Impeccable vs Taste Skill: which Claude design skill to use, tested

Short answer: pick Taste Skill for a clean landing page at a modest cost, and Impeccable when the page has to be memorable, has to stay truthful about your product, or is an app or dashboard rather than a marketing page. In our test on the same brief, Taste Skill produced a calm, tidy page for $2.28 in 7 minutes; Impeccable produced a far bolder page, and the only one that invented no product facts, for $7.49 in 24 minutes.

Both are among the most popular design skills for Claude Code, Codex and Cursor, and people often ask which to install. They solve the same problem, generated pages that all look alike, in very different ways. This page compares them on what they are, what they produced from one brief, what they cost and which jobs each one suits.

Try it on a real task: Impeccable is free for 25 runs a month.

#Impeccable vs Taste Skill at a glance

#The test: one brief, both skills

We gave Claude Code the same brief for each run: the landing page for Ledgerline, a fictional bookkeeping app for freelance designers, with a hero, three steps, features, one $12 plan, an FAQ and a footer. It had to be one self-contained HTML file with Google Fonts as the only external resource, and it could not invent testimonials, customer counts, logos or awards. Both runs used Claude Code 2.1.278 with Claude Opus 5 on 9 October 2026, with the skill's files in the project and the Skill tool switched off so no other installed skill could interfere. Each skill ran once, so this is an honest example rather than a benchmark; a second run would produce different pages.

#What each skill did before writing code

Impeccable ran its context command, found no product file, and wrote a PRODUCT.md that separates what the brief said from what it did not. It listed the unknowns explicitly ("which banks… data export… card-required-or-not… must not be invented"), then rolled a seeded concept from a shortlist grounded in the subject. It landed on a Swiss departure board: money arriving and leaving as a timetable, with the quarterly tax date as the next departure.

Taste Skill declared its design read in one line ("SaaS landing for solo freelance designers… with a calm, type-led, honest-tool language"), set its dials to variance 7, motion 5 and density 4, lowering variance "because this is a money app and trust matters", and resolved two conflicts with the brief in writing: no icon library, and a working mini-product in the hero instead of a fake screenshot.

#Results

†Impeccable's own anti-pattern detector, run on both finished pages, leaving out its cramped-padding rule (false positives on pages that compute padding with calc()) and its design-system rules (which only apply to a page with its own DESIGN.md). With every rule counted: 55 and 6. Most of Impeccable's remaining findings are all-caps body text, which is its departure-board style rather than an oversight.

For reference, the same brief with no skill cost $0.98, took 3.3 minutes, invented similar product facts and was the only page that scrolled sideways on a phone.

#Where Impeccable wins

• It doesn't make things up. Its rule "Ask before replacing factual copy or adding claims", and the PRODUCT.md it writes when it cannot ask, kept every unsupported claim off the page. For a real product, this matters more than any visual difference: an invented "no card needed" or a list of supported banks can mislead buyers.
• A memorable direction. The departure-board concept turns the product's idea into the page. Taste Skill's page is good; Impeccable's is the one you remember.
• It covers more jobs. Four modes (persuade, operate, read, experience) and 24 commands mean it works on dashboards, settings screens and docs, and on existing interfaces: /impeccable audit and /impeccable critique review what you have without rebuilding it.
• It leaves a design system behind. The DESIGN.md it writes keeps later changes consistent.

#Where Taste Skill wins

• Cost and speed. About a third of the cost and a third of the time for a finished landing page.
• Stricter about generic tells. It bans the em dash outright, rations the middle dot and lists dozens of AI-design patterns to avoid. Its page had none of the template chrome the no-skill page used.
• Restraint. A calm, readable page that suits a finance product and would pass most brand reviews. Impeccable's condensed capitals and split-flap tags are bolder but slower to read.
• Simplicity. One file you can read in an evening and adjust. No binary, no commands to learn.

Its two weak points in our run: it invented product facts like the other skills, and it missed two items from its own pre-flight check. Both are reasons to review its output, not reasons to avoid it.

#Which should you use?

Can you use both? Not on the same page at once: their rules overlap and sometimes disagree, so one direction skill per page works best. A combination that makes sense, though we did not test it here, is building with Taste Skill and then running Impeccable's audit command on the result.

Whichever you choose, put the real product facts in your brief. It was the single biggest difference between a page you could ship and one you would have to rewrite.

#Install both

# Impeccable: detects your agents, then run /impeccable init inside the agent
npx impeccable install

# Taste Skill
npx skills add https://github.com/Leonxlnx/taste-skill --skill "design-taste-frontend"

# Or either one from SkillGild, for Claude Code, Codex, Cursor or Gemini CLI
skillgild install impeccable --agent claude-code
skillgild install design-taste-frontend --agent claude-code

Read the full tests: Impeccable and Taste Skill. For eight more frontend skills, including Anthropic's frontend-design skill (the best all-round default in our test), see the top 10 Claude frontend and design skills.