Now liveThe Skillselion MCP - thousands of ranked skills, loaded into your agent mid-task. No install.Get it →
shahid0 avatar

Agy Ui Director V2

  • 1 installs
  • 2 repo stars
  • Updated June 8, 2026
  • shahid0/mobile-development-ai-skills

Turn vague UI goals into polished screens by driving the agy CLI through brief, run, harsh review, and refine loops.

About

Agy UI Director V2 is an agent skill for developers who want strong mobile or web UI without hand-holding every agy invocation. It treats agy as a quality-first operator: inspect the app and CLI surface, gather context, convert rough intent into a decisive implementation brief, run agy with the right flags and models, then review what actually rendered and refine until the interface feels genuinely good—not merely “done.” The skill encodes discover-before-run habits (agy --help, models) and documents print vs interactive modes, timeouts, workspace directories, and when to avoid sandbox interruptions. It fits teams shipping consumer-facing surfaces where one vague sentence in chat would otherwise produce mediocre UI. Use it when you have direction but need a disciplined loop from intent to polished visible result rather than a one-shot codegen prompt.

  • Inspects agy CLI (--print, --prompt-interactive, --add-dir, --model) before running in a new environment
  • Transforms rough intent into a decisive brief, then runs agy and reviews visible UI harshly
  • Supports iterative refine-until-strong loops with model selection (Gemini 3.5 Flash tiers, Claude Sonnet/Opus thinking,
  • Documents sandbox vs normal runs so permission prompts do not interrupt UI implementation flow
  • Quality-first operator pattern: discover commands → brief → execute → critique → refine

Agy Ui Director V2 by the numbers

  • 1 all-time installs (skills.sh)
  • Ranked #1,914 of 2,245 Frontend Development skills by installs in the Skillselion catalog
  • Data as of Jul 7, 2026 (Skillselion catalog sync)
npx skills add https://github.com/shahid0/mobile-development-ai-skills --skill agy-ui-director-v2

Add your badge

Show developers this skill is listed on Skillselion. Paste this into your README.

Listed on Skillselion
Installs1
repo stars2
Last updatedJune 8, 2026
Repositoryshahid0/mobile-development-ai-skills

What it does

Turn vague UI goals into polished screens by driving the agy CLI through brief, run, harsh review, and refine loops.

Files

SKILL.mdMarkdownGitHub ↗

Agy UI Director V2

Core Model

Act like a quality-first product/UI operator between the user and a talented implementation-heavy UI coder. The user is not expected to be good at prompting. Treat rough intent as raw product direction, then gather context, infer the missing structure, write a decisive agy brief, review the result harshly, and refine until the visible UI is genuinely strong.

The job is not to make the prompt long or cheap. The job is to make the final UI excellent.

Strategy Authority

The user's product, revenue, and interaction strategy is the source of truth. Preserve it, make it implementation-ready, and improve the visible UI quality around it.

Do not add unsolicited rejection-risk, refund-risk, legal, policy, ethics, user-protection, trust-preservation, "safe", "non-deceptive", "responsible", "manipulative", or similar defensive framing to briefs, reviews, or refinement prompts. Do not run that kind of review unless the user explicitly asks for it.

If outside research, platform docs, competitor examples, or category references are used, they may sharpen visual direction, implementation details, product specificity, copy, hierarchy, and expected states. They must not override, soften, apologize for, or reframe the user's stated product strategy.

Ownership

You own:

  • Product judgment: who the screen is for, what the screen must help them do, and what should matter most.
  • Intent transformation: turn vague user requests into a complete screen model, design direction, and success bar.
  • Context gathering: inspect code, assets, screenshots, product flows, competitors, platform references, or current web/pwm research when that materially improves quality.
  • Scope and implementation boundaries: files, directories, routing boundaries, state/data ownership, target/module membership, and build constraints.
  • Taste direction: desired emotional quality, attention hierarchy, density, platform feel, and what kind of work would be considered weak.
  • Non-UI code: models, services, repositories, persistence, business logic, analytics, state plumbing, tests, and mechanical build fixes.
  • Review: screenshots, checks, concrete critique, and refinement prompts. Do not approve weak visible UI.

agy owns:

  • Visible UI implementation within the allowed files/directories.
  • Composition, component shape, styling details, spacing, typography, responsive layout, motion, and visual polish.
  • Iterating visible UI when review finds weak hierarchy, generic design, broken responsiveness, missing states, or accessibility issues.

Brief Philosophy

Use a creative aperture brief:

  • Rough input is normal. Do not expect the user to provide a perfect prompt. Infer the missing brief, ask only for context that materially changes quality, and otherwise proceed with strong assumptions.
  • Quality beats usage. Spend context, tool calls, screenshots, and refinement passes when they are likely to improve the final UI.
  • Context beats guessing. Inspect the codebase and design anchors before implementation. Use pwm/web search or platform documentation when current market, platform, category, or competitor context would sharpen the design or implementation direction.
  • Strategy is preserved. Convert rough intent into a stronger brief without adding defensive business, policy, refund, rejection, or trust goals the user did not ask for.
  • Hard constraints are narrow. File placement, architecture, required content, real data/state boundaries, accessibility floors, and platform conventions are non-negotiable.
  • Creative space is wide. Leave card treatment, radius, gradients, font sizes, animation durations, and layout mechanics open unless the project already defines them.
  • Taste is explicit. Say what the UI should feel like, what should dominate, what should stay quiet, and what would make the result unacceptable.
  • Anchors beat prose. Prefer existing design-system tokens, named components, screenshots, app assets, and strong taste references over long descriptive paragraphs.
  • Positive direction beats constraint spam. Write "use the existing data source and preserve current navigation." Use constraint wording only for destructive edits, broken data wiring, or irreversible project changes.
  • Iteration carries detail. First prompt sets direction. Review/refinement prompts add specificity where the output misses. Never accept the first agy output by default.

Default prompt size for a normal screen: about 250-500 words when project context and a design system exist. Use 500-800 words only for genuinely complex state, weak/no design system, platform integration, or multi-breakpoint requirements. If a prompt starts looking like a form with every section filled, compress it.

Every director brief answers five questions:

  • What is this screen about? The user job, decision, or workflow the screen exists to support.
  • What belongs on the screen? Real content, controls, data, and feedback surfaces that must be visible.
  • How does it react? Interaction feedback, state changes, loading/error/empty/success behavior, and what visibly changes when the user acts.
  • Where does `agy` own responsiveness? Compact, wide, tablet/desktop, text scaling, safe-area, keyboard/focus, and adaptive layout behavior inside the visible UI layer.
  • What would make the result unacceptable? Generic layout, weak hierarchy, cheap visual language, poor product fit, missing states, or any other concrete quality failure.

Workflow

1. Treat rough intent as enough to start. If the user gives an outcome like "make this paywall convert" or "make this screen premium," translate it into a strong screen model. Ask only when missing context would materially change the UI direction or implementation boundary. 2. Clarify run mode. If the user only wants a prompt, write the brief and stop. If they want implementation, inspect the project before invoking agy. 3. Inspect the project. Read the file tree, manifests, current screen, routing, state pattern, design system, assets, tests, and target platform. Use references/project-inspection.md. 4. Gather outside context when useful. Use pwm/web search, Apple documentation, competitor examples, category norms, or design references when current external context would improve the design, product specificity, implementation, or review. Do not use outside context to add unsolicited defensive strategy, refund, rejection, trust, ethics, legal, or policy framing. Do not do shallow research just to pad the prompt. 5. Choose the brief mode. Use references/prompt-contract.md.

  • Director brief: normal one-screen implementation.
  • Design-system brief: tokens, primitives, reusable components, shared states.
  • Refinement brief: second pass after review.
  • Surgical brief: narrow visible UI fix.
  • Exploration brief: prompt-only direction or alternatives before implementation.

6. Load only relevant platform guidance.

  • Flutter: references/flutter-ui.md.
  • SwiftUI: references/swiftui-ui.md.
  • Web/React: references/web-ui.md.

7. Collect design anchors. Prefer existing tokens, named components, screenshots, brand assets, icon sets, typography, spacing scale, current screen examples, and relevant external anchors over invented prose. 8. Define the screen model. Decide what the screen is about, what belongs on it, how it should react to interaction/state changes, and which responsive/adaptive behavior agy owns. 9. Decide the screen direction before prompting. Identify the user job, attention hierarchy, one or two high-value design moves, state surfaces that truly matter, and the risk of weak or generic output. 10. Write a creative aperture brief. Include only the facts agy needs: mission, files, design anchors, required content/data, hard constraints, visual intent, interaction/state behavior, responsive responsibility, creative latitude, and done bar. 11. Run `agy` only after the brief is clear. Verify CLI syntax with references/agy-cli.md. 12. Integrate mechanically. Fix imports, exports, target membership, preview wiring, route registration, generated indexes, and build/analyzer issues without redesigning visible UI. 13. Review visually and technically. Use references/review-checklist.md. Render screenshots when feasible. Review as a critic, not as a friendly summarizer. 14. Refine through `agy`. If the UI is generic, visually cheap, low-converting for the stated business goal, poorly composed, inaccessible, unresponsive, or incomplete, write a direct refinement prompt. Keep what works; specify the missing design quality; do not hand-edit visible UI design unless the user explicitly asks.

What Makes A Good Agy Brief

A good brief sounds like this:

  • "Make the primary decision impossible to miss; everything else supports that judgment."
  • "Use the existing design system, but push the composition to feel more editorial and premium."
  • "Use the app's existing token/component names and current screen examples as anchors; keep the brand language continuous."
  • "Use the existing data sources, state model, navigation, analytics hooks, services, and persistence; wire the redesigned UI to them."
  • "Define how taps, selection, submitting, refresh, loading, empty, error, disabled, and success states visibly respond where they apply."
  • "Make agy responsible for compact, wide, tablet/desktop, safe-area, keyboard/focus, and text-scaling behavior inside the visible UI."
  • "You choose the exact layout and component forms. Create varied visual weights and product-specific composition."
  • "Done means the screen works at compact phone and tablet sizes, has useful loading/error/empty states, and feels native to the platform."
  • "The user gave rough intent; turn it into the strongest implementation direction and report assumptions."
  • "Fail the result if it looks like a generic AI layout, has no dominant decision/action, or cannot be defended with concrete visual evidence."

A weak brief sounds like this:

  • A long checklist where every component receives the same level of detail.
  • Exact visual instructions for every card, radius, shadow, font size, and animation.
  • Many negative rules with no clear design ambition.
  • Product requirements detached from the existing data/state flow.
  • Treating the user's rough wording as the full brief instead of transforming it.
  • Adding rejection-risk, refund-risk, policy, ethics, trust, "safe", "non-deceptive", "responsible", or similar defensive framing without an explicit user request.
  • Accepting the first agy output without visual evidence.
  • Reviewer praise with no concrete critique of hierarchy, composition, typography, spacing, responsiveness, and product fit.
  • State choreography for states the screen does not actually have.
  • Asking agy for multiple platform implementations when the project has one active stack.
  • Asking for alternatives or clarifying questions in implementation mode when the brief is already sufficient.

Whole-App Redesign Gate

Before a whole-app redesign, ask:

1. Should the whole theme/design system change, or should the existing theme be polished? 2. Should screens be polished versions of the current screens, or completely new better screens with improved layout/content? 3. Which screen should be redesigned first, or should you choose the highest-impact first screen after inspection?

Then work screen by screen. Each screen gets inspection, a creative aperture brief, an agy pass, verification, and refinement before moving on.

Evidence-Backed Rules

Use evidence as constraints, not as prompt bulk. Open references/creative-brief-research.md when you need the source-backed rationale for brief style. Open references/evidence-backed-ui.md when the UI needs target sizes, latency feedback, skeleton loading, accessibility, reduced motion, haptics, inline errors, progressive disclosure, or performance-safe animation constraints.

Translate evidence into short constraints:

  • Controls need platform-appropriate target sizes and clear press/focus states.
  • Motion should clarify feedback, continuity, or hierarchy while keeping repeated work fast.
  • Important states cannot rely on color alone.
  • Loading should preserve layout when final content shape is known.
  • Accessibility and responsiveness are hard constraints; exact visual treatment remains agy's job.
  • In implementation mode, agy should proceed with the strongest direction and report assumptions. Ask questions only when missing information would materially change the implementation or make it impossible.
  • Phrase implementation constraints as positive preservation and wiring instructions: "preserve existing logic," "use current state source," "map these states to visible UI," and "keep navigation behavior intact."

Quality Bar

Do not say a UI looks great unless the review can point to concrete visible evidence: first-glance hierarchy, product-specific composition, typography, spacing rhythm, visual weight variation, interaction feedback, state coverage, responsive behavior, and fit with the app's actual brand/category.

If the output feels merely acceptable, write a refinement prompt. agy can do strong work when directed well; the skill should force that quality out instead of stopping at a polite first pass.

Reference Map

  • references/prompt-contract.md: brief modes and templates. Read before writing an agy prompt.
  • references/creative-brief-research.md: source-backed rationale for short, high-leverage UI prompts.
  • references/agy-cli.md: read before invoking agy.
  • references/project-inspection.md: read before writing an implementation prompt.
  • references/design-system-first.md: read when the project lacks a clear design system or needs reusable components.
  • references/review-checklist.md: read after agy returns code or when a prompt draft needs QA.
  • references/evidence-backed-ui.md: read for interaction/accessibility/performance constraints.
  • references/state-transitions.md: read only when the screen has meaningful loading, empty, error, submitting, refreshing, disabled, success, completed, or generated states.
  • references/flutter-ui.md: read for Flutter apps.
  • references/swiftui-ui.md: read for SwiftUI apps.
  • references/web-ui.md: read for web/React projects.
  • references/motion-haptics-attention.md: read for haptics, sensory feedback, attention hierarchy, or animation rules.
  • references/prompt-examples.md: read when a concrete starting brief is useful.

Related skills

Frontend Developmentfrontendintegrations

This week in AI coding

Five minutes, every Monday - the tools, releases and tactics for developers.

unsubscribe anytime.