Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/yousuf-developer/liquid-glass-carousel/agents-mdgit clone --depth 1 https://github.com/Yousuf-developer/liquid-glass-carouselWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/instructions/yousuf-developer/liquid-glass-carousel/agents-md)<a href="https://agentmods.dev/instructions/yousuf-developer/liquid-glass-carousel/agents-md"><img src="https://agentmods.dev/badge/instructions/yousuf-developer/liquid-glass-carousel/agents-md.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.01246 | $0.01246 |
| Opus 5 | $0.00623 | $0.00623 |
| Sonnet 5 | $0.00249 | $0.00249 |
| Haiku 4.5 | $0.00125 | $0.00125 |
Grade A, and why
liquid-glass-carousel AGENTS.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 5d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 50 lines — stays where its author put it; the contents beside it link to each section on GitHub.
This is NOT the Next.js you know
This version has breaking changes — APIs, conventions, and file structure may all differ from your training data. Read the relevant guide in node_modules/next/dist/docs/ before writing any code. Heed deprecation notices.
Project notes for AI assistants
WebGL portfolio carousel: three.js + GSAP core with a thin React/Next.js wrapper. Human docs: README.md (overview) and HOW-IT-WORKS.md (architecture walkthrough) — read the latter before changing engine behavior.
File map
| File | Owns |
|---|---|
lib/carousel/config.js |
PROJECTS image list + all tunables (CONFIG, INTERACT, LENS, FOCUS, ENTRY, UI_ANIM). Data only, no logic. |
lib/carousel/engine.js |
Everything on the canvas: renderer, infinite row, scroll, lens shader (inline GLSL), focus mode, entry animation, render loop. Framework-free — no React imports, ever. |
lib/carousel/gui.js |
lil-gui dev panel (hidden by default). Mutates the config objects live. |
Components/CarouselSection.jsx |
React overlay only: heading, counter, "View" cursor label, Close button, viewport gate (<1025px shows a black screen and never boots WebGL). |
app/page.js |
Just renders <CarouselSection />. |
Routing common requests: add/change images → config.js PROJECTS. Scroll feel / snap / panel size → config.js CONFIG. Drag + click-vs-drag thresholds → config.js INTERACT. Lens look → config.js LENS (uniforms mirror it 1:1). Entry/focus choreography → ENTRY / FOCUS. Overlay text or breakpoint → CarouselSection.jsx. Motion/render behavior → engine.js.
Architecture invariants — do not break these
- The engine talks to React only via callbacks (
onActiveChange,onFocusChange,onEntryDone,onModeChange) and the returned handle (closeFocus,replayEntry,refreshLayout,setInteraction,destroy). Don't import React into the engine or reach into engine internals from the component. - GSAP animates plain numbers, never meshes. Timelines tween values in
drop[],pEntry[],growArr[],focusScale,focusState.lensFx;layout()reads them each frame and derives final mesh transforms. Keep new animations in this pattern or motion will fight itself. - One easing system. Scroll motion is a single lerp (
scroll += (target - scroll) * ease), whereeasepicks betweenEASE/SNAP_EASE/TOUCH_EASEper frame. Wheel, drag, momentum and the settle-snap all movetargetonly — nothing may movescrolldirectly or add a second tween on it. (History: layered easing/snap systems here caused jumpy scrolling and were rebuilt twice.) - The settle-snap triggers on idle time (
SNAP_IDLE_MSsince the last input), not on remaining distance or velocity. Distance/velocity gating fired at different moments for fast flicks vs slow scrolls and felt inconsistent — don't reintroduce it. - Cursor state has one owner. All cursor writes go through
updateCursor()(deduped, runs every frame viarefreshHover). It readshoverPanel(raw geometry); the trailing label readsoverPanel, which adds its own gates. Never setel.style.cursordirectly. - A drag must not open focus.
onPointerUpsetssuppressClickpastCLICK_SLOP(TOUCH_CLICK_SLOPfor touch) andonClickconsumes it. Keep that handshake intact when touching either handler. - The FBO is sized in device pixels (
W * dpr), including inonResize. Sizing it in CSS pixels makes everything blurry on retina. - Textures need mipmaps + anisotropy (set in the load callback). Panels render ~80px tall during entry; plain linear filtering looks mushy.
- 1 world unit = 1 px (orthographic camera). All layout math assumes this.
- Input gating: wheel/click are ignored while
focusState.active || entryActive || entrySettled.focusState.activestays true until the close timeline finishes, even thoughonFocusChange(false)fires at close start (so the overlay UI syncs with the returning cards). Preserve that ordering. - Panel indices can be any integer —
centerForIndex/nearestIndexuse an unbounded index (source = idx mod N, loop =floor(idx / N)) so the row can target the nearest copy of a panel across the infinite wrap. Don't clamp them to0..N-1.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 5d ago First seen · 50 lines · 1,246 tokens per session scan A 70255c72bc96
liquid-glass-carousel AGENTS.md is an instructions file published in the GitHub repository Yousuf-developer/liquid-glass-carousel (144 stars, last pushed 24d ago), licensed MIT. It adds 1,246 tokens to every session, about $0.0062 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other instructions, from other repositories
recharts AGENTS.md
AGENTS.md instructions for recharts/recharts: This is Recharts repository. Recharts is a React-based charting library. The goal is to provide a simple, declarative, and composable way to build charts. We value consistency, usability, and performance. Accessibility is important.
wonder-blocks CLAUDE.md
Claude Code instructions for Khan/wonder-blocks, covering wonder blocks design system, technology stack, coding conventions, file & export naming and typescript.
Canvasight AGENTS.md
Instructions for Niall-Young/Canvasight, covering agents.md, project context, working rules, implementation standards and design standards.
cos-design CLAUDE.md
Claude Code instructions for jiaxiantao/cos-design, covering cos-design — claude code context, for library consumers (user projects), quick reference, when to use and constraints.
Mimikyu CLAUDE.md
Instructions for 3x-haust/Mimikyu, covering mimikyu (따라큐) - figma pixel perfect publisher, 개요, 프로젝트 구조, 기술 스택 and 워크플로우.
screenshot-to-code CLAUDE.md
Instructions for abi/screenshot-to-code, a project described as: Drop in a screenshot and convert it to clean code (HTML/Tailwind/React/Vue).