Node Map Anchoring

# Node-map anchoring (shared fragment)

Not a standalone prompt — a shared instruction block injected into the image prompts when a Figma layer map is available. It anchors each localized issue to concrete Figma layer IDs while keeping all user-facing text in plain language.

**Inputs (placeholders, substituted at call time):**
_None — this prompt takes no placeholders._

**Source:** [`supabase/functions/_shared/analyze-prompts.ts`](https://github.com/Soba-noodl/qualia/blob/main/supabase/functions/_shared/analyze-prompts.ts)

---

~~~text
STRUCTURAL ANCHOR — USE FIGMA LAYER IDs FOR LOCALIZED ISSUES:
A NODE MAPS appendix is provided below with one entry per image (keyed by 0-based image index). Each entry lists the visible Figma layers in that frame with stable ids, names, types, and frame-local design-unit bounds.

For every LOCALIZED engine/accessibility/prototype/cross_frame finding:
1. Identify the layer(s) in the relevant image's node map that the issue applies to. Match by NAME and TYPE first — node names like "Sign in CTA" or "Header / Logo" are the strongest signal. Use bounds only as a tiebreaker.
2. Output ALL applicable layer ids in a "layer_ids" array (strings, copy verbatim from the node map). If a single layer covers the issue, return a one-element array. If multiple related layers (e.g. a button group), include all of them — the frontend draws the union rectangle.
3. Output "layer_ids": null when the issue is genuinely global (whole-screen palette, hierarchy, etc.) or when no node in the map clearly matches. Do not invent ids that are not in the appendix.
4. Always also output "box_2d" in the normal [ymin, xmin, ymax, xmax] 0-1000 format as a fallback in case the layer ids can't be resolved on the webapp.

Format: "layer_ids" is a JSON array of strings (e.g. ["1:42", "1:43"]) or null. NEVER mention layer ids, node names, or coordinates inside user-facing text ("issue", "suggestion", "why_it_matters", "one_big_thing"). Those fields stay in plain language.
~~~