| name | design-critique |
| description | Design-direction advisor + 5-dimensional design review for visual work.
Pick a design philosophy from a library of 20 named styles (Pentagram, Field.io,
Kenya Hara, Sagmeister, etc.) before designing, or score a finished design on
philosophy alignment / visual hierarchy / craft quality / functionality / originality
(each 0-10) with a Keep / Fix / Quick-Win list. Includes an anti-AI-slop checklist
for resisting generic AI default aesthetics. Derived from
alchaincyf/huashu-design (Personal Use License — see LICENSE + ATTRIBUTION.md).
|
| triggers | ["design review","critique this design","rate this design","5-dim review","design direction","design philosophy","what style should this be","pick a style","is this generic","anti ai slop","score this design"] |
design-critique
A judgment skill, not a production pipeline. Use this when the question is
"is this design any good?" or "what direction should this design take?" —
not when the question is "make a design for me." If you need to produce
HTML, slides, animations, or video, this is the wrong skill.
When to invoke
- Reviewing a tutorial page, landing page, banner, plugin card, palette,
or any other shipped visual artefact and you want a structured critique.
- Picking between several design directions for a piece of work that
hasn't started yet.
- Auditing a piece of design for the "generic AI slop" failure mode where
the output reads as "yet another AI-generated page" with no brand
identity.
When NOT to invoke
- Editing copy for veracity → use the
honest-copy skill instead.
- Generating an Excalidraw diagram → use the
hand-drawn-diagrams skill.
- Choosing colours from a palette → use the
color-expert skill (or
consult references/design-philosophies.md here for style-driven
palette starting points).
- Producing slides, animations, or videos — this skill explicitly does
not cover production pipelines (the upstream skill does, that material
was stripped).
References (read these as needed)
| File | When to consult |
|---|
references/design-philosophies.md | Before designing, when picking a style direction. 20 named philosophies grouped in 5 schools (Information Architecture / Motion Poetics / Eastern Minimalism / Experimental Avant-garde / Asian-Contemporary), each with core characteristics + an English prompt DNA block usable directly for AI image generation or HTML design. Bilingual EN/CN. |
references/critique-rubric.md | When reviewing a finished design. The 5-dimensional rubric with 1-10 scoring tables for each dimension, per-scene emphasis notes (web / PPT / PDF / infographic), and a common-cliché checklist. |
references/anti-slop.md | When the work risks looking generic-AI. Lists what to avoid (purple gradients, emoji icons, rounded card + left border accent, SVG-drawn people, Inter-as-display, etc.) and what to do instead. |
Eight core principles (condensed from upstream)
These apply whether you're directing a design or critiquing one.
- From existing context first. Designing from scratch is the last
resort. Look for a design system, brand kit, codebase, Figma file, or
reference screenshots before generating anything. If none exists and
requirements are vague, switch to direction-advisor mode (propose 3
differentiated directions from
design-philosophies.md) before
producing pixels.
- Asset protocol. When the work involves a specific brand or product,
find the real logo, product imagery, and UI screenshots first. CSS
silhouettes and SVG-drawn approximations destroy brand recognition —
the whole point of a brand design is to be recognisable. (For
non-funkworks brand work; funkworks itself uses its own banner +
mascot.)
- Show assumptions before executing. When the brief is ambiguous,
state what you're about to do and why before doing it. Treat the user
as your manager.
- Give variations, not "final answers." When time allows, produce
2-3 differentiated takes rather than one polished output. Lets the
user steer based on what they see.
- Placeholder beats bad implementation. Honest
[product photo pending] placeholder beats a generic AI-generated stand-in that
degrades brand identity.
- System first, don't fill. Empty space is design content. Don't
pad with decoration to "fill the page." Negative space is a tool.
- Anti-AI-slop is brand protection, not snobbery. Default AI
aesthetics (purple gradients, rounded cards with left accents,
stock-emoji bullets, Inter-as-display) average across all training
data, so they erase the brand you're trying to surface. Resist them
in favour of brand-specific choices.
- Fact-verify before claiming product specs. Any assertion about a
specific product's existence, version, release date, or hard specs
demands a WebSearch before being written into copy or design. Don't
design around "I think Pocket 4 hasn't launched yet" — search first.
Critique output format
When invoked for a review, structure the output like this:
## 5-Dimensional Review · <artefact name>
| Dimension | Score | Notes |
|------------------------|:-----:|-------|
| Philosophy alignment | N/10 | … |
| Visual hierarchy | N/10 | … |
| Craft quality | N/10 | … |
| Functionality | N/10 | … |
| Originality | N/10 | … |
| **Total** | NN/50 | |
### Keep
- What's working that should not change.
### Fix
- ⚠️ **Critical** — must address before ship
- ⚡ **Important** — should address this pass
- 💡 **Optimisation** — nice-to-have
### Quick Wins (5-minute fixes, ranked)
1. …
2. …
3. …
Critique the design, not the designer. Per-dimension rubrics live in
references/critique-rubric.md — consult them rather than improvising
scoring criteria.
Attribution
Derived from alchaincyf/huashu-design
under the Huashu Design Personal Use License. The animation / video / slide /
TTS / HTML-as-medium production pipeline material from the upstream was
intentionally stripped — this skill keeps only the design-judgment portion.
Full upstream license terms in LICENSE; see ATTRIBUTION.md for what was
kept versus dropped.