verify

name	verify
description	Check mflux (MLX) availability, per-family runtime presence, cached weights, and report readiness status.
allowed-tools	Bash(command -v *), Bash(python *), Bash(python3 *), Bash(uv *), Bash(pip *), Bash(du *)

/verify

Check whether this plugin's prerequisites are met and report readiness.

Locate Plugin

PLUGIN_ROOT=$(python3 -c "
from pathlib import Path
import sys, os

if os.environ.get('JACK_TAR_MLX_ROOT'):
    print(os.environ['JACK_TAR_MLX_ROOT']); sys.exit()

home = Path.home()
for base in [home / '.claude' / 'plugins' / 'cache']:
    for p in base.rglob('jack-tar-mlx/.claude-plugin/plugin.json'):
        print(str(p.parent.parent)); sys.exit()

dev = Path.cwd() / 'plugins' / 'jack-tar-mlx'
if dev.exists():
    print(str(dev)); sys.exit()

print('NOT_FOUND')
" 2>/dev/null)

Step 1: Runtime check — per catalogued entry point

command -v mflux-generate-flux2
command -v mflux-generate-z-image-turbo
command -v mflux-generate-qwen

Report per-family presence. A missing entry point on an otherwise-working install usually means the installed mflux predates that family:

Family	Minimum mflux version
qwen (mflux-generate-qwen)	>= 0.11
z-image (mflux-generate-z-image-turbo)	>= 0.13
flux2 (mflux-generate-flux2)	>= 0.15 (the Runpod klein primary repo needs >= 0.16)

If ALL entry points are absent, report:

PLUGIN: jack-tar-mlx
VERSION: 0.1.0

DEPENDENCIES:
  mflux:           NOT_READY (run: uv tool install --upgrade mflux)

STATUS: NOT_AVAILABLE
REASON: mflux is not installed

Step 2: Version

There is no plain mflux entry point and no --version flag. Resolve the installed version with fallbacks, in order:

uv tool list 2>/dev/null | grep mflux

If empty:

pip show mflux 2>/dev/null

If empty:

python3 -c "import importlib.metadata as m; print(m.version('mflux'))" 2>/dev/null

Surface the resolved version in the report, and flag if it is below any family minimum from Step 1 whose entry point is present.

Step 3: Weights check

Shell to the wrapper's own Python helper — do NOT re-implement the snapshot-completeness check in bash:

python3 "$PLUGIN_ROOT/src/generate_image.py" --check-weights

This prints, per catalogued model, either READY (<repo>) or NOT_READY (run: hf download <repo>), using the same HF-cache snapshot-completeness check the deckhand detection seam uses.

Step 4: HF cache location + disk usage

Resolve the HF hub cache dir per precedence (HF_HUB_CACHE env var directly → HF_HOME/hub → ~/.cache/huggingface/hub) and report its path plus disk usage:

du -sh "$HF_HUB_DIR" 2>/dev/null

Step 5: Report status

If mflux runtime is present but no model is READY from Step 3:

PLUGIN: jack-tar-mlx
VERSION: 0.1.0

DEPENDENCIES:
  mflux:           READY (version X.Y.Z)

ENTRY POINTS:
  mflux-generate-flux2:          READY / NOT_READY
  mflux-generate-z-image-turbo:  READY / NOT_READY
  mflux-generate-qwen:           READY / NOT_READY

WEIGHTS:
  mlx/flux2-klein-4b:   NOT_READY (run: hf download Runpod/FLUX.2-klein-4B-mflux-4bit)
  mlx/z-image-turbo:    NOT_READY (run: hf download filipstrand/Z-Image-Turbo-mflux-4bit)
  mlx/qwen-image:       NOT_READY (run: hf download OsaurusAI/Qwen-Image-mflux-4bit)

HF CACHE: <resolved hub dir> (<disk usage>)

STATUS: NOT_AVAILABLE
REASON: mflux is installed but no catalogued model has cached weights

If mflux runtime is present AND at least one model is READY:

PLUGIN: jack-tar-mlx
VERSION: 0.1.0

DEPENDENCIES:
  mflux:           READY (version X.Y.Z)

ENTRY POINTS:
  mflux-generate-flux2:          READY

WEIGHTS:
  mlx/flux2-klein-4b:   READY (Runpod/FLUX.2-klein-4B-mflux-4bit)

HF CACHE: <resolved hub dir> (<disk usage>)

CAPABILITIES:
  image:           READY

STATUS: FULLY_AVAILABLE
REASON: mflux runtime + N cached model(s) available

If mflux runtime is entirely absent, use the Step 1 NOT_AVAILABLE report and skip Steps 2-4.