Design To Code Check

Design-to-code check

A skill for reviewing the alignment between a design specification and its code implementation, producing a structured discrepancy report catalogued by dimension and severity.

Before you begin: verify references

Confirm that every path in this skill's frontmatter references: exists relative to this SKILL.md. If any is missing, stop: the install is incomplete, usually because a flattening installer (for example npx skills install) dropped the repo-root knowledge-notes/ directory. Tell the user to reinstall by a method in 1-INSTALL.md and run verify-install.sh from the install root. Proceed without the references only if the user explicitly says to, and then say in the output that it was produced without the pack's reference material.

Context

Design-to-code alignment reviews catch two different categories of problem. The first is implementation error — the developer built something different from what was specified, either by mistake or because the specification was unclear. The second is specification ambiguity — the design did not define behaviour completely enough for the developer to implement it correctly, and the developer made a reasonable guess that turned out to be wrong.

Both categories matter, but they require different responses. An implementation error needs to be corrected in the code. A specification ambiguity needs to be corrected in the design and documented, so the same guess does not get made again.

This skill produces a report that distinguishes between the two.

---

Configuration

If .ds-ops-config.yml exists, follow the configuration-and-recurring knowledge note (../../knowledge-notes/configuration-and-recurring.md) for loading, integration fallbacks and recurring runs. This skill reads:

• severity.* — discrepancy severity overrides, especially specification_gap and missing_interaction_state
• system.framework and system.styling — pre-select framework-specific checking guidance
• integrations.figma — the design specification (see below)
• integrations.chromatic — visual regression data as a supplementary signal
• integrations.github — the component source
• gates.design_to_code — if running as part of component-to-release, determines which findings block release

Auto-pull integrations

Figma MCP (integrations.figma.enabled: true, or a Figma link in the request):

• Official Figma MCP: get_design_context on the component's node for its properties, variants, layout and styles, and get_variable_defs for the variables that node binds, which is how the spec names its tokens. Both are selection- or node-scoped, so ask for the node link if the file key alone is given.
• Figma Console MCP: figma_get_component_for_development for the dev-ready spec, figma_capture_screenshot for the rendered reference.
• Record the node id with every value taken from Figma; it goes in the log as evidence.

Chromatic (integrations.chromatic.enabled: true):

• Pull the latest visual snapshots for the component being checked
• Use visual diffs between the Chromatic baseline and the current implementation as a supplementary signal for Dimensions 1–3 (spacing, colour, typography): it surfaces differences the manual review should confirm, never marks a dimension as matching

GitHub (integrations.github.enabled: true):

• If the component implementation is in the configured repo, pull the component source files directly
• Identify the component's last update date and recent changes to contextualise findings

Step 1: Gather the comparison materials

Ask for or confirm (skip questions already answered by auto-pull):

• The design reference: Figma file link, exported specs, or described specification (skip if pulled from Figma MCP)
• The implementation reference: component in code (React, Vue, Twig, etc.), a link to a running implementation, or a description of what was built. Note the styling approach — CSS custom properties, SCSS variables, Tailwind utility classes, or CSS-in-JS — as this affects how token references are identified during the check.
• The component or screen being reviewed
• Whether this is a first-pass review or a follow-up check after a previous round of corrections
• Any known areas of concern the review should pay particular attention to

If both design and implementation are available directly, proceed to the check. If only one is available, note in the report which side of the comparison is inferred rather than directly inspected — dimensions that depend on the inferred side are reported as "not checked", not ✅.

Step 1b: Design specification checklist

Before running the check, verify the design specification is complete enough to check against. Incomplete specs are the root cause of Type II (specification gap) findings — catching them upfront reduces noise in the report.

Design specification completeness checklist:

• All interactive states are defined (default, hover, active, focus, disabled, error, loading)
• Spacing values are specified using token names, not pixel values
• Colour values are specified using token names, not hex values
• Typography is specified using type scale tokens
• Responsive behaviour is defined for at least two breakpoints
• Focus indicator style is specified
• Content overflow behaviour is defined (truncation, wrapping, scrolling)
• Touch target sizes are specified for mobile breakpoints

If the specification fails this checklist: Note the missing items and proceed with the check. Missing specification items will appear as Type II findings in the report — but flagging them upfront sets the right expectation: these are design gaps, not implementation errors.

Step 2: Run the check across all dimensions

Review alignment across five dimensions. For each dimension, the goal is not to produce a list of every difference — minor sub-pixel differences in a rounding pass are not discrepancies worth reporting. The goal is to identify differences that affect visual consistency, user experience, or system integrity.

Dimension 1: Spacing and layout

Check:

• Padding and margin values: do they match design specifications, and are they using the correct spacing tokens?
• Element alignment: horizontal and vertical alignment of components within their containers
• Gap between elements in flex or grid layouts
• Component sizing: width and height where specified, or proportional behaviour where not fixed
• Responsive behaviour: does the implementation respond to breakpoints as specified?

Where the spec names a token (or the system has one for the value), a raw value in the implementation is a discrepancy even when the number matches: it won't theme or track the scale. Log it against the property the spec covers. Sweeping the whole component for raw values regardless of spec is token-compliance's job, with its per-styling-approach rules for what counts as a token reference (SCSS variables, var(), Tailwind utilities versus arbitrary values, Emotion helpers); apply the same rules here and don't widen the check beyond the spec'd properties.

Dimension 2: Colour and visual treatment

Check:

• Background colours, border colours, text colours: are they referencing the correct design tokens?
• Shadow and elevation: correct values, correct token references
• Border radius: correct values, consistent with the design system's radius scale
• Opacity: correct values and applied to the correct element
• Gradient or background treatments if present

Flag raw colour values on the properties the spec covers, as in Dimension 1.

Dimension 3: Typography

Check:

• Font family: correct typeface applied
• Font size: correct size, using the correct type token
• Font weight: correct weight at each text role
• Line height: correct leading, using the correct token or documented value
• Letter spacing: correct tracking where specified
• Text alignment: left, centre, right, or justified as designed
• Text truncation or overflow handling: does the implementation handle long strings as designed?

Dimension 4: Interactive states

Check:

• Default state: visual treatment matches design
• Hover state: correct treatment applied on hover
• Active/pressed state: correct treatment on click or touch
• Focus state: visible, compliant focus indicator applied (this is also an accessibility check)
• Disabled state: correct reduced-prominence treatment
• Loading state: if designed, correctly implemented
• Error state: if applicable, correctly applied and correctly associated with the relevant element
• Empty state: if applicable, correctly implemented

Interactive states are the most commonly under-implemented dimension. Flag any state that was designed but is not present in the implementation.

How states are checked from source. A state exists in the implementation when the source has a rule for it: :hover, :focus-visible (or :focus), :active, :disabled or [disabled]/[aria-disabled="true"], [aria-busy="true"] or a loading prop branch, [aria-invalid="true"] or an error prop branch, an empty-state branch in the template. Read those selectors and branches and compare their values with the spec. A designed state with no rule or branch is "not implemented". What those rules render (the actual hover colour on screen, the focus ring's visibility over a background) can only be confirmed in a running build or Storybook; if none was used, report the values as compared from source and the rendering as "not checked", not ✅.