ztx4-design-to-code

Deep implementation master for Design-to-Code Pipeline (offline) with 16 targeted Codex skills, code, tests and no MCP. Additive to all existing Perfect_AI versions.

SKILL.md

GitHub

Design-to-Code Pipeline (offline)

Convert permitted local Figma exports/reference assets to structured, accessible, maintainable UI without MCP.

Mandatory sequence

  1. Require a local export, screenshot, token file or SVG with known rights and source.
  2. Normalize layers, bounds, parent-child containment, auto-layout, token aliases and typography.
  3. Derive semantic components and flex/grid constraints rather than absolute positioned pixel replica.
  4. Generate framework-specific code with logical properties and available assets only.
  5. Compare screenshots by viewport; validate interactions not represented in static files separately.

Skill routing

Choose only technique files needed for this project from DEEP-IMPLEMENTATION-v2/references/design-to-code.md. Never indiscriminately load all 300+ skills into context.

Runbook

Inputs: local Figma JSON or exported images/SVG, tokens and rights; no MCP.

Implementation: normalize local layers, reconstruct auto layout as Flex/Grid, map semantics and tokens, generate TSX/CSS with responsive/RTL constraints.

Tests: parse invalid exports safely, no executable unsafe SVG, no absolute-layout dependence, screenshot comparison and keyboard affordances.

Acceptance gates

  1. No MCP required; only local files are read.
  2. Exported code uses semantic layout and responds beyond design artboard size.
  3. Sanitize imported SVG and verify licensing.

Runnable assets

extensions/DEEP-IMPLEMENTATION-v2/examples and tools (installed main package). See RUNBOOK-FA.md.

Sources