vrm-avatar-polish-looplisted
Install: claude install-skill Star-957/vrm-avatar-polish-loop
# VRM Avatar Polish Loop
## Objective
Produce visible, stable avatar improvements without overwriting the accepted model or confusing automated evidence with human acceptance. Keep every experiment reversible and every claim tied to one exact source/capture epoch.
## Automatic user-friendly entry
1. Before proposing a fix or editing, invoke `review-loop` in `companion-preflight` mode when installed.
2. Keep that pass read-only. Orient across the whole relevant model: source formats, licence/privacy, dependencies, meshes, materials, armatures, bones, weights, morphs/shape keys, expressions, physics, colliders, toggles, accessories/props, runtime integration, and existing evidence.
3. Do not ask implementation questions until orientation is complete. First distinguish a from-zero build from an existing-artifact workflow. For an explicit from-zero build, replace file inventory with a zero-artifact requirements/environment preflight and then continue to consequential clarification. Only when an existing artifact is intended but no artifact or location was supplied, ask one blocking locator/input question.
4. Then use `clarify-then-build` when installed. Ask only choices that materially change scope, privacy/publication, accepted appearance/behavior, source authority, production target, promotion, or irreversible work. Decide reversible implementation details autonomously and record assumptions.
5. Build only an isolated candidate after the direction is clear.
6. After candida