Improve UI
Use this skill to identify the highest-leverage UI improvements for one coherent product surface.
This is an evidence-based audit. Prefer no finding to an unsupported finding.
Do not redesign immediately. Report findings first unless the user already selected a specific improvement to apply.
Use follow-ds-guidelines for design-system conformance audits, heuristic-evaluation for Nielsen-method usability evaluations, and visual-consistency-check for cross-screen consistency checks.
Output Mode
Choose the output mode from the user's request:
- If the user asks to
audit,find improvements, orreview, return findings only. - If the user asks to
improve,fix, orapply, identify the finding being addressed, make only that change, and summarize it. - If the user asks for broad improvement, return the findings first and ask which one to apply.
- If no supported finding exists, say so instead of making cosmetic changes.
Scope
Honor the user's selected target.
If the request is broad, narrow to one coherent surface:
- one screen
- one flow
- one component set
- one Figma Make prototype area
- one family of sibling frames that support the same task
State the selected scope before auditing.
Evidence Sources
Use only evidence available in the current context:
- selected Figma frames, layers, components, or variants
- connected libraries, components, variables, styles, or tokens
- sibling screens or existing product patterns
- attached screenshots, briefs, or reference files
- Figma Make app behavior visible in the current project
- user-stated goals, constraints, or non-goals
Do not infer hidden code, analytics, design-system rules, or product strategy unless provided.
Reconstruct The Local System
Before findings, summarize:
## Design Language
- Audited surface:
- Available design sources:
- Repeated local patterns:
- Connected system elements:
- Explicit exceptions:
Write None visible when a category is unavailable.
Absence of documentation is not a finding.
Finding Gate
A candidate becomes a finding only when all three are present:
Contract: A visible pattern, stated requirement, connected system element, sibling-screen convention, or direct contradiction establishes what should happen.Evidence: The selected target visibly violates or weakens that contract.Correction: One concrete change can be made without inventing product intent.
Reject candidates when:
- the issue is only personal taste
- the correction requires a product decision the user has not made
- evidence supports multiple equally plausible corrections
- the change primarily concerns hidden implementation, performance, metadata, or code architecture
- the target is outside the selected scope
Findings Report
Return at most three findings.
## Design Language
- Audited surface:
- Available design sources:
- Repeated local patterns:
- Connected system elements:
- Explicit exceptions:
## Findings
| # | Problem | Evidence | Proposed change | Scope | Confidence |
| --- | --- | --- | --- | --- | --- |
## Improve First
- <Highest-leverage finding and why, or "No supported recommendation.">
## Next Step
- Ask which finding to apply, unless the user already chose one.
If no finding survives, write:
## Findings
No supported findings were found.
## Improve First
No supported recommendation.
Applying A Selected Finding
When the user selects a finding to improve:
- make only the selected improvement
- preserve unrelated frames, copy, layout, variants, and components
- use existing styles, variables, and components when available
- state any uncertainty before changing product intent
- verify the changed target against the original evidence
Quality Check
Before finalizing:
- Every finding has contract, evidence, and correction.
- Findings are limited to the selected scope.
- Proposed changes are deterministic and actionable in Figma.
- No finding depends on unavailable code or scripts.
- Figma Design and Figma Make concerns are separated.
- The report distinguishes visible evidence from assumptions.