design-critique
Skill2 starsChanged 12 days agoMITView on GitHub
Use whenever someone wants a design judged, in any words, any length or language: "what do you think", "is this good", "why does it feel off", "tear this apart", "will people get it", "which one is better", "give me feedback"; critique, review, audit, score or compare a screenshot, mockup, wireframe, HTML page, live site, flow, dashboard, chart, form or a described design; findings with severity, anchored scores, a fix order, what to keep; not for building or applying fixes, even a final polish pass (do the work instead), logo or identity critique (brand-identity) or code review.
42 other projects use a version of this.
Adopted by
Most recently taken up by
- cassianodiniz/Titantheir own version, first seen Nov 16, 2026
- Vujavujavuja/Stillmotion-onbrandtheir own version, first seen Nov 16, 2026
- emiliacurie/arely-skillstheir own version, first seen Nov 16, 2026
- cassianodiniz/casstheir own version, first seen Nov 16, 2026
- Hydralune/frame-forgetheir own version, first seen Nov 15, 2026
- nisarth/claude-skills-storetheir own version, first seen Nov 8, 2026
Used with
Projects with it also use color-system, ux-writing and 8 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
launchapp-devSkillcolor-system
Build a comprehensive color system with palette generation, semantic mapping, and accessibility compliance.
Also in 3 projects that use this
brianciouSkillux-writing
UX writing guidelines for reviewing and optimizing interface copy. Use when the user asks to review, improve, or rewrite UI text including button labels, error messages, onboarding flows, or any interface copy. Apply when auditing microcopy for web apps, mobile apps, or any digital product requiring clear, user-centered communication.
Also in 5 projects that use this
jpoindexterSkilldesign-system-audit
Audit code for design system violations - grid spacing, tokens, existing components
Also in 3 projects that use this
nandanosqlSkilldesign-brief
Create a design brief that clearly communicates the product context, user needs, constraints, and success criteria to the design team — bridging PM requirements and design execution.
Also in 3 projects that use this
GitHubSkillaccessibility-audit
Audits actual components and routes against applicable WCAG 2.2 A/AA criteria using consumer-owned design-system evidence and repository tools. Use when asked for an accessibility audit, a11y check, keyboard or focus review, contrast check, or screen-reader verification.
Also in 4 projects that use this
soilmassSkillmotion-design
Create animations and motion graphics using industry standards. Use when asked to animate UI elements, create video content, implement transitions, or build motion systems. Implements W3C Design Tokens, Material Design 3 baseline, and Lottie specification compliance.
Also in 3 projects that use this
Other versions42 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 42
- anthropics/knowledge-work-plugins★ 36ksame name
- cuellarfr/design-skills★ 94same name
- adityareddy-dev/deskstudio★ 43same name
- tdimino/claude-code-minoan★ 41same name
- launchapp-dev/animus-cli★ 39same name
- ArnavPuri/designskills★ 17same name
- henrique-simoes/Istara★ 15same name
- aiatelie/ai-atelie★ 9same name
- mae616/ai-template★ 9same name
- shayan-m81/layerwright★ 8same name
32 more not listed
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 repository25 other files it gives its agents.
- agent-interface-designSkill
- brand-identitySkill
- chart-designSkill
- color-paletteSkill
- data-tablesSkill
- design-system-builderSkill
- form-designSkill
- generated-ui-restyleSkill
- interface-behaviorSkill
- journey-and-service-mappingSkill
- layout-structureSkill
- practice-and-assessmentSkill
- presentation-designSkill
- pricing-and-packagingSkill
- responsive-layoutSkill
- sales-conversationsSkill
- showing-workSkill
- spacing-and-groupingSkill
- teaching-interfacesSkill
- typeface-selectionSkill
- typesettingSkill
- ui-motionSkill
- ui-polish-passSkill
- user-research-basicsSkill
- visual-hierarchySkill
The file
We split this skill into the rules it gives an agent
3 rules. 0 are common, 3 are distinctive, and 0 are very specific to this project.
0
Common
Most of the 36 edited versions we read keep these.
3
Distinctive
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 38
When a count is asked for (three fixes, two options, top five), give exactly that count.
- Distinctive: line 73
Scores: Only for audits or when asked: 0 to 4 per applicable dimension with a one-line reason, whole numbers, anchors in report format.
- Distinctive: line 82
Every number shown was computed from what was supplied.
“Common” is measured: kept by at least 6 in 10 of the 36 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Design critique
Judge a design like a careful colleague: what it is for, what works, what fails and for whom, how bad each problem is, what to change first. The default failure is vague advice (make it pop, add whitespace) and long unranked lists. The fix is evidence, severity, an order of attack, and honesty about what could not be seen.
Do not produce
- A finding with no element, number or reason behind it.
- A flat list mixing taste, function and polish; a critique with nothing kept.
- Color and type nitpicks on a sketch or gray wireframe.
- Conventional polish demanded of a deliberate style (brutalist, dense, playful) instead of judging it by its own intent.
- A redesign or rewrite when a critique was asked for, or a critique when something was asked to be built.
- A survey or opinion poll where watching a few people would answer the question.
- Manipulative patterns recommended as conversion tactics.
- Invented evidence: a contrast ratio, size, count, analytics figure, user quote or test result that was not supplied or computed from what was supplied.
When not to use
- Building, fixing or restyling: do the work (applying fixes includes the last small polish pass on a nearly finished screen); give a critique only if one was asked for, never in its place.
- Logos, wordmarks, identity systems (brand-identity, if installed); code correctness, security or performance review.
- Research programs (surveys, statistical tests): give a short recommendation only.
When facts are missing
- Infer purpose, audience, main task and fidelity from the material first. Ask once only if the chat is interactive, the question is cheap and the answer would change the verdict; give a default with it. Never reply with only questions.
- Otherwise deliver on stated assumptions, in one line ("assuming a first-time visitor deciding whether to sign up; high fidelity").
- Nothing to inspect (no markup, no description detail)? Say what cannot be judged, critique what is given, list what to send. Do not invent the design.
- A number you cannot compute is "not checked: contrast", never an estimate dressed as a measurement. Label a rough estimate as one.
- The user's brand, design system, platform conventions and stated constraints win over generic taste.
Deliver
- Open with the verdict or the top finding. No preamble, no restating the design, no design theory.
- When a count is asked for (three fixes, two options, top five), give exactly that count. Honor any length cap by dropping the lowest-value points, and count rather than assert.
- Scale to the ask: a one-line question gets a verdict plus the few findings that matter; an audit gets the report in report format. Short is not small: keep evidence.
- Notes beyond what was asked: at most two short lines (assumptions, what was not checked).
- Compute instead of asserting: contrast ratios, target sizes, line lengths, counts of font sizes, colors and button styles, clicks to the goal, from the supplied code or description.
- Each finding: what is on screen, who is affected and how, a concrete change. Mark taste as taste.
- Severity on every finding; blockers first. Weigh reach: a minor problem on every screen can outrank a major one on a rare path.
- Group repeats into one finding with a count. Report patterns, not every instance.
- Name two to four things that work and should stay, with a reason each.
Judgment calls
What the design is for comes first. Check the first screen before details: a newcomer can name the offer and their next action. Failing that is usually the top finding. On an inner page they can tell where they are and how to leave or search.
Match depth to fidelity. Sketches and wireframes: purpose, structure, flow, content priority, task fit. High fidelity adds type, color, spacing, states, copy. Shipped adds accessibility, consistency, perceived speed. Polish notes on a rough piece only if asked, labeled premature.
Taste or function. Lead with function (the task, errors, legibility, access). A preference ("I don't like the blue") becomes a question about the goal it should serve, plus contrast and brand fit; keep it labeled as preference. When a team is arguing over taste, recommend settling the criteria before looking at options, one named decision-maker, and a presenter who recommends one option with reasons rather than asking "which do you like?".
Deliberate style. Judge against its own intent; flag only what harms use (contrast, legibility, navigation, focus, labels). If intent is unstated and the style hurts the goal, ask about intent rather than rule.
Cognition over theory. Tie a finding to what a person must notice, read or remember, and to a measurable fact (distance, contrast, seconds, clicks), not to a named law or a rule of thumb about reading.
Comparing variants. Decide for the stated audience and task, cite the specific differences, say how sure you are. Say "close call, test it" only when the evidence is really balanced, and name the cheapest test.
Does it already work? Say so plainly and list only real issues. Do not invent problems to look thorough.
Finished versus formed. A screen can have every state and still be unformed. Judge whether the main task works end to end and whether anything in it was clearly decided for this audience, not only whether it is polished.
Comprehension questions. If a finding hinges on whether people understand something, recommend three to five people doing real tasks without help, then a ranked list of what failed; fixes should remove things before adding explanations. Use analytics or a larger study only when the question is volume.
Audits at scale. Inventory the screens, sample key flows and one screen per pattern, check consistency (duplicate components, one-off values, mismatched labels), report patterns with counts, give a deliverable shape. For a huge set, deliver the first slice in full, then continue.
Vague feedback. Rewrite each phrase as an observable issue, the reason, and a fix ("too busy" becomes "six type sizes and four accents compete in the first viewport; keep one accent, cut to three sizes"). If it is pure preference, say so and ask what it should achieve.
Manipulation. Hidden or late costs, forced accounts, pre-checked consent, decline buttons worded to shame, false urgency or scarcity, cancel harder than sign-up, defaults that work against the person: report as findings with an honest alternative. If asked for them, decline that part in one sentence and offer two or three honest routes. Describe what is on screen; do not label on a hunch.
Scores. Only for audits or when asked: 0 to 4 per applicable dimension with a one-line reason, whole numbers, anchors in report format. A blocker caps the overall; two 4s do not offset a 0. Mark dimensions that cannot be assessed.
Checks
checks holds short passes for interaction, state and errors, accessibility, content, data displays and manipulative patterns. Load it when the design contains those parts. For full procedures use visual-hierarchy, spacing-and-grouping, layout-structure, typesetting, color-palette, form-design, ui-motion, chart-design and data-tables, each if installed; otherwise use ordinary judgment. Detailed component accessibility semantics and multi-chart dashboard layout are outside these skills: use ordinary judgment.
Before sending
- First line is the verdict or the top finding.
- Every number shown was computed from what was supplied; estimates are labeled.
- Findings are ranked, each with severity and a fix; taste is labeled.
- The count and length match the ask; at most two notes.
- Nothing was built or redesigned that was not asked for.
- What was not checked is stated.
References
- references/report-format.md: load for a full review, audit or scored review; order, finding template, severity labels, score anchors, compact template.
- references/checks.md: load when judging flows, forms, states, errors, accessibility, content, charts, tables, dashboards, pricing, consent and cancellation, or a feature where a model generates or suggests content or software acts for the person.
- references/accessibility-review.md: load for an access review or audit, or any ask whether something is accessible or compliant; method, proof labels, blockers, what scanners miss.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
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.

