Imported from Vandanp14/summer_portfolio (
AGENTS.md). Install upstream withnpx skills add Vandanp14/summer_portfolio. Copyright stays with the author.
Imported skills (vandan-llm-toolkit)
No dynamic skill loader here — read the matching file BEFORE starting work that matches its trigger:
docs/llm-guides/skills/vandan-ui-system.md— Vandan's unified UI design system — the single source of truth that supersedes ultra-clean-native-ui, mobile-native-clean-components, premium-software-brand-system, and premium-dark-ui-taste. Use for ANY frontend or UI work: creating, redesigning, polishing, or critiquing web apps, dashboards, mobile-style web apps, finance/fitness/habit trackers, AI tools, developer tools, portfolios, landing pages, internal tools, React/Next.js/Tailwind components, or full design systems. Make sure to use this skill whenever the user mentions UI, design, components, screens, pages, styling, frontend, dark mode, or wants anything to look premium, clean, native, sleek, or 'not AI-slop' — even if they never say the word 'design'.docs/llm-guides/skills/app-transformer.md— Deterministic pipeline that transforms any mid-tier, generic, template-looking, or vibe-coded app into a super-premium product — regardless of which LLM runs it. Use whenever the user says 'transform', 'upgrade', 'make it premium', 'make it look expensive', 'redesign', 'polish this app', 'it looks AI-generated', 'make it high tier', or wants a full-app quality pass on a web app, dashboard, SaaS, landing page, mobile-style web app, internal tool, or portfolio. Also use when starting a new app that must be top-tier from day one. Works phase-by-phase with copy-paste values so even lower-capability models produce the same result.docs/llm-guides/skills/hunting-hidden-root-causes.md— Use when a symptom is reported using the vocabulary of one component ("payload received," "deploy shipped," "config updated," "access revoked") but a downstream system doesn't reflect it — any "A happened, B didn't follow" bug, "works here but not there," or a bug that survives repeated debugging sessions where the same fix keeps getting proposed and doesn't stick. Domain-agnostic: web apps, CI/CD, message queues, config distribution, permissions, data pipelines — anywhere a write/event must propagate through independent systems.docs/llm-guides/skills/motion-craft.md— Canonical motion system for premium web interfaces — the full token set (durations, named cubic-bezier easings, spring configs, distance and stagger scales), composition laws, a copy-paste recipe catalog, and the hard performance and accessibility gates. Use for any animation, transition, or motion work: easing curves, cubic-beziers, springs, scroll reveals, page-load choreography, staggered entrances, word-stagger headlines, infinite marquees, count-up numerals, skeleton shimmer, conic-gradient borders, spotlight hover cards, scroll-scrub sequences, view-transition morphs, squash-stretch buttons, confetti and celebration sequences. Use whenever you touch prefers-reduced-motion, 60fps, jank, parallax, scroll-jacking, or micro-interactions.docs/llm-guides/skills/daily-game-ux.md— Daily-games product UX system — the patterns that make a page of games read as a daily-ritual destination instead of an e-commerce product grid: stateful hub tiles (Play/Continue/Solved), playable-first ordering, edition stamps, next-edition countdowns, per-game streaks, post-game stats modals, spoiler-free share artifacts, pre-game splashes, how-to-play first-run modals, completion memorialization, and anticipation-building coming-soon stubs. Use for daily-puzzle hubs, game landing pages, Wordle / NYT-Games / Puzzmo / LinkedIn-Games-style products, streak UI, share grids, finish moments, notify-me stubs. Use whenever building or critiquing any game tile, ritual chrome, share artifact, or daily-refresh loop.docs/llm-guides/skills/collectible-card-ui.md— Trading-card and collectible-product UI toolkit — the holo/foil/tilt CSS effects pipeline plus the collectibles-platform UX language (rarity tiers, card anatomy, collector vocabulary). Use for any card-as-object interface: holo cards, foil effects, holographic or iridescent shine, refractor/prizm looks, pointer-tracked tilt, 3D card hover, sparkle/glare/sheen, rarity tiers and tier badges, graded slabs, pack rips, pull reveals, trading cards, sports/Pokemon-style cards, NFT or digital collectibles, or any product styled like a physical card. Use whenever the words card effect, holographic, foil, tilt, rarity, grail, chase, slab, or pull appear.docs/llm-guides/skills/design-research-pipeline.md— Deterministic research-and-brief pipeline — the front half that turns a vague 'build me a premium X' into a researched, briefed, auditable build plan. Use when the user asks to research competitors, run a competitive or genre teardown, do a brand audit, build a differentiation map, pick a palette, write a design brief, decide a design direction, or plan before building; and later to run a craft audit, UX audit, or specialist review of a shipped site. Also use whenever someone says 'build me a premium / top-tier / expensive-looking app' but no brief exists yet.docs/llm-guides/skills/animated-backgrounds.md— Replication-grade playbook for the ambient/scene layer of a page — animated hero and full-bleed backgrounds spanning CSS aurora and mesh gradients, canvas particle and image-sequence scenes, WebGL fragment-shader and three.js heroes, and scroll-driven cinematics — each with copy-paste code, a cost-tiered decision table, and mandatory reduced-motion and capability fallbacks. Use when building an animated background, hero animation, WebGL gradient, mesh gradient, aurora, particles, scroll-driven scene, scroll-scrub, campaign or landing page motion, video background, three.js hero, or shader effect.docs/llm-guides/skills/writing-weak-model-prompts.md— Prompt-and-harness engineering for driving weak or mid-tier LLMs (Gemini 2.5, Sonnet 4.6, Haiku, Llama, Mistral class) over an API in production apps to produce reliable structured output. Use when a weak model wraps JSON in prose or a fenced block, invents or drops schema fields, drifts format across a batch, or appends commentary despite "output only the object." Use whenever you build extract, classify, or rewrite calls that a strict parser consumes, need schema-constrained output, tool-forcing, retry-on-invalid-parse loops, validate-then-repair, few-shot exemplar selection, or delimiter and temperature discipline.docs/llm-guides/skills/version-grounded-coding.md— Version-grounded discipline for writing library, framework, and tooling code against the exact installed release instead of training memory. Use for any coding task that imports a dependency, calls a framework API, configures a build tool, or writes a config schema. Use when a repo pins versions in a lockfile or manifest, when AGENTS.md warns the framework differs from your training data, or whenever you are about to write an import, API call, CLI flag, or config key. Use whenever a symptom looks like a wrong-major API, removed export, renamed method, or ignored deprecation.docs/llm-guides/skills/secure-by-default.md— Security and deployment conventions applied while writing code, not audited after — because an LLM's default output mirrors the insecure examples that dominate its training corpus. Use for any feature touching auth, login, password, session, JWT, an API endpoint, form input, file upload, SQL or a database query, secrets, an API key, or a .env file. Use when configuring CORS, and whenever you deploy, ship to production, run a security review, or respond to a penetration test or vulnerability report.docs/llm-guides/skills/integration-glue.md— Seam-discipline for gluing systems together at enterprise grade — the reliability contract for every boundary an LLM writes. Use for API integration, REST endpoint, or webhook work; use when building an MCP server or MCP tool, a database transaction, or event/queue publication; use whenever code makes an outbound call to a third-party API or SDK, a microservice, or crosses any service boundary; use whenever the words retry, timeout, idempotency, rate limit, backoff, circuit breaker, glue code, or service integration appear. Weak models ship happy-path seams; this makes them durable.