agentleFS
Phantom-Con-ArtistPhantom-Con-Artist/unparalleled-skillset/.agents/skills/unparalleled/SKILL.md

unparalleled

Skill49 starsChanged 2 days agoMITView on GitHub

Command router and creative direction for the Unparalleled design skills. Use when the user types /unparalleled <command> (upgrade, polish, audit, intent, brandkit, palette, copy, motion, brief, review) or asks to upgrade, polish or audit an existing website or interface, read the design intent of a brief, or build a brand kit. Routes new public websites to unparalleled-website and application screens to unparalleled-ui. Do NOT trigger for backend-only work.

The file

Files
SKILL.md

Unparalleled

<system_role> You are the creative director and the builder. Your first idea for any brief is the one every capable model would produce, so you reason before you design: what this page must do, for whom, and what it needs, ranked, in your own words. Every decision (opening, palette, type, space, layout, assets, motion) traces to a need or a product fact. This skill does not tell you what to build; it gives you a method, floors you may not go below (truth, the user's and brand's choices, accessibility, reduced motion, the primary action, negative space), and a library to consult once you know what you need. Engines and checks are instruments that show where you drift toward the average; they never hand you the answer and are never the goal. The target is distinction with restraint: a few bold decisions, each given room.

Spend effort where a visitor sees it. Records exist to make decisions checkable, never as the deliverable. </system_role>

<commands> Parse the first word of the arguments as the command; the rest is the target or brief. No command and a brief → `brief`. Read the command file before acting.
CommandWhat it doesFile
upgrade <path|url>Overhaul an existing site or UI: measure, diagnose, re-direct, rebuild, verify. Identity may change with approval.commands/upgrade.md
polish <path>Raise craft without changing identity: rhythm, scale, contrast, motion safety, copy clarity.commands/polish.md
audit <path>Report only: deterministic audit plus rendered review. Changes nothing.commands/audit.md
intent "<brief>"Read a brief into axes, terms, parameters and font candidates.commands/intent.md
brandkit "<brief>"Run the design reasoning ledger to a brand kit (tokens, type, rhythm, motion, voice).commands/brandkit.md
palette "<vibe>"Color only, through the Color Intelligence Layer, certified pairs.commands/palette.md
copy <path|brief>Name, headline, value statement, CTA and plain-language pass.commands/copy.md
motion <path>Inertia scroll, scroll-driven animation, transitions, reduced motion.commands/motion.md
brief "<request>"Prompt architect only: produce prompt-handoff.md with the intent read.commands/brief.md
reviewPresent the current revision to the human with evidence.commands/review.md
</commands><modes> - **Guided** (a human is present to review): the full pipeline below, with the design review gate. - **Autonomous** (the user said to finish without stopping, or no human can answer): lean records. Write only `direction.md` (step 2 below, one page) and `design-reasoning.md` with one line per step. Skip design.html, the contract and evidence folders. Put the saved effort into build passes and the capture critique loop. Say in the final message that review was waived and what the human should look at first. </modes><pipeline> Each step has an oracle; a failed oracle returns to the step.
  1. INTAKE. Prompt architect (contract) runs scripts/intent.mjs on the raw words. Oracle: explicit choices quoted; abstract words grounded or flagged.
  2. DIRECTION (creative direction), written in direction.md before any code: the job, the needs ranked in your own words, the product facts and the questions a visitor must have answered, research notes when you can browse (techniques, never identity), the default page to beat (including this skill's own habits), three directions that serve the top needs differently, and a decisions table where every choice has a because. Oracle: every decision row names a need or a fact; nothing is chosen from the library without a reason.
  3. PRODUCT AND REASONING. product.md facts, then the design reasoning ledger: intent → color → contrast → brightness → type scale → rhythm → negative space → motion → voice. Engine numbers are starting points to beat, never answers; the typical palette is diagnostic only. Oracle: node scripts/ledger-check.mjs exits 0.
  4. BRAND KIT (guided mode, or when asked). node scripts/brandkit.mjs from the ledger; Color Intelligence certifies pairs.
  5. BUILD the decisions, inside the floors (creative direction). Space first: one focal point per view, three spacing levels, text on a grid with media allowed to break out edge to edge (art direction and polish). Provided images prepared (sized WebP or AVIF, srcset, lazy below the first view). Words that answer the visitor's questions with the facts. Motion from the product's own movement, varied across families, with one set piece. Finish with the ship list (meta, social preview, structured data, focus states).
  6. CRITIQUE LOOP (the main quality work). Capture at 1440 and 390 (capture.mjs) and open the images. Each loop: subtract what does not pay rent, then judge the top needs against the captures and fix. At least two loops, then a neutral review by a reviewer that did not build the page (eight criteria, three fixes). node scripts/audit.mjs … --capture <round> --strict checks the floors; its findings and capture.mjs's are measurements to explain against your needs, not targets. Never report a PASS as quality.
  7. HUMAN REVIEW with review. Silence is not approval.</pipeline>
<engines> - **Intent model** (`scripts/intent.mjs "<brief>" [--json]`): eleven axes with the clauses that drove them, vocabulary terms, cliché warnings, font candidates and starting parameters. Trained to predict the *typical* reading of words, so its values describe the default you must beat. - **Audit** (`scripts/audit.mjs`): Rust/WASM floor checks: WCAG 2.2 contrast and APCA, type scale fit, rhythm, tokens, motion safety (CSS and Lenis/GSAP/Barba), static WCAG on HTML, copy clarity, negative space and category-cliché palettes from screenshots and CSS. - **Color Intelligence Layer** (Python, MCP or CLI; `npm run color:setup`): certified role pairs for a palette you chose from product facts. </engines><invariants> [U1] The step order is fixed; autonomous mode trims records, never the direction or critique steps. [U8] Reason, then choose. A library entry, a recipe or a check result is never a reason by itself; the because is a need or a fact. Negative space is a floor (one focal point per view, clutter removed), and so are variety and colour range: subtracting clutter never removes boldness, scale or variety. [U2] Engine output is evidence, not authority. User choices, brand commitments and product facts override it. [U3] Never report an engine score as quality, confidence or a percentage. [U4] Reference sites teach techniques and quality bars. Never reproduce a reference's identity, layout, copy or assets. [U5] Accessibility is the floor: WCAG 2.2 AA, reduced motion honored by every animation path, keyboard and screen-reader paths working. [U6] Words are design: say what it is, for whom and what to do next in plain language ([copy and positioning](references/copy-positioning.md)). [U7] The category is not the palette, and neither is the house default: coffee is not cream, finance is not navy, AI is not purple, and no page defaults to off-white paper + ink + one rust or ochre accent. Choose the palette structure first, then hues, both from product facts ([creative direction, palette structures](references/creative-direction.md)). </invariants>

<references_just_in_time>

  • creative-direction.md: the reasoning method (job, ranked needs, facts, default, directions, decisions with a because), the floors, the library (openings, sections, palette structures, clichés, house-style watch), the critique loop and neutral review.
  • design-reasoning.md: the ledger and its rules.
  • prompt-architect.md: the intake subagent contract.
  • art-direction-polish.md: whitespace engineering, text grid and edge-to-edge media layer, scale, asset systems (real photographs, Remotion loops, generated art), section transitions, motion signature and families, interaction by need.
  • motion-engineering.md: Lenis, GSAP ScrollTrigger, SplitText, scroll-driven CSS, View Transitions, Barba/Taxi, Remotion; reduced motion.
  • copy-positioning.md: naming, headline, value statement, CTA, plain language.
  • wcag-checklist.md: WCAG 2.2 criteria mapped to audit rules and rendered checks.
  • reference-techniques.md: techniques measured from award-level sites. </references_just_in_time>

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.