| name | screen-implementer |
| description | Reimplement ONE screen from a design bundle as production code in the target codebase's stack. Visual fidelity to the prototype, but native to the target. Designed for delegated waves where each screen is a subordinate lane handled by a fresh task agent. |
Use this skill to reimplement ONE screen from a Claude Design bundle as production code in the target codebase. Subordinate-lane discipline: one screen per task agent.
Required reading (per screen)
The session charter names:
- the bundle path
- the target codebase
- the screen-to-source-to-target mapping
- the production stack to use (vanilla HTML/CSS/JS, React, Vue, native, etc.)
- the viewport floor for mobile fidelity
General order:
- The wave prompt file (charter)
- The bundle's shared chrome / shell prototype (typically
screens/chrome.jsx or equivalent)
- The bundle's shared detail-card / partial pattern (typically
screens/artifact-panel.jsx or equivalent)
- The bundle's canonical token file
- Your assigned screen's prototype source — read top to bottom
- The target codebase's existing house-style page (charter names which one)
- Reference screenshots in the bundle's
uploads/ if present
Implementation rules
- Match the target codebase's stack. The charter names it. Do not introduce a different framework.
- Token-only design values. Every color, font, spacing, radius must come from the bundle's token system. Zero hard-coded values.
- All themes work. If the bundle ships multiple themes, every theme renders correctly via the toggle pattern.
- Shared chrome / shell. Extract the bundle's chrome pattern into a shared partial in the target codebase the first time it's needed; subsequent screens consume it consistently.
- Mock data first. Screens render with realistic mock data inline. Real API wiring is a later wave unless the charter explicitly asks otherwise.
- Mobile-aware. Layout must not break below the viewport floor the charter specifies.
- Accessible. Semantic HTML, alt text on imagery, focus states, keyboard reachable controls (theme toggle, navigation, etc.).
- Bundle-specific copy preserved. Taglines, hero quotes, sub-lines, persona-voice fragments must survive verbatim where the prototype includes them.
Per-screen workflow
- Read your assigned screen's prototype + chrome + tokens
- Sketch the production-stack structure as a comment first
- Write the screen file
- Verify token consumption (grep for hard-coded color values; should find none in production output)
- Verify theme toggle (open mentally; trace which classes/attrs flip)
- Hand off to the fidelity reviewer (separate task, separate skill —
fidelity-audit)
- Fix any Stage 1 (spec compliance) gaps; re-review until PASS
- Fix any Stage 2 (code quality) issues; re-review until APPROVED
- Append to
wave-N-report.md:
## <screen-id> — DONE
- Source: <bundle-path>
- Target: <implementation-path>
- Token fidelity: confirmed (zero hard-coded values)
- Theme support: confirmed (themes: <list>)
- Mock data approach: <one line>
- Stage 1 audit: PASS
- Stage 2 audit: APPROVED
Hard rules
- Never introduce a runtime not in the charter's named stack.
- Bundle prototype is reference, not code to copy.
- Stay in your assigned screen — do not edit other screens or shared files outside the chrome partial unless the charter explicitly asks.
- No real API wiring unless the charter authorizes it.
- Never
git add -A.
Deliverable per screen
- The screen file at the target path the charter mapped
- (First screen only) the shared chrome partial extracted from the bundle's chrome prototype
- An entry in
wave-N-report.md per the format above