agentleFS
emiliacurieemiliacurie/arely-skills/skills/design-critique/SKILL.md

design-critique

Skill4 starsChanged 3 months agoMITView on GitHub

Runs a design critique panel — four different designers, each with a distinct perspective — on a UI's visual craft (typography, color, spacing, hierarchy, alignment, consistency, flow, and design-system adherence). Works on a live/localhost URL, screenshot image(s), or a Figma file, and handles multi-screen flows. The four lenses are chosen adaptively for what's shown. Delivers all four critiques plus a synthesis in chat. Use when the user says "critique", "design critique", "crit this", "review this design", "feedback on this design", "panel review", "how's the design", or "design feedback". This is visual-craft critique. Usability testing and heuristic audits are separate concerns.

42 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use color-system, presentation-design and 10 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-devSkill

    color-system

    Build a comprehensive color system with palette generation, semantic mapping, and accessibility compliance.

    Also in 3 projects that use this

  • ArnavPuriSkill

    presentation-design

    Design slide decks as HTML/CSS (custom or reveal.js): slide layouts on a fixed 16:9 canvas, presentation typography, slide charts, builds and transitions, speaker notes, and PDF export. Use for decks presented or shared from the browser; for native .pptx files use a PowerPoint-specific tool. Trigger phrases: "presentation", "slide deck", "slides", "pitch deck", "keynote", "reveal.js", "slide design", "presentation template", "speaker notes"

    Also in 3 projects that use this

  • brianciouSkill

    ux-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

  • jpoindexterSkill

    design-system-audit

    Audit code for design system violations - grid spacing, tokens, existing components

    Also in 3 projects that use this

  • agentes100xSkill

    brand-identity

    Proporciona la fuente única de verdad para lineamientos de marca, tokens de diseño y voz/tono. Usa esto para mantener consistencia visual y de comunicación.

    Also in 3 projects that use this

  • nandanosqlSkill

    design-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

The file

We split this skill into the rules it gives an agent

6 rules. 0 are common, 6 are distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 36 edited versions we read keep these.

  • 6

    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
  1. Distinctive:

    Only ask *which* design to critique if it's genuinely ambiguous.

    line 43
  2. Distinctive:

    Every point ties to a goal, a user, or a named principle.

    line 96
  3. Distinctive:

    Prefer observations and questions over directives.

    line 97
  4. Distinctive:

    No AI writing tells: no em-dashes, no "X, not Y" contrast constructions.

    line 100
  5. Distinctive:

    Keep output tight: capped findings, one line each, no restating the checklist, no preamble.

    line 109
  6. Distinctive:

    No web search or extra round-trips.

    line 110

“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.

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

Design Critique Panel

Act as a panel of four different designers giving a real critique. The user hands over a design (often a multi-screen flow) as a localhost/live URL, screenshot image(s), or a Figma file. Cast four relevant designer voices, critique through each lens, then synthesize.

This is a public, shared skill. Users run it on their own accounts, so keep token cost low by default. Economy is built into every step below — follow it, don't skip it.

Step 1 — Resolve the input (do the work yourself; don't ask for selectors or paths you can find)

  • Localhost / live URL → run scripts/capture.py <url> [<url> ...] to save resolution-capped PNGs to the scratchpad, then Read them. This is the cheapest, highest-control path. For a flow, capture each route the user names and keep them in the order given.
  • Screenshot image(s) → if the user gives file paths, run scripts/capture.py --downscale <path> ... to cap resolution before Read. If images are pasted directly into chat, they're already billed at full size — note that paths are cheaper next time, then proceed. A flow = multiple images read as an ordered sequence.
  • Figma → take a rendered screenshot of the frame(s) (mcp__figma-desktop__get_screenshot or mcp__figma-console__figma_take_screenshot) and, only if a design system check is needed, pull a small token list (get_variable_defs / figma_get_token_values). Never call get_file_data or deep design-context dumps — those cost 50k+ tokens. Multiple frames = the flow.
  • Only ask which design to critique if it's genuinely ambiguous. If it's a flow, confirm the screen order so the panel can judge transitions and cross-screen consistency.

Cost heads-up (only on heavy paths). Stay quiet on the normal path. Only when about to do something genuinely token-heavy — a Figma file read, a full-page capture of a long page, or a flow of more than ~6 screens — give a one-line heads-up with the rough cost and a cheaper alternative, then wait. Example: "Reading the full Figma file is pricey (~50k tokens). Frame screenshots or localhost links are roughly 5x cheaper — want me to go that way?" Everything cheap just runs.

Step 2 — Drip three quick questions (one at a time, plain language)

Announce up front that you have up to three quick questions, then ask one at a time. No fill-in-the-blank templates. Skip any the user already answered in their request.

  1. Goal, audience, stage — what is this flow for, who's it for, and is it early concept or near-ship? (Sets the bar; the panel critiques against goals, and stage controls how nitpicky to get.)
  2. Design system to apply — a named system, a CSS/token file, Figma variables, or "none / universal principles"? If a system is named or detectable, inspect it and check adherence; otherwise critique against universal fundamentals.
  3. Focus or "do not touch" areas (optional) — anything specific to zoom in on or leave alone?

Step 3 — Cast the panel (adaptive)

From the roster in references/critique-checklist.md (Part A), pick the four most relevant designer perspectives for what was shown (e.g. add a Content/UX-writing voice for a text-heavy screen, a Conversion voice for a landing flow). Give each a short role label. The four are chosen per run, not fixed.

Step 4 — Analyze

Evaluate against references/critique-checklist.md (Part B), scoped to the stated goal, the design system, and the flow as a whole (not just isolated screens). Each designer works through their own lens; overlap between them is fine and useful.

Step 5 — Deliver the panel critique in chat

Output only — no saved file. Use this structure:

  1. What I'm looking at — one line: the flow and screen order, stated goal, audience, stage, and which standard is being applied.
  2. The panel — the four designers chosen, each with a role label and one line on why they're in the room.
  3. Each designer's critique — one block per designer, in their own voice:
    • What's working — 1–2 genuine strengths through their lens.
    • Findings — their 3–4 highest-value issues, each tagged Blocker / Major / Minor / Polish, tied to a principle or the stated goal (never personal taste), with where it is (screen + element) and a concrete, rationale-backed direction. One line each. The designer owns the final call.
  4. Where the panel agrees — points raised by 2+ designers = high-confidence priorities.
  5. Where they disagree — genuine tensions surfaced as decisions for the user, not resolved silently.
  6. Top 3 next steps — highest-leverage fixes across all four voices, in order.

Critique etiquette (non-negotiable)

  • Critique the work, never the person.
  • Every point ties to a goal, a user, or a named principle. No "I'd make it blue."
  • Prefer observations and questions over directives; the designer owns the decision.
  • Match depth to stage. Don't nitpick spacing on an early concept that might get cut.
  • Name the problem clearly rather than silently redesigning it.
  • No AI writing tells: no em-dashes, no "X, not Y" contrast constructions.

Token economy (keep it cheap)

  • Image resolution is the main cost driver (≈ width × height ÷ 750 tokens). Capture at ~1000px wide, viewport height, never full-page 4K.
  • Downscale provided image paths before Read.
  • Figma = screenshot + small token list only. No file dumps.
  • Role-play the four voices in one pass. Do not spawn subagents.
  • Keep output tight: capped findings, one line each, no restating the checklist, no preamble.
  • No web search or extra round-trips.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.