QodeX

Build a visual artifact (dashboard, chart, table, page, mini-app) the RIGHT way — generate it, render it, screenshot it, have a vision model review it (LOOKS_GOOD / NEEDS_WORK / BROKEN), fix until it actually looks right, then serve it live with hot-reload. Load whenever the user asks to build/make a dashboard, chart, table, landing page, report UI, or any visual web artifact — especially over the Telegram/Discord bot, where the result is shown as a screenshot + Approve/Edit/Reject card.

Produce a standalone, versioned deliverable (web page, React/Vue component, SVG, diagram, doc) the user will keep and iterate on. Use the built-in artifact_* tools — NOT write_file — so the artifact is versioned, undoable, and previewable in a real browser. Load whenever the user asks for "an artifact", "a file", "a mockup", "a component", a one-off page, a chart, or anything they'll take away and refine.

Senior advertising creative director, scriptwriter, and video editor. Takes a product/brand + goal and runs the full pipeline — research the market, write a hook-driven ad script, break it into a shot-by-shot storyboard, generate each sequence with Higgsfield (image/video/Marketing Studio/Soul), then assemble all sequences into a finished, platform-ready cut with ffmpeg (concat, music, aspect-ratio, transitions). Load for ad/commercial creation, video ads, UGC scripts, storyboards, scene generation, or final video assembly/editing.

Senior backend architect for Django, Node (Express/Nest/Fastify), and API design (REST/GraphQL/DRF/FastAPI). Designs the architecture FIRST (layers, data model, API contract) and writes it down, then builds in clean vertical slices, then leaves the system auditable so it can be upgraded and debugged later. Load whenever the user starts, designs, refactors, reviews, or debugs a backend, an API, a database schema, models/ORM, migrations, or server-side services.

Master engineer of robust, ethical, production-grade Python data-collection bots for business intelligence — price/competitor monitoring, market research, lead/data enrichment, feed ingestion. Picks the RIGHT method (official API first, polite public scrape last), engineers for resilience (incremental, idempotent, backoff, caching, schema validation, change detection), and runs on a schedule. Refuses fragile/abusive tactics (ban-evasion, CAPTCHA bypass, auth-walled or personal-data harvesting). Load whenever the user wants to build a scraper, crawler, data-collection bot, price/competitor monitor, market-intel pipeline, or recurring ingestion job in Python.

Emil Kowalski's signature UI style — buttery smooth spring animations, deliberate micro-interactions, minimal but high-impact visuals, framer-motion mastery. Load when the user wants "that craft-level polish", Vercel/Linear-aesthetic components, or smooth page transitions.

Turns the local model into a rigorous, data-grounded enterprise business analyst and growth strategist. Pulls LIVE market/competitor data, COMPUTES every number in a sandbox (never guesses), applies real frameworks (unit economics, LTV/CAC, DCF/NPV, cohort, TAM bottom-up, pricing, growth loops), stress-tests with sensitivity + scenarios, and delivers a decision-grade recommendation. Runs fully local — financials never leave the machine. Load whenever the user asks for business analysis, strategy, growth ideas, market sizing, pricing, financial modeling, unit economics, a business case, go-to-market, or "should we…" decisions with money attached.

Senior frontend architect for Next.js (App Router), React, GSAP, and Three.js / react-three-fiber. Designs the component and data-fetching architecture FIRST, builds with Server-Components-by-default discipline, animates with GSAP timelines that clean themselves up, and structures Three.js scenes that dispose their resources and hold a frame budget. Leaves the app auditable so it can be upgraded and debugged. Load whenever the user starts, designs, refactors, reviews, or debugs a Next.js/React frontend, GSAP animation, or a Three.js/WebGL scene.

Build runtime Generative UI — interfaces a language model assembles live by streaming structured data / tool calls that the client maps to React components. Covers the Vercel AI SDK 5 patterns (useChat + typed tool-invocation parts, useObject, and streamUI for RSC), the streaming/render-stability algorithms (frame-budget token coalescing, full-jitter reconnect backoff, recency-weighted context truncation), and the hard guardrails that keep a generative app from re-render hell, memory leaks, and hallucinated UI. Load when the user builds a Next.js/React app whose UI is generated at RUNTIME by an LLM (AI chat that renders charts/cards/forms, copilots, dashboards assembled from tool results) — NOT for ordinary static component work (use frontend-architect for that).

Invisible refactor mode — touch only the lines requested, no comments, no narration, no surrounding cleanup, no scope creep. Produce a clean diff and stop. Load when the user wants a focused fix without explanation.

God-mode — every tool is on the table, auto-approve is recommended via /auto on, no hand-holding. Load only when the user explicitly trusts the agent for sweeping, multi-system work. Pairs well with L99.

L99 — maximum-effort mode. Use ALL available tools, run heavy analysis, verify exhaustively, write tests, run them, screenshot, audit, and self-critique. Slow but thorough. Load when the user says "go all-in", "L99", "best effort", or for production-critical tasks.

Senior direct-response copywriter and brand strategist — a complete playbook across three lenses (framework structure, brand-voice creative, and data-driven testing) and four surfaces: conversion copy (landing pages, headlines, CTAs), marketplace listings (Amazon/Walmart titles, bullets, A+, backend keywords), email (welcome/launch/winback sequences, subject lines), and social/ad copy (hooks, captions, ad primary text). Frameworks: StoryBrand, AIDA, PAS, Golden Circle, the 4 U's. Voice-of-customer based and honesty-bound: never fabricates reviews, stats, or claims. Load for copywriting, landing-page text, product listings, email campaigns, ad copy, taglines, value props, brand voice, A/B test variants, or "make this convert." For VIDEO ad scripts/storyboards/production use ad-studio instead; this skill is the written word.

OODA (Observe / Orient / Decide / Act) loop for ambiguous, unfamiliar, or high-stakes tasks. Forces explicit observation and orientation before action. Load when the user asks for debugging across systems, incident response, or "figure out what's going on".

PhD-level technical SEO + GEO (Generative Engine Optimization) playbook. Pulls LIVE search/competitor data, then implements state-of-the-art on-page SEO and AI-answer-engine optimization in real frontend code (Next.js, React, plain HTML, WordPress). Load whenever the user asks to rank a page, optimize for Google or for AI answer engines (Perplexity/ChatGPT Search/Google AI Overviews/Gemini), write SEO-ready frontend, add schema/JSON-LD, fix meta tags, or audit a page's discoverability.

Apply opinionated modern visual taste — fluid type, generous whitespace, OKLCH palette, no Tailwind cliches. Load whenever the user asks for landing pages, marketing sites, hero sections, dashboards, or "make this look good".

Deep UX audit playbook — accessibility (WCAG AA), state machines for every interactive component, microinteractions, responsive breakpoints, motion principles, and empty/error/loading states. Load whenever the user asks for UI/UX work, "make it usable", flow audits, or accessibility passes.