Factory: Turn Your Factory into a Runnable 3D Discrete-Event Simulation with One Sentence

Install

npx skills add AirsupHQ/factory-skill -g -a claude-code

Factory turns your factory description into a runnable 3D simulation page, but before you press /factory for the first time, confirm three things: you have Claude Code (or any host that loads the Agent Skills format), your machine has Node 20+, and you are willing to let it generate an HTML file. If you need a precise simulation of an existing production line or integration with real MES data, the official evidence does not state these capabilities—evaluate it first as a rapid demo and teaching tool.

Who It's For

Engineers, production planners, or educators who want to quickly turn a production line (e.g., "2,000 e-bike motors a day") into a visual, runnable simulation.
Users of Claude Code who want a single sentence to trigger the full flow of "write the process route → size the stations → headless line run → output a single-file HTML."
Not suitable if: your environment cannot run scripts with Node 20 or above, or your browser does not support WebGL2 (the page will not render). The official documentation does not state which input languages it supports, whether it can import existing CAD/BOM data, or whether the output can be used directly for decision-making—these are all unknowns, so verify them yourself.
Note: this is not just visual decoration—behind the scenes it is a real discrete-event simulation that outputs throughput, takt time, work in progress (WIP), lead time, utilization, and bottleneck displays.

Your First Try in 3 Minutes

Prerequisites: Node 20 or higher installed locally; a WebGL2-capable browser to open the generated page; Claude Code available.
Install: The officially recommended main path is a global install via the skills CLI:
```
npx skills add AirsupHQ/factory-skill -g -a claude-code
```
The official documentation also mentions: for a per-project install, drop the -g -a claude-code flags; or use git: git clone https://github.com/AirsupHQ/factory-skill.git ~/.claude/skills/factory. These are secondary methods, not the main path of this guide.
Let the session pick up the new Skill: The official docs state that Claude Code automatically detects newly installed skills in a running session; but if the skills folder did not exist before the session started, you need to run /reload-skills first.
Start a low-risk task: In Claude Code, type:
```
/factory a line that makes 2,000 e-bike motors a day
```
This is a purely generative task: it does not modify your existing files, but it will produce a new HTML file (this is its intended side effect). If you don't want files left in your working directory, switch to an empty scratch directory before triggering it.
Success signals: Claude will sequentially write the process route, size each station, check the line headlessly, then output a self-contained HTML file. Open it in a browser and you should see a tilt-shift miniature-style 3D factory page showing throughput, takt, WIP, lead time, utilization, and bottlenecks.
Optional: The official docs also provide two scripts, capture.mjs and record.mjs (requires an optional Playwright install), for taking screenshots/recording the page. You don't need to install these for a first try.

Use Cases

Case 1: Rapid visualization of a new line's capacity plan. You are planning a motor assembly line and want to know where the bottleneck is for a given capacity target. Type a one-line description such as /factory a line that makes 2,000 e-bike motors a day in Claude Code. Claude will write the process route, size the stations, run a headless check, and produce a single-file HTML simulation; you can read from the page's bottleneck display and utilization which station is the constraint. This is as far as you can confirm—the official documentation does not say whether you can adjust parameters and automatically re-run for comparison, so treat it as a one-shot generation per run.

Case 2: Animation recording for teaching/presentations. If you need to show your team the operating rhythm of a line (WIP buildup, takt matching), first confirm you have optionally installed Playwright, then use the official record.mjs script to record the page, or capture.mjs for screenshots. The evidence only goes as far as "these two scripts exist and depend on Playwright"; the scripts' specific command-line arguments are not covered in the official documentation, so inspect the scripts themselves before using them.

Boundaries of Use

File access: The intended output of /factory is a new HTML file. The official documentation does not state whether it reads/writes other files, executes other commands, or accesses the network—these behaviors are unknown, so do not assume it is completely side-effect free.
Runtime prerequisites: Node 20+ (the scripts use only built-in modules; the official docs state no npm install and no API key are needed); a WebGL2 browser; Playwright is only needed when using the capture/record scripts.
Client scope: The official documentation says it targets Claude Code but follows the Agent Skills format (a SKILL.md at the root, with ordinary scripts and reference files alongside), so any host that loads this format can use it; the skills CLI's -a <agent> flag can install it to other agents, and the Claude desktop/web apps can import it as a zip under Settings > Capabilities > Skills. Do not infer that it "only works with certain agents"—the official docs provide no exhaustive list.
Language and text type: This is a skill that generates simulation pages; it is unrelated to writing/text-processing skills. The official documentation does not state which languages the one-sentence description supports.
Key information not yet confirmed: The simulation's accuracy/assumptions, whether custom process steps and layouts are supported, whether the generated HTML can be opened offline, and the full parameter list of /factory—none of these are stated on the official page, so verify them yourself before use.