skills

Pixel-perfect website cloner. Given a URL, faithfully reproduces the page as a single self-contained HTML file. Uses Playwright to render the page in a real browser, then directly extracts the rendered DOM + all computed CSS — no manual rewriting. Handles static HTML, JS-rendered SPAs, Webflow, Next.js, and any framework automatically. Zero creative interpretation — reproduces exactly what exists. Use when asked to "clone this site", "copy this page", "replicate this URL", "pixel-perfect clone", or user provides a URL and says "make it look exactly like this".

Converts cloned website HTML files (from frontend-ui-clone) into production-ready React/Next.js projects. Extracts components, decomposes CSS, sets up routing, and wires navigation so buttons and links actually work. Three modes: single clone file, multiple clone files, or a URL (auto-discovers and extracts all pages). Outputs a runnable Next.js App Router project with `npm run dev` ready to go. Use when asked to "convert to React", "make a Next.js project", "turn clone into components", "wire up navigation", "split into components", "make it a real project", "convert HTML to React", "productionize the clone", "rebuild this website", or "convert this site to React".

Converts negative and vague prompt constraints into specific, actionable positive directives — in-place. Follows a three-phase compiler pipeline inspired by VCC: Scan (locate constraint spans) → Transform (expand each span into a precise directive) → Backfill (substitute spans back; all other text is preserved character-for-character). Handles both explicit negations ("don't X") and vague qualifiers ("appropriate length"). Triggers on: "refine prompt", "positive prompt", "/dont-to-do".

Compares overlapping or duplicate skills to identify the best one for a given use case. Always starts with a functional analysis (feature coverage, tool access, specialization): if a clear winner emerges, recommends it directly with cited evidence. When skills are functionally tied, or partially overlapping with a test task provided, runs both in parallel isolated subagents on the same task — each receives only its own instructions, zero cross-contamination — then presents outputs side-by-side with a synthesized verdict. Use when: "which skill is better", "compare these skills", "find best skill for X", "which firecrawl should I use", "compare frontend skills", "pick the best skill for me". Proactively suggest when the user seems unsure which of several similar skills to use.

High-aesthetic frontend UI generator trained on the design DNA of Linear, Vercel, Stripe, Raycast, Resend, Supabase, Anthropic and 10+ other benchmark products. Analyzes reference URLs or screenshots to extract design DNA (color, typography, spacing, motion), generates pixel-perfect production code, and saves extracted styles as reusable template skills. 26 curated palettes, anti-AI-slop enforced: real design systems, typographic craft, 8px grid discipline. Use when asked to "build UI", "design a page", "clone this style", "make it look like X", or "generate frontend". Proactively suggest when user shares a URL/screenshot for inspiration.

Product-grade icon designer and curator. Two modes: (1) load icon design principles and style guidelines for professional product icons — anti-emoji, real design systems, optical consistency; (2) given a query describing icon usage and context, search existing icon libraries or generate custom SVG icons and save as SVG/PNG. Supports 6 style archetypes (geometric, rounded, duotone, thin, filled, organic) calibrated against Lucide, Phosphor, Heroicons, Tabler, and SF Symbols. Use when asked to "design an icon", "find an icon for X", "generate SVG icon", "icon design principles", or any icon-related design request. NOT for emoji or decorative use.

Intelligent web search skill that autonomously decides whether a user query requires web search. Searches for text content or reference images based on query analysis. Returns clean, formatted results with quality filtering. Triggers on: queries containing specific people, recent events, trending topics, named styles, or requests that reference real-world entities needing current info.

Brand Quill — Rewrites everyday text into the aesthetic voice of iconic brands. 13 built-in styles: MUJI (restraint), Apple (minimal confidence), Aesop (literary sensory), Patagonia (honest activism), Diptyque (Parisian narrative), Snow Peak (nature craft), IKEA (everyday warmth), Nike (action empowerment), Suntory (liquid philosophy), Hobonichi (gentle everyday), Hermès (poetic craft), Rolls-Royce (facts as poetry), Guinness (patience and substance). Supports mood control (positive/negative/neutral) to shape emotional direction. Supports ZH/EN/JA multilingual output. Specify scene: social media, Vlog, product, bio, slogan. Use when asked to "rewrite copy", "write in X style", "文案改写", "风格文案", "帮我写个无印风的文案", "make this sound like Apple", or any copywriting style request.

Generate structured Product Requirements Documents (PRDs) via guided conversation. Defaults to a lean MVP PRD. Use --full for the complete version with additional depth. Collects context in up to 2 rounds of Q&A, drafts each section for user confirmation, then assembles the final PRD. Supports export in Markdown, Confluence wiki, Feishu/Lark, and HTML formats. Triggers on: "create a PRD", "write PRD", "write requirements", "product requirements document".

Turn your git commit messages into poetry. Reads the diff, auto-selects a poem style based on change complexity, and commits with a poetic message. Supports haiku, modern verse, Shakespearean sonnet, and opus (epic formal verse). Triggers on: "commit poet", "poetic commit", "poem commit", "write a poem for my commit".