Poc
Build fast, clickable app mockups (mobile/web), prototypes and visualizations to validate an idea or demo a concept to a client — including live, during a conversation. Use THIS skill whenever the user wants to: build a clickable app mockup, show how something could look or work, visualize an idea for a client, or make a quick mockup. Trigger on phrases like "aplikacja mobilna do X", "apka do Y", "zbuduj mock", "pokaż jak to mogłoby wyglądać", "zwizualizuj pomysł", "szybki mockup dla klienta", "makieta aplikacji", "wireframe ekranu", "mobile app for X", "app for Y", "wireframe a screen", "build a mock", "show how this could look", "prototype for the client", "quick mockup", "zróbmy PoC", "proof of concept", "quick prototype" — even if the user never says the word "PoC". Especially: when the user describes an app and wants to see or click through it, this is the skill. It is for THROWAWAY / validation artifacts and concept demos, not for building the real production product (use the `kickstart` skill for that).From its SKILL.md
npx -y skills add n02b3rt/Agent-Build-Kit --skill pocAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 19 days oldThe repository was created 19 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
7.2 KB, ~1.6k tokens by cl100k_base, as published. Nobody here has run it
PoC — fast clickable mockups for a PM
Turn an idea into something tangible, live in a conversation. A PoC is not a product — mock the data, cut corners, skip edge cases. One thing matters: does it show ONE thesis well enough to get a "wow". Working-and-now beats perfect-in-a-week.
(Technical feasibility or building it for real → not this skill, use kickstart.)
Live fast path — SPEED FIRST (this is the default)
On /poc live: <idea> the client is often watching. Get to a published EMPTY mock in seconds, then add features on request. Do exactly this, nothing more:
- Pick the shell from the idea: mobile app →
assets/mobile-shell.html; website / web app / responsive →assets/web-shell.html. - Copy it to a working file. For the empty scaffold you only change
<title>, the brand name, and one--accent(domain hints below) — nothing else. - Publish via
Artifact. Reply with the link in one line + "what do we build first?". Keep thefile_path+ URL.
At init, do NOT (this is what makes it slow):
- Do NOT read
mock-canvas.md,design-styles.md,domains.md, or loadartifact-design/dataviz. Everything you need is inline in this file. Reading them first just makes the client wait. - Do NOT write an analysis — no "Thesis / Surface / Direction / Assumptions" essay. Skip it entirely.
- Do NOT deliberate over the app name or ask about language. Default silently: name from the idea, content language = the language of the user's request (Polish for PL clients). At most one short line if you assumed something.
- Do NOT invent features. The first result is the empty, polished shell. Full stop.
The whole init = pick → copy → set title/accent → publish → one-line link + "what first?". Apply the anti-slop and golden rules below silently — don't narrate them.
Phase 2 — features: each user message = one change. Edit the same file, republish to the same file_path (same URL), no need to repeat /poc. The shell's header comment shows how to add screens/sections — glance at that if needed, not the reference files.
Running on an existing app / repo? Different job: read the app's current code and build an HTML replica of its key screens (so the client sees "what we have today"), then iterate. That's the one case where you read existing code before building.
Always report state — one line, so the user knows what's happening
No walls of text. After every step, a single compact status line:
- After init:
🟢 Canvas live · <link> · empty — tell me what to add (e.g. "add the home screen"). - After each feature:
✅ Added: <screen/section> · same link · now has: <comma-list of screens>. - If you had to assume something (name, language, accent): one short clause, e.g.
(assumed English + name "Ostrzyż" — say to change). The user must always see three things: what's live, the link, and the next move. Nothing else.
Anti-slop — inline, non-negotiable
The shell is already well-designed; keep it that way. Flat by default, one accent from the domain, real data (never placeholders). Banned: gradient heroes/buttons, glow/blobs behind content, decorative dots or ✨, everything-centered, emoji as section headers, glassmorphism, a shadow under everything, cream #F4F1EA+serif+terracotta.
Icons: inline SVG line-icons ONLY — NEVER emoji. Emoji as UI icons read as a kids' app. The mobile shell ships an SVG sprite (#i-calendar, #i-user, #i-scissors, #i-chevron, …) — use <svg viewBox="0 0 24 24"><use href="#i-…"/></svg>; need a new icon → add a <symbol> to the sprite (24×24, stroke, fill:none). When a control is a <button>, keep border:none so it doesn't render browser box-borders.
Proportions — baked into the shells' tokens, use them (don't eyeball): 8pt spacing --s1..--s8 = 4/8/12/16/24/32/48/64 → card padding 16, section gaps 24–32, ≥8px between taps. Touch targets ≥44px. Body ≥16px, line-height 1.4–1.5; headings 1.5–2× body via --t-*. Web body text max-width:66ch. Full detail: references/design-fundamentals.md.
Accent quick-pick: health → calm blue/teal · fintech → navy or money-green · fitness → coral (often dark) · productivity/B2B → indigo/blue · education → warm friendly · dev/tech → mono / high-contrast.
Golden rules
- Build only what was asked. Zero features ahead of time.
- "App" = show the product UI, never a marketing hero/landing/pricing (unless explicitly asked).
- Artifact = the app only — no side panels or "assumptions" in the mock; assumptions go to the user in chat.
- Real data from the client's domain; mock content in the client's/product's language (Polish for PL clients).
- Same link all session — republish the same file.
- Don't gold-plate the shell — spend energy on features.
- Assemble from the shell's widgets — both shells ship ready blocks (KPI tiles, avatar, chips, toggle, segmented, input, progress, mini bar-chart, callout, table…). Use them instead of hand-writing CSS; the full list is in the shell's header comment. Faster and consistent.
- Web shell: responsive CSS uses
@container, not@media.
Need more? (optional — read on demand, NOT before the first mock)
Reach for these only when a specific need comes up — never as a warm-up:
references/mock-canvas.md— deep guide: incremental command vocabulary, building blocks, web@containerdetails. Read when the shell's header comment isn't enough.references/design-styles.md— full direction library. Read only for an elaborate/branded look beyond the flat default.references/design-fundamentals.md— proportions cheat-sheet (8pt grid, touch targets, type scale, line length). The essentials are already inline above + baked into tokens; read for depth.references/domains.md— per-domain "what to build / data / pitfalls".artifact-designskill — only for a bespoke visual beyond the shell.dataviz— only when adding a chart.web-artifacts-builder— complex multi-view React.kickstart— when the idea is validated and it's time to build for real.
What ships with it: 6 files
56.6 KB alongside SKILL.md
assets/
- mobile-shell.html21.5 KB
- web-shell.html13.2 KB
references/
- design-fundamentals.md2.5 KB
- design-styles.md8.0 KB
- domains.md3.7 KB
- mock-canvas.md7.6 KB