Generating Code Connect

Scope

Create Code Connect files, write figma.config.json, validate mappings, and optionally publish

Command line first

• Prefer rg, jq, and ast-grep over manual inspection
• Use CLI tools or tiny scripts to validate and inspect outputs before editing by hand
• When extracting or transforming JSON, use jq instead of ad-hoc parsing

Inputs to collect

• Repo path, default current working directory
• Figma file URL
• Framework, auto-detect or ask React vs Angular
• Component scope, comma-separated list or Enter for all
• Output directory, default codeConnect
• Permission to install @figma/code-connect locally if missing

Default behaviors

• Generate all components when the user hits Enter on scope
• Prefer local @figma/code-connect over global installs
• Do not overwrite existing Code Connect files unless the user agrees

Workflow

0. Pre-flight checks

• Parse the file key from the Figma URL
• Validate the token against the target file without printing it:
  • curl -s -o /dev/null -w "%{http_code}\n" -H "X-Figma-Token: $FIGMA_ACCESS_TOKEN" https://api.figma.com/v1/files/<fileKey>
• Check for @figma/code-connect with scripts/check_figma_cli.sh
• If the token check returns 403 but the user says it is valid:
  • Locate the token source without printing it: rg -l -g '.env*' -g 'superconnect.toml' "FIGMA_ACCESS_TOKEN|figma_access_token"
  • Ask the user to set the token in the current shell and re-run the check
  • Confirm the token is non-empty without printing it: echo ${#FIGMA_ACCESS_TOKEN}
  • Do not echo or paste the token value in any command or response

1. Detect framework

• Read package.json and look for react or @angular/core
• If ambiguous, ask the user to choose

2. Extract Figma data with figma-scan.js

• Use the bundled script for reliable extraction:

  node scripts/figma-scan.js <figmaFileUrl> \
    --token $FIGMA_ACCESS_TOKEN \
    --output codeConnect/.figma-evidence \
    --index codeConnect/.figma-evidence/figma-components-index.json
  

• The script handles all Figma API complexity including:
  • Fetching componentPropertyDefinitions via /v1/files/:key/nodes endpoint
  • Parsing variant properties from component names
  • Extracting text layers for figma.textContent()
  • Extracting slot layers for figma.children()
  • Normalizing property names to camelCase
• Do NOT use raw curl + jq for extraction - the API has edge cases the script handles
• See references/figma-extraction-spec.md for output format details

Evidence output structure

Each evidence file contains:

{
  "schemaVersion": "figma-component@1",
  "componentSetId": "6:4335",
  "componentName": "Button",
  "variantProperties": { "Size": ["Compact", "Base", "Wide"] },
  "variantValueEnums": {
    "Size": { "normalizedKey": "size", "enums": ["compact", "base", "wide"] }
  },
  "componentProperties": [
    { "name": "Disabled", "type": "BOOLEAN" },
    { "name": "Label", "type": "STRING" }
  ],
  "textLayers": [{ "name": "Label", "type": "TEXT" }],
  "slotLayers": [{ "name": "Icon", "type": "FRAME" }]
}

Evidence quality gate

Before proceeding to generation, verify evidence is not empty:

for f in codeConnect/.figma-evidence/*.json; do
  props=$(jq '.variantProperties | length' "$f")
  comps=$(jq '.componentProperties | length' "$f")
  if [ "$props" -eq 0 ] && [ "$comps" -eq 0 ]; then
    echo "WARNING: Empty evidence for $f"
  fi
done

If evidence is empty, the script may need --layer-depth increased or the Figma component genuinely has no configurable properties.

3. Extract code component inputs

• For Angular, find all inputs in each component:

  rg "= input[<\(]" projects/zap/src/lib/components/button/button.component.ts
  

• For React, find props interface or destructured props:

  rg "interface.*Props|type.*Props|function.*\(\{" src/components/Button.tsx
  

• Update codeConnect/mapping.json with discovered inputs, see references/mapping-spec.md

Mapping Figma to code

Create figmaToCodeMap in mapping.json to link Figma property names to code input names:

{
  "figmaToCodeMap": {
    "Size": "size",
    "Disabled": "disabled",
    "Label": "text"
  }
}

Use these heuristics for matching:

1. Case-insensitive match first (Size → size)
2. Remove spaces and normalize (Has Icon → hasIcon)
3. Common synonyms (Label → text, Title → heading)
4. If ambiguous, prompt the user for confirmation

4. Read the relevant Figma docs and extract rules

• Read bundled references first:
  • references/figma-code-connect-overview.md
  • references/figma-code-connect-understand-files.md
  • references/figma-code-connect-react.md or references/figma-code-connect-html.md
  • references/figma-code-connect-config-file.md
  • references/figma-code-connect-common-issues.md
• If more detail is needed, consult the authoritative sources linked in each reference
• Enforce these rules from the docs:
  • figma.connect uses an object literal
  • example uses only props defined in props
  • Use figma.children for slots and figma.textContent for text layers
  • Keep JSX or templates valid and minimal

Angular-specific syntax rules

Use Angular property binding syntax in examples:

// ✅ CORRECT - Angular property binding
html`<zap-button [disabled]="props.disabled" [size]="props.size">`

// ❌ WRONG - HTML attribute syntax
html`<zap-button disabled=${props.disabled} size="${props.size}">`

For content projection with slots, use figma.children:

props: {
  icon: figma.children('Icon'),
},
example: ({ icon }) => html`<zap-button>${icon}</zap-button>`

5. Write Code Connect files directly

• Follow references/code-connect-format.md for exact file structure
• Read evidence from codeConnect/.figma-evidence/.json
• Read component info from codeConnect/mapping.json
• Write files directly to codeConnect/.figma.ts (Angular) or .figma.tsx (React)

For each component:

1. Read the evidence file to get Figma properties
2. Read the mapping entry to get selector, componentClass, figmaToCodeMap
3. Build the props object mapping Figma properties to code props
4. Build the example template using Angular property binding syntax
5. Write the complete Code Connect file

Prop mapping rules:

• figma.enum('PropName', {...}) for variantProperties
• figma.boolean('PropName') for BOOLEAN componentProperties
• figma.string('PropName') for STRING componentProperties
• figma.textContent('LayerName') for textLayers
• figma.children('LayerName') for slotLayers
• figma.instance('PropName') for INSTANCE_SWAP componentProperties

URL must include node-id

Every figma.connect() call MUST include the node-id query parameter:

// ✅ CORRECT - includes node-id from componentSetId
figma.connect('https://www.figma.com/design/FILE_KEY/Name?node-id=6-4335', {...})

// ❌ WRONG - missing node-id
figma.connect('https://www.figma.com/design/FILE_KEY/Name', {...})

Use the componentSetId from evidence files, converting colon to hyphen (6:4335 → 6-4335).

Import path discovery

Do not guess import paths. Derive them from the actual package:

# Find the package name
jq -r '.name' projects/zap/package.json

# Find exported components
rg "export \* from|export \{" projects/zap/src/public-api.ts

Write files to codeConnect/.figma.tsx for React or codeConnect/.figma.ts for Angular

6. Write figma.config.json at repo root

• Follow references/figma-code-connect-config-file.md for config options
• Use parser: react for .figma.tsx, html for .figma.ts
• Use include globs for Code Connect files and exclude node_modules, dist, .next

7. Validate using the two-tier validator

• Run node scripts/validate_code_connect.js
  • Single file: node scripts/validate_code_connect.js --code codeConnect/button.figma.tsx --evidence codeConnect/.figma-evidence/button.json
  • Batch: node scripts/validate_code_connect.js --code-dir codeConnect --evidence-dir codeConnect/.figma-evidence
• If evidence is missing, stop and ask for it before proceeding

Do not generate empty Code Connect files

If props cannot be determined, either prompt for manual input or skip the component:

// ❌ NEVER generate this - provides no value
figma.connect('...', {
  props: {},
  example: () => html`<zap-button></zap-button>`
})

8. Offer to publish

• Ask before running figma connect publish
• If user wants a dry run, use figma connect publish --dry-run

Prerequisites

For figma-scan.js, ensure dependencies are available:

npm install commander chalk json-stringify-pretty-compact

Notes

• Write Code Connect files directly - do not use a generator script
• Avoid global installs unless the user explicitly prefers them
• Keep references one level deep from SKILL.md
• Use forward slashes in all paths
• Never print the raw FIGMA_ACCESS_TOKEN value
• The figma-scan.js script filters hidden components (prefixed with . or _)
• Use componentSetId from evidence for node-id URLs, not individual variant IDs