agentsclimarketplace

Design critique

Skill bharat3645/The-Ideal-Harness/skills/design-critique

Pre-emit self-critique for UI/visual/design output (components, CSS, copy, layouts) — anti-slop gates, taste dials, and a design-token contract check, run BEFORE the work is shown, not after. Use whenever producing or editing anything with a visual/UX surface.From its SKILL.md

Install
npx -y skills add bharat3645/The-Ideal-Harness --skill design-critique

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 5 stars5 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.2 KB, as published. Nobody here has run it

Design critique

A checklist run against your OWN output before you show it, not a linter run after the fact. This is the one home for several related ideas (hallmark's pre-emit self-critique, a slop-gate list, astryx-style token discipline) folded into a single skill rather than several overlapping ones or a whole new source module — the value here is procedural judgment, which is what a skill is for; building deterministic tooling around subjective taste would be false precision.

Slop gates — catch these before emitting

  • Generic AI-visual-cliché markers: purple/blue gradient backgrounds with no reason, emoji used as section icons, everything centered regardless of content type, decorative rounded corners/shadows applied uniformly instead of purposefully, stock-phrase headers ("Unlock the power of...").
  • Padding/spacing that wasn't decided, just defaulted. If every gap is the same "safe" value, that's the tell of not having actually looked at the layout.
  • Copy that describes instead of communicates. "This section shows your recent activity" above a list literally showing recent activity is filler, not information.

Pre-emit self-critique (do this last, right before showing the work)

Look at what you are about to output and ask, plainly: would someone who does this for a living ship it, or does it read as generated? If the honest answer is "generated," name the specific reason (not "needs polish") and fix that one thing before emitting — vague dissatisfaction doesn't converge, a named defect does.

Taste dials — calibrate, don't default

State (even briefly, to yourself) where this piece sits on the axes that matter for its context: formal ↔ playful, dense ↔ spacious, bold ↔ restrained. A dashboard for financial data and a landing page for a kids' app should not default to the same visual voice.

Token-contract check

If the project already defines a design-token system (CSS custom properties, a Tailwind config, a theme object), every color/spacing/type value in new work should reference a token, not a one-off magic value — an unreferenced #3b82f6 next to a token file that defines --color-primary is a drift bug, not a style choice. If no token system exists, this check doesn't apply — don't invent one uninvited.

Boundary

This is a self-review pass, not a build gate — there's no CI enforcement here, deliberately: taste resists deterministic scoring, and a false-precision linter for it would produce confident-wrong verdicts. It also isn't a replacement for testing the actual rendered result in a browser when the work is inspectable that way — this catches what looking at code alone can catch. Motion/animation taste is a large enough topic to warrant its own skill rather than sprawling this one — see motion-design for anything with a transition, animation, or micro-interaction in it.

Keep looking

Skills are one crate of 627,801. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.