Mouse Trail Masking Reveal
Reusable Agent Skills for building soft mouse/cursor-trail masking reveal effects on websites.
This repository packages the Mouse Trail Masking Reveal workflow as two portable skills:
image-layer-alignment-validator: checks whether two raster layers are spatially compatible before using them in reveal, before/after, morph, or mask-compositing effects.cursor-reveal-hero: builds, repairs, tunes, and validates a canvas-based soft cursor-trail reveal hero effect.
The skills are intentionally provider-neutral. They use the open SKILL.md Agent Skills pattern and include provider adapter notes for Codex, Claude Code, Gemini CLI, Kimi Code CLI, Qwen Code, GLM-based coding agents, and generic LLM coding assistants.
Demo
Demo by @sergekost:

Watch on YouTube: youtu.be/BppodCPasDc.
Full-quality MP4: media/sergekost-cursor-reveal-demo.mp4.
Why This Exists
The idea came after seeing a manual Figma workflow by @thedesignely. That post demonstrated the design direction manually in Figma. This package turns the same kind of visual workflow into reusable LLM-agent skills for website work: first validate image-layer alignment, then build the interactive cursor reveal implementation.
Repository Layout
mouse-trail-masking-reveal/
├── skills/
│ ├── image-layer-alignment-validator/
│ └── cursor-reveal-hero/
├── adapters/
│ ├── codex/
│ ├── claude-code/
│ ├── gemini-cli/
│ ├── kimi-code-cli/
│ ├── qwen-code/
│ ├── glm-agents/
│ └── generic-agent/
├── docs/
├── media/
└── scripts/
Quick Start
Install the two canonical skills into the location your agent scans:
mkdir -p .agents/skills
cp -R skills/image-layer-alignment-validator .agents/skills/
cp -R skills/cursor-reveal-hero .agents/skills/
Then ask your coding agent:
Use $image-layer-alignment-validator to compare my base and reveal images.
Then use $cursor-reveal-hero to build the interactive cursor trail reveal hero.
Demo Media
media/sergekost-cursor-reveal-demo.mp4: demo by @sergekost.media/sergekost-cursor-reveal-demo-preview.gif: README preview by @sergekost.- YouTube demo: youtu.be/BppodCPasDc.
media/thedesignely-figma-manual-reference.mp4: reference video attributed to @thedesignely.
See docs/media-and-attribution.md before redistributing the videos.
Provider Support
See docs/provider-matrix.md and the adapters/ folder.
The short version:
- Codex: use
.agents/skills/or a user/admin skill location. - Claude Code: use filesystem-based custom skills.
- Gemini CLI: use
.gemini/skills/or.agents/skills/. - Kimi Code CLI: use
.kimi/skills/,.agents/skills/, or another configured skills directory. - Qwen Code: use
.qwen/skills/or project/user skill locations. - GLM agents: use the Agent Skills folder in the host agent, such as Claude Code, OpenCode, OpenClaw, AutoClaw, or another GLM-powered coding shell.
- Generic agents: paste or mount the
SKILL.mdfolders as task-specific instructions with scripts/resources available.
Requirements
For image-layer-alignment-validator:
- Python 3
- Pillow
- numpy
- opencv-python
For cursor-reveal-hero smoke checks:
- Node.js
- Playwright
The skills themselves are Markdown-first and can still guide an agent without these dependencies; the scripts need them for deterministic checks.
Privacy
This package contains no API keys, tokens, credentials, or machine-specific paths. Scripts run locally by default. See PRIVACY.md and SECURITY.md.
License
Code and skill text are released under the MIT License. Third-party media may have separate rights; see docs/media-and-attribution.md.
No comments yet
Be the first to share your take.