Code to Figma

Generate a project-specific Figma export pipeline: a walker that reads compiled HTML and CSS, resolves class → token bindings, and pushes a structured JSON artifact to a GitHub Gist that the tokens-sync-to-figma Figma plugin consumes.

The pipeline is intentionally one-directional and CI-anchored. During setup, assess the project once, generate scripts, and wire CI. After that, every push that touches configured token or template paths automatically updates the Gist without a Figma API key. The bundled plugin rebuilds page structure using existing Figma variables; token-value changes require a separate collection update and readback before Figma parity can be claimed.

Commands

For an unqualified setup/sync request, default to /code-to-figma setup when no figma-sync.config.json or walker scripts are found. Preserve explicit read-only status requests: report a missing pipeline rather than creating one.

/code-to-figma setup

1 — Assess the project

Read in this order before generating anything:

1. Package manifest — package.json, pyproject.toml, etc. Identify the framework (Next.js, Eleventy, SolidJS, plain HTML) and package manager.
2. CSS / token files — find the compiled output, not the source. Common paths:
   • Tailwind v4: out/assets/css/tailwind.css or equivalent build output
   • Custom CSS: look for a file with --token-name: value; custom property declarations
   • Style Dictionary: tokens.json, variables.css, or generated output in dist/
3. Component CSS — look for a components.css or utilities.css alongside the main CSS file
4. Built HTML — the compiled output page, not source templates. Common paths: out/index.html, dist/index.html, _site/index.html. Next.js: a default build emits no single HTML file; require output: 'export' (yields out/index.html) before proceeding — see the Next.js note in references/walker-patterns.md. If no static HTML artifact can be produced, stop and tell the user rather than guessing a path.
5. Token naming convention — read the CSS custom property names, extract the prefix-to-group mapping (e.g. beige-* → palette/beige/, fs-* → typography/scale/)
6. Section structure — scan the HTML for how sections are delimited: <section id="...">, [data-section], <article>, header/footer landmarks, etc.

Ask one focused question if two genuinely different walker shapes are possible (e.g. sections identified by ID vs by class). Otherwise infer and state the choice.

2 — Generate and verify the scaffold

Create or update the project-specific walker, DTCG 2025.10 token converter, generic Gist pusher, figma-sync.config.json, package scripts, GitHub Actions workflow, Gist, secrets, first sync, and Figma plugin connection. Load references/setup-scaffold.md for the file specifications and references/ci-and-gist-setup.md for the canonical Gist, secret, CI, and plugin setup commands.

Required local checks before relying on CI:

node scripts/tokens-to-figma/convert-to-dtcg.mjs
git status --short -- scripts/tokens-to-figma/*.w3c.json
node scripts/figma-export/walk-<site>.mjs | jq -e '.sections | type == "array"'

Keep the core boundary visible: CI produces figma-export.json and checks the committed <project>-tokens.w3c.json; the tokens-sync-to-figma plugin consumes only figma-export.json. Use the DTCG artifact separately to maintain Figma variables. Do not require a Figma API key in CI.

Setup is complete when the generated scripts run, the Gist contains the validated export, CI is wired, and the user has verified the plugin connection and variable values through collection reconciliation. If Figma access or manual steps are pending, report pipeline setup separately from Figma verification.

/code-to-figma sync

1. Confirm figma-sync.config.json exists and the walker path is valid.
2. Build the current site with the project's build command, regenerate node scripts/tokens-to-figma/convert-to-dtcg.mjs, and review the .w3c.json diff, including changes to existing values and aliases. Then run node <walker> > figma-export.tmp.json.
3. Validate: jq -e '.sections | type == "array"' figma-export.tmp.json must exit 0; stop on failure.
4. Confirm GIST_TOKEN is already exported from a secret manager per references/ci-and-gist-setup.md; never type the PAT on the command line.
5. Push: node scripts/tokens-to-figma/push-to-figma.mjs < figma-export.tmp.json
6. Reconcile token changes with the Figma collection using the reconciliation procedure. Gist authorization does not authorize writes inside Figma. Report export uploaded and Figma token parity verified/unverified as separate statuses; without authorized Figma access or a manual readback, parity is unverified and the pending deltas are listed.
7. Delete figma-export.tmp.json when done, then report sections and nodes exported, the Gist URL, and the separate Figma verification status. Claim Figma parity only after collection values and the plugin result are verified.

/code-to-figma update

Use this when the framework output, token naming convention, section selectors, or CSS build paths changed enough that the existing walker may be stale.

1. Re-run the setup assessment against current compiled HTML/CSS and token files.
2. Update walk-<site>.mjs, convert-to-dtcg.mjs, figma-sync.config.json, and CI paths together so tokenPath(), explicit token types, and file paths stay aligned.
3. Regenerate the DTCG token artifact: node scripts/tokens-to-figma/convert-to-dtcg.mjs.
4. Validate the walker output: node <walker> > figma-export.tmp.json && jq -e '.sections | type == "array"' figma-export.tmp.json.
5. Push through /code-to-figma sync or CI after reviewing the script and .w3c.json diffs, then reconcile collection changes and report export and Figma parity separately.

/code-to-figma status

Report:

Skill Boundaries

This skill does not edit Figma files. The plugin (tokens-sync-to-figma) is the Figma-side consumer — this skill produces the artifact it reads. The plugin binds to an existing Design Tokens collection; creating that collection once is a prerequisite, and its optional agent path hands off to figma-use (see references/ci-and-gist-setup.md).

Behavioral evals

Authors/reviewers only: use the scenario catalog when explicitly evaluating this skill; skip it during normal task execution. It holds description-only routing prompts and isolated sync/readback fixtures; withhold labels and assertions from the executing agent. The catalog is not evidence of a passing run, and synthetic readbacks do not verify live Figma integration. Converter behavior is fixture-tested by tests/walker-patterns-snippets.test.mjs.

Reference Files

Operating Principles

• Read the compiled output, not the source. Token bindings only become resolvable in built HTML+CSS. Source templates may use variables that haven't been substituted yet.
• tokenPath() and explicit $type mappings are the contract. The walker and DTCG converter must use the same path function, and every exported token must have an intentional DTCG type. Never infer $type from the raw CSS value; fail on unknown taxonomy or non-conforming values.
• Walker is project-specific; pusher is generic. The walker understands the project's HTML shape; the pusher only knows the Gist API. Keep them separate.
• Commit the DTCG JSON. The .w3c.json file is the human-readable diff surface for token changes. It belongs in the repo, not in .gitignore.
• node not pnpm in CI. pnpm writes a script header to stdout when running a lifecycle script, which corrupts a > file.json redirect. Always invoke the walker with node directly in CI steps.