Skip to content

Implement a new design

ID ap-implement-new-design
Kind skill
Selection optional
Category Design / Creative
Targets claude, codex, cursor

Build a chosen visual direction, refine it through bounded fresh-context critiques, and verify the final interface.

Use when you have a visual direction to build and refine, either as a static prototype or in your actual application.

Optional. Select during initialization or when offered in an update, or install separately when you need it.

agents-pack install ap-implement-new-design --dry-run
agents-pack install ap-implement-new-design --yes
Use ap-implement-new-design to build the selected editorial direction as a
standalone static HTML prototype. Include mobile views and local demo
interactions. Use at most two improvement rounds.
Mode When to request it Result
Static prototype HTML mockup, concept preview, or a Studio batch A portable page or self-contained folder with local assets and draft design notes
Application Build or change the actual site, or explicitly integrate a selected prototype Implementation using the application’s framework, routes, components, and real behavior

Choosing a favorite prototype does not switch modes. In prototype mode, edits stay inside the concept directory and leave application files and root DESIGN.md unchanged. Static output is checked for direct-file and offline use.

The skill accepts a selected direction, references, an existing design system, or a clear brief. If the direction is unclear, it explores options and asks for a choice before changing the application. It uses ap-frontend-design for implementation and ap-frontend-review for final checks when available.

When exploration supplies a creative packet, the implementation records and checks its content-native generator, first-viewport silhouette, type roles, probable generic fallback, explicit rejections, and connection to the primary product task before coding. The packet is binding unless it conflicts with legibility, accessibility, or working behavior.

Before screenshots are captured, the rendered design is checked after fonts load at desktop and narrow widths. Accidental text overlaps, near-tangencies, clipped glyphs, and heavy weights applied to too many roles are corrected while preserving any deliberate, readable overlap named in the direction.

Standalone defaults are a target of 8.5/10 and up to three improvement rounds after the initial build and assessment. Studio supplies its own remaining budget instead. Refinement stops when the target is reached, the budget is exhausted, or further changes bring no meaningful improvement.

Every assessment uses a new critic conversation with only the fixed brief, current visuals, role instructions, assessment prompt, and optional fixed references. Previous designs, feedback, scores, source code, and targets stay out of that context. See fresh critiques.

Expect artifact or preview locations, significant refinements, the applicable visual verdict, rounds used, stopping reason, and separate functional check results. If final edits invalidate the last visual score and no assessment budget remains, the final appearance is reported as unassessed.

If isolated visual critique is unavailable, authorized implementation and available checks can continue, with independent critique marked incomplete. An explicit freeze makes later checks audit-only, including reporting any remaining functional defects. Deployment is a separate action.