Children
Use a single use_figma call to find the correct page and classify all children:
-
Page discovery: Iterate all pages, switch to each, use figma.currentPage.findOne(n => n.id === "<nodeId>") to find the page that owns the node. Do NOT use getNodeById() — it returns nodes cross-page. Mention which page you found (e.g., "Working on page [page name]") so the user can catch wrong-page issues early.
-
Classify section children: screens (everything visible that isn't a label or note), labels (TEXT nodes), notes (INSTANCE ≤ 450px or FRAME named "Dev Note" ≤ 450px).
-
Scan page-level TEXT: Check figma.currentPage.children for TEXT nodes within the section's absoluteBoundingBox. Flag these as pageLevel: true (canvas-absolute coords).
-
Validation: If design-organize was used (expect labels) but 0 labels found — you're on the wrong page. Re-run discovery.
-
If user chose "let me pick": List screens with names and dimensions for selection.
Store one FRAME screen ID as <childId> for subsequent page-switch preambles.
Step 3: Version Safety
Ask the user to save a named version in Figma before proceeding. Wait for confirmation.
Step 4: Determine Width → Reposition → Place Notes
This is the core step. The order is critical — reposition BEFORE placing notes to prevent overlaps.
4a: Determine note width
- Component path: Search design system → import → list all variants, ask user which one (don't auto-pick default). Create ONE test instance, confirm dimensions. Record
noteWidth.
- Default template:
noteWidth = 400px.
- Improve copy (Path C): Skip to 4d.
4b: Reposition screens and labels
Widen gaps between screens to fit notes. For each screen left-to-right:
- Screens WITH notes: gap =
screen.width + 50 (noteGap) + noteWidth + 200 (nextScreenGap)
- Screens WITHOUT notes: gap =
screen.width + 200 (standard gap)
Move each screen's label to follow it. Page-level labels need canvas-absolute coordinates: sectionAbsBounds.x + screenRelativeX.
Verify with get_screenshot that gaps are visible.
4c: Place notes
For each selected screen, place a note at screen.x + screen.width + 50, screen.y.
Component notes: Import chosen variant, createInstance() per screen, append to container.
Default template notes: Create template frame (400px wide, auto-height, vertical layout, "Dev Note" name, Inter fonts, warm fill {1, 0.98, 0.94}), clone per screen, remove template. Must run in a single use_figma call (template reference doesn't persist).
Verify with get_screenshot.
4d: Improve existing copy (Path C)
- Read all existing notes via
use_figma — walk each note's children to extract text nodes with their IDs and content
- Match notes to screens by x-proximity (note.x near screen.x + screen.width)
- Screenshot each screen for visual context
- Generate improved copy — focus on what devs need: component behavior, state, data sources, interactions
- Present suggestions in a table for confirmation:
| Screen | Current | Suggested |
|---|
| Dashboard | "Add notes here..." | "Main dashboard. Stats row: real-time metrics..." |
- Bulk-update confirmed notes in one
use_figma call — load each text node's exact font before changing .characters
Step 5: Validate and Present
Screenshot the section. Verify notes are positioned correctly, not overlapping, readable.
Done. Added/updated X dev notes across Y screens.
Want to adjust anything?
Next step: Want me to optimize these screens for MCP? (I can invoke mcp-optimize)