replica-capabilities

The bundled tool does the whole scan. It installs nothing, changes no config, and makes no network calls: it runs --version probes and checks a few paths.

python doctor.py                 # human-readable report
python doctor.py --json          # for the orchestrator to read
python doctor.py --target android   # only what one target needs

(doctor.py is next to this SKILL.md, under ~/.agents/skills, ~/.claude/skills, ~/.claude/plugins or the repo's skills/. Standard library only, so plain python or python3 runs it.)

What it decides

For each target type it prints [YES] or [no ] and, when no, the missing tools with where to get them:

target	ready needs	note
web	Node + npm	static sites and web apps. The core target.
desktop	Node	Electron on Node alone; Rust for Tauri, .NET or Python/Qt for native
android	Java + (Flutter or Gradle) + Android SDK	an emulator or an adb device to test
game	Node	web-canvas (Phaser, Three.js, PixiJS); Godot for a native engine
mashup	Node (or Godot)	an original blend of two games' mechanics; own engine and assets

How to use the result

1. Run it and show the user the plain report.
2. Read the --json targets block. If the target they asked for is not ready, do not start the build. List exactly what to install (the tool already names each one and its site), and offer either a target that is ready or to proceed after they install the missing pieces.
3. Write the JSON to replica/capabilities.json. replica-recon and replica-architect read it so the stack they pick is one this machine can build, test and run.
4. Never install anything yourself. The user installs tools and SDKs; you only report and re-scan.

Reporting

A short summary: OS and CPU count, the ready targets, and for the requested target either "ready" or the install list. Then hand back to the orchestrator's Wave 0.