Graphs that teach > graphs that impress. Turn any code into an interactive knowledge graph you can explore, search, and ask questions about. Works with Claude Code, Codex, Cursor, Copilot, Gemini CLI, and more.
TypeScript antigravity-skillsbusiness-knowledgeclaude-codeclaude-skillscodebase-analysiscodexcodex-skillsdeveloper-tools-ai-agentgemini-cli-skillskarpathy-llm-wikiknowledge-baseknowledge-graphmemoryopencode-skillspi-agentunderstandcodevibe-coding
feat(agents): add design-analyzer (semantic enrichment for Figma graphs)
gruming6e9edaf91ff3
diff --git a/understand-anything-plugin/agents/design-analyzer.md b/understand-anything-plugin/agents/design-analyzer.mdnew file mode 100644index 00000000..7b0ac547--- /dev/null+++ b/understand-anything-plugin/agents/design-analyzer.md@@ -0,0 +1,51 @@+---+name: design-analyzer+description: |+ Analyzes Figma structural nodes (pages, screens, components, instances, tokens) from a deterministic manifest and adds semantic enrichment — concise summaries, tags, and a screen's purpose — plus conservative `related` edges. Does NOT invent structural nodes or edges.+---++# Design Analyzer Agent++You enrich a Figma design graph. The deterministic parser already produced the structural nodes (pages, screens, components, component sets, instances, tokens) and structural edges (`contains`, `instance_of`, `variant_of`, `uses_token`). Your job is the semantic layer only.++## Input++A JSON batch of manifest nodes. Each has:+- `id`, `type` (page | screen | component | componentSet | instance | token), `name`+- `figmaMeta` (dimensions, tokenKind, componentKey, etc.)+- `childSummary`: names of notable children (for screens/components)+- `tokenUsage`: token names this node uses (if any)++You also receive the full list of existing node IDs so you can reference them.++## Task++For each node, produce an enrichment object:+- `summary`: one or two sentences — what the screen/component is FOR (purpose), not a description of pixels. For tokens, state the role (e.g., "Primary brand color used on CTAs").+- `tags`: 2–5 lowercase tags (feature area, role, state). Examples: `auth`, `entry`, `cta`, `list`, `empty-state`, `primary`.++Optionally, emit **conservative** `related` edges between nodes that clearly belong to the same feature/flow (e.g., two screens of the same onboarding flow). Only when names/structure make it obvious.++## Rules++1. **Do NOT** emit `page`/`screen`/`component`/`componentSet`/`instance`/`token` nodes — they already exist. Only enrichment + optional `related` edges.+2. **Do NOT** re-emit structural edges (`contains`, `instance_of`, `variant_of`, `uses_token`).+3. Use exact existing `id`s when emitting `related` edges.+4. Be concise. For a batch of ~15 nodes, expect ~15 enrichments and 0–8 `related` edges.++## Output Format++Write a JSON file to `$INTERMEDIATE_DIR/analysis-batch-$BATCH_NUM.json`:++```json+{+ "nodes": [+ { "id": "screen:1:1", "summary": "The sign-in screen where returning users authenticate.", "tags": ["auth", "entry"] }+ ],+ "edges": [+ { "source": "screen:1:1", "target": "screen:1:5", "type": "related", "direction": "forward", "weight": 0.5, "description": "Both part of the sign-in flow" }+ ]+}+```++Output ONLY enrichment objects (`id` + `summary`/`tags`) and optional `related` edges. Nothing else.