Something people experience
The exhibit you see in a museum or at an event, where video and sound react to how people move. Projection mapping, installations, stage visuals, experience-driven apps.
The goal is to make a visitor feel something.
interactive‑
experience
Camera, sensors, human movement — decide what you are building first.
Something people experience, or a tool that makes people better at a movement? They need different technology, different users, different buyers and different definitions of success. These skills settle that first — then design the right one.
>/plugin marketplace add takaoumehara/interactive-experience-skills
01Two kinds
Two products that look alike and are not
The most common failure in this field is arguing about TouchDesigner versus Unreal, or about pose-estimation accuracy, before deciding which of the two you are making.
A tool that makes people better
For things where form matters — dance, martial arts, sports, yoga. It films you, evaluates what you did, tells you what to fix and designs your practice.
The goal is that the person actually improves.
02Architecture
One director, two specialists
The two specialists do the design work. The director on top only decides which way you go. They are deliberately not split further: every extra description in the system prompt makes the hardest call — experience or improvement — worse.
- 00
embodied-product-directorDecides which of the two you are making, with the reason. Appears only when you have not decided yet.
- ①
interactive-experience-collectiveExhibits, installations, stage visuals and experience apps: projection, latency, audio, venue economics.
- ②
movement-learning-system-designerForm evaluation, practice design and tools that help instructors — who uses it, who pays, what is actually going wrong today.
Skill instructions are in English; the reference library is still in Japanese (translation in progress). Claude reads both and answers in your language. Japanese trigger phrases are kept, so Japanese prompts start the skills too.
03Install
Two lines, then just describe it
The plugin installs all three skills. Open a new session and write what you are building normally — the right skill starts on its own.
claude code
/plugin marketplace add takaoumehara/interactive-experience-skills
/plugin install interactive-experience-skills@interactive-experienceRecommended. Installs the three skills as one plugin.
bash
git clone https://github.com/takaoumehara/interactive-experience-skills.git
cd interactive-experience-skills
./install.shAlso installs the maintainer commands /motion-idea, /refresh-skills, /scout-skills and /skills-routine. Don't install both the plugin and the script copies.
bash
./package.sh # writes dist/<skill>.zipUpload each zip in your assistant's skill settings.
Then try
prompt
Design a projection mapping piece that reacts to a dancerprompt
Design the MVP for an app that compares a karate
punch against the instructor'sprompt · no direction yet
I have two spare webcams and want to build
something around martial arts practice04What you get
A design decision, not code
What comes back is what to build, roughly what equipment and how much of it, what to validate first, and what not to build yet. Implementation is a normal Claude Code job afterwards.
It always commits to one
Never “well, it could be either.” It picks experience or training tool, and gives the reason.
Numbers, not “immersive”
5,000–8,000 lm for 3 m in a dark room. Feedback within 100 ms. A side camera for step depth. About 107,000 characters across 18 references, loaded only when needed.
It says no clearly
No judging pain or injury. Silence when unsure. Guardian consent before technology for children. And no belief that better pose accuracy means better learning.
Before / After
| Before | After | |
|---|---|---|
| How it starts | “We have two cameras — what can we build?” | “Which problem makes a second camera worth its price?” |
| Artwork or tool | Never settled; implementation starts anyway | Settled first, with the reason |
| Technical answer | “An immersive AI-powered experience” | 5,000–8,000 lm · 100 ms · one camera is enough |
| Building alone | A degraded version of the real thing | Possibly the final and best form |
| Pose estimation | “More accuracy means people learn more” | Accuracy and learning are unrelated; redesign for learning |
05FAQ
Frequently asked questions
Does it write code?
No. It returns design decisions: what to build, what equipment, what to validate first and what not to build yet. Ask Claude Code to implement afterwards.
Do I have to call the director myself?
No. Describe what you are building and the right skill starts. The director only appears when you have not decided between an experience and a training tool.
Can I use it in Japanese?
Yes. Each skill keeps its Japanese trigger phrases, and the reference library is written in Japanese. Claude answers in the language you use.
Won't the prices and hardware numbers go stale?
Yes, so volatile passages are marked. When the skill reads one, it checks the web before answering.
/refresh-skills batch-verifies them and /scout-skills proposes new candidates without editing references.Will it assess injuries or rehabilitation?
No. Pain, injury and range of motion are medicine. It will not step into rehabilitation without a clinician involved.
How accurate is the routing between the three skills?
An earlier check was reported as 97% with 11% ambiguous cases, but its method, model, date and raw results were never committed, so treat it as to be re-measured. Each skill ships
evals/evals.json with near-miss cases for re-checking.