SKILL.md
glb-compressor CLI
Command-line tool for compressing GLB/glTF 3D model files. Supports glob patterns, configurable presets, mesh simplification, and batch processing. Built on dreamcli, so --help, --version, --quiet, and --json are framework-owned root flags.
Binary Name
glb-compressor (installed via npm/bun) or bun run cli (from source).
Usage
glb-compressor <files...> [options]Options
| Flag | Description | Default |
|---|---|---|
-o, --output <dir> |
Output directory | Same as input (with -compressed suffix) |
-p, --preset <name> |
Compression preset | default |
-s, --simplify <0-1> |
Mesh simplification ratio (e.g. 0.5 = 50%) | None |
-k, --keep-nodes |
Preserve node hierarchy, node and material names, for parts moved at runtime by name (wheels, doors) | false |
-f, --force |
Overwrite existing output files | false |
-q, --quiet |
Suppress informational output (root flag, for scripting) | false |
--json |
Emit machine-readable JSON output (root flag) | false |
-h, --help |
Show help text | |
-V, --version |
Show version |
Presets
| Preset | Behavior |
|---|---|
default |
Conservative, preserves all detail |
balanced |
Moderate animation quantization, 24 Hz resample |
aggressive |
Strong animation quantization, 15 Hz resample (best for avatars) |
max |
Aggressive + supercompression + lower vertex precision |
Examples
# Compress a single file
glb-compressor model.glb
# Aggressive preset to an output directory
glb-compressor model.glb -p aggressive -o ./out/
# Batch compress with glob, overwrite existing
glb-compressor *.glb -f -p balanced
# Quiet mode for CI/scripts (exit code 0 = success, 1 = failure)
glb-compressor model.glb -q -p max
# Simplify mesh to ~50% vertex count
glb-compressor model.glb -s 0.5
# From source (development)
bun run cli -- model.glb -p aggressive -o ./compressed/Output Naming
Output files are named <input>-compressed.glb. When -o is specified, files are placed in that directory. Without -o, output is written alongside the input file.
Exit Codes
| Code | Meaning |
|---|---|
0 |
All files compressed successfully |
1 |
One or more files failed, or invalid simplify ratio |
2 |
Parse or constraint error (e.g. unknown preset) |
Pipeline
The CLI runs the same 6-phase pipeline as the library:
- Cleanup - dedup, prune, remove unused UVs (+ flatten/join/weld for
static)
- Geometry - merge by distance, remove degenerate faces, auto-decimate
- GPU - instancing, vertex reorder, sparse encoding
- Animation - resample keyframes, remove static tracks, normalize weights
- Textures - compress to WebP (max 1024x1024)
- Final - gltfpack (preferred) or meshopt WASM fallback
Skinned models are auto-detected and take a conservative path that skips transforms known to break skeleton hierarchies.
Skinned Model Awareness
When a GLB contains skins (e.g. avatars), the CLI automatically skips: flatten, join, weld, mergeByDistance, reorder, quantize, and auto-decimate. This prevents broken skeleton hierarchies, vertex weight denormalization, and mesh clipping artifacts.
Dependencies
- gltfpack (optional): If found in
$PATH, used for final compression (best
results). Falls back to meshopt WASM if unavailable.
- sharp: Used for texture compression to WebP.