agentleFS
Discriminative-shoji607Discriminative-shoji607/ui-skills/skills/improve-ui/SKILL.md

improve-ui

Skill3 starsChanged 5 months agoMITView on GitHub

Audit an existing product surface against its own design evidence, identify verified UI problems, and write self-contained implementation plans for another agent. Strictly read-only on product source. Use when asked to review, refine, improve, or clean up an interface without replacing its identity; investigate design-system drift; or prepare a design handoff.

We split this skill into the rules it gives an agent

18 rules. About 0 are common, 14 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈14

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Never modify product source.

    line 12
  2. Distinctive:

    Do not install dependencies, run formatters, commit, push, or otherwise mutate the working tree.

    line 13
  3. Distinctive:

    Do not update design documentation.

    line 14
  4. Distinctive:

    Use rendered evidence only when the user provides it or explicitly requests visual inspection.

    line 15
  5. Distinctive:

    Start from the surface's routes and layouts.

    line 22
  6. Very specific:

    Check for DESIGN.md, repository guidance, and surface-local design documentation.

    line 28
  7. Very specific:

    Write None documented under Explicit exceptions unless a cited source explicitly identifies the exception.

    line 43
  8. Distinctive:

    Before applying the proof gate, inspect every traced surface's user-facing labels, active-state presentation, responsive branches, and …

    line 47
  9. Distinctive:

    Search results, repetition, and implementation differences produce candidates, not findings.

    line 51
  10. Distinctive:

    Before reporting, re-open every cited source and try to falsify each candidate.

    line 65
  11. Distinctive:

    Only findings that survive this pass may enter the table.

    line 74
  12. Distinctive:

    Delete any finding that does not include every required column, including Confidence.

    line 100
  13. Very specific:

    If no candidate survives, write No supported findings were found.

    line 104
  14. Distinctive:

    If findings survive, stop and ask which to turn into plans.

    line 106
  15. Distinctive:

    Read references/plan-template.md.

    line 110
  16. Distinctive:

    Before writing, re-open every cited source, record the current commit when available, identify exact reusable primitives and exemplars, and …

    line 112
  17. Distinctive:

    Do not invent values when the repository provides a token or component contract.

    line 114
  18. Distinctive:

    If asked to reconcile, recheck existing plans against current source and documented decisions.

    line 116

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Improve UI

Audit one coherent product surface against the system that actually governs it. Preserve the product's identity, reuse existing owners, and prefer no finding to an unsupported one. Write plans only for changes the user selects; another agent executes them.

Boundaries

  • Never modify product source. Create or edit files only under design-plans/.
  • Do not install dependencies, run formatters, commit, push, or otherwise mutate the working tree.
  • Do not update design documentation. Record accepted documentation changes in the plan for its executor.
  • Use rendered evidence only when the user provides it or explicitly requests visual inspection.
  • Make every plan self-contained; its executor has no context from the audit or conversation.

1. Select the surface

Honor the user's scope. If the request is broad, select one deployable application and one coherent surface family representing a primary product task. State the selection; do not synthesize the whole repository into one product.

Start from the surface's routes and layouts. Trace the rendered path through compositions, shared components, variants, resolved tokens, and styles. Do not begin with a repository-wide search for inconsistencies.

A connection exists only when it is proven through rendering, imports, props, resolved configuration, CSS inheritance, or a generated artifact loaded by the surface. Shared names, similar tokens, repository proximity, and conceptual relationships do not establish a connection. Exclude other applications, previews, configurators, generated registries, legacy systems, and enterprise variants unless they participate in the traced path.

2. Reconstruct the local system

Check for DESIGN.md, repository guidance, and surface-local design documentation. Use a source only after proving it is current and governs the selected surface; drafts, proposals, migrations, and task lists describe future intent unless explicitly accepted and current. Absence of design documentation is not a finding.

Inspect only the tokens, variables, themes, primitives, variants, and compositions relevant to the traced path. Resolve aliases and variants to their definitions. Classify an implementation as local or legacy only when the repository says so.

Record:

## Design language
- Audited surface:
- Design sources:
- Documented decisions:
- Governing owners and consumers:
- Explicit exceptions:

Write None documented under Explicit exceptions unless a cited source explicitly identifies the exception.

3. Prove findings

Before applying the proof gate, inspect every traced surface's user-facing labels, active-state presentation, responsive branches, and sibling variants for internal contradictions. Treat the results only as candidates.

A finding is in scope only when its correction primarily changes visual presentation, interface copy, layout, component styling, or conformance to a documented design rule. If the correction primarily changes whether product behavior works, reject it.

Search results, repetition, and implementation differences produce candidates, not findings. Keep a candidate only when all three proofs exist:

  1. Contract — Cite a binding design decision for this property and scope, or a direct contradiction in user-facing presentation or content within the same task. “Prefer,” “generally,” names, omissions, repetition, and absence of an exception do not establish a contract.
  2. Runtime — Prove that the cited owner, value, or behavior reaches the affected surface through the traced runtime path. Do not compare separate ownership layers or lifecycle states.
  3. Correction — State one change required by the evidence. If it depends on an existing token, variant, primitive, or exemplar, name it exactly. If the evidence cannot determine the correct choice, the intended condition is ambiguous, the proposal contains alternatives, or the correction requires inventing product intent, reject the candidate.

Source can prove token, typography, color, spacing, layout, copy, component-variant, responsive-presentation, and explicit design-contract violations. It cannot turn functional behavior, state management, or interaction correctness into design findings. Hierarchy, prominence, density, clarity, discoverability, usability, and perceived coherence require rendered or user evidence.

Discard accessibility and HTML/ARIA semantic findings unless the user explicitly requests them. Discard broken routes, redirects, data wiring, action failures, metadata, package API, performance, architecture, and code-quality findings unless the user requested them or a product-specific design contract governs them.

Assign confidence only after all proofs pass. Reuse an existing owner when the evidence supports it; do not create a shared primitive from repetition alone.

4. Vet findings

Before reporting, re-open every cited source and try to falsify each candidate. Delete it when:

  • The problem does not exactly match the cited implementation.
  • The rule does not govern that property and surface.
  • Counterevidence shows the difference is valid or deliberate.
  • The evidence supports multiple corrections.
  • The correction invents product intent.
  • Another finding describes the same root problem.

Only findings that survive this pass may enter the table.

5. Report

Order surviving findings by confidence, user impact, reach, and correction cost. Stop at three.

Use this structure:

## Design language
- Audited surface:
- Design sources:
- Documented decisions:
- Governing owners and consumers:
- Explicit exceptions:

## Findings
| # | Problem | Evidence | Proposed change | Scope | Confidence |
| --- | --- | --- | --- | --- | --- |

## Improve first
<Highest-leverage finding and why, or no supported recommendation.>

Evidence must establish the contract, runtime relationship, and deterministic interface consequence. Proposed change must contain one correction. Delete unsupported or overlapping rows before returning.

Delete any finding that does not include every required column, including Confidence.

Under Improve first, select one surviving finding with the strongest evidence and highest leverage. Never combine findings.

If no candidate survives, write No supported findings were found. under ## Findings and No supported recommendation. under ## Improve first.

If findings survive, stop and ask which to turn into plans. If the user already selected a finding or explicitly requested a plan for a described improvement, continue with that scope. If asked to fix or improve directly, offer a plan; never implement it.

6. Specify selected changes

Read references/plan-template.md. Write one plan per selected change, never one per symptom.

Before writing, re-open every cited source, record the current commit when available, identify exact reusable primitives and exemplars, and trace affected surfaces. Reconcile an existing plan instead of duplicating it.

Do not invent values when the repository provides a token or component contract. Introduce a primitive only after proving why the existing system cannot express the decision and which consumers should share it.

If asked to reconcile, recheck existing plans against current source and documented decisions; update stale evidence, affected surfaces, and status.

Download now (ZIP)

Trending

Steady: 0 new projects took it in the last 4 weeks, none the 4 weeks before.thin data

Week of Aug 30: 0 new projectsWeek of Sep 6: 0 new projectsWeek of Sep 13: 0 new projectsWeek of Sep 20: 0 new projectsWeek of Sep 27: 0 new projectsWeek of Oct 4: 0 new projectsWeek of Oct 11: 0 new projectsWeek of Oct 18: 0 new projects
Aug 2026Sep 2026Oct 2026
New projects taking a version of this file each week, over the 8 weeks our latest run read: only 0 in all, so this is a hint, not a trend.
  • Other versions4 projects use a version of this.

    Word for word · 2

    1. ibelick/ui-skills★ 10k100% same
    2. julianromli/faiz-skills★ 16100% same

    Edited · 1

    1. sanmidable/skills★ 298% same

    Wrote their own version · 1

    1. transitive-bullshit/skills★ 3same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repository5 other files it gives its agents.

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

Reviews can't be read right now.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.